How to read the claim ledger

Few provisions of federal law have generated more argument per word than the twenty-six words of 47 U.S.C. 230(c)(1). The provision occupies less than a page of the United States Code, yet the claims made about it fill volumes of commentary, testimony, and argument across every medium. This article adds no new theory of what the provision ought to do. It tests eight claims that circulate about it against three kinds of authority: the statutory text, the case law construing that text, and the constitutional doctrine that operates alongside it.

Each claim appears in the strongest form its holders would recognize. That choice is method, not courtesy. A weakened version of a claim proves nothing when it falls, because its holders disown the weakened version. The strongest version is the one that has survived contact with objections, and it is the one worth testing. Claims pressed from one direction receive the same length and the same seriousness as claims pressed from the opposite direction. Equal treatment is a rule of this article, not a gesture toward balance.

Every claim carries a verdict. The scale is spare. False marks a claim that the text, the cases, or the doctrine do not support. Contested marks a claim on which serious readings of the same evidence reach different conclusions, usually because the dispute turns on prediction or on a judgment the law leaves open. No verdict stands without an authority beside it. Each verdict names the specific language, decision, or holding that decides it, quoted or paraphrased closely enough for a reader to check the work. A verdict that cannot name its authority is not a verdict. It is an opinion, and this article is not a collection of opinions.

The eight claims cover the full circuit of the argument. One holds that the provision conditions its shield on neutrality. Another holds that the law sorts services into publishers and platforms and protects only the latter. Two more make competing predictions about repeal, one that ending the provision would end censorship and one that ending it would improve moderation. Others hold that the shield covers only large firms, that it immunizes unlawful material, that the courts struck it down with the rest of the decency title, and that the Supreme Court has sustained it. Each is tested on the same terms.

The findable artifact of this article is the ledger itself: each claim, its verdict, the specific statutory text, case, or doctrine that decides it, and the article in this series that carries the full treatment of the underlying question. A reader who arrives holding one of the claims in circulation should leave knowing what the text, the case law, and the constitutional doctrine actually support, with each verdict sourced.

The ledger is built for checking, not for persuading. A reader need not accept any verdict on trust; each entry points to the exact words or decision that produced it, and the series articles named in each entry carry the extended analysis. Disagreement remains possible, and where the law leaves a question open the ledger labels it contested rather than forcing a false certainty. The discipline runs in both directions: a claim pressed against the provision’s protections faces the same textual test as a claim pressed in their favor, and the same verdict scale governs both.

A note on what correction means here. Several claims below are false in the precise sense that the provision does not say what the claim says it says. That finding does not settle whether the concerns behind the claims are legitimate. The immunity the provision creates shields conduct that many people find indefensible, and pointing at statutory language does not answer the unease that produces. A claim can fail as a reading of the text while succeeding as an expression of a genuine grievance, and this article treats those as separate questions. The temptation to grade every claim false is resisted throughout, and where predictions are genuinely contested rather than wrong, the article says so and states the mechanism on each side. Correcting the record is part of making a reader competent in the field where the gap between the words and the argument is widest, and competence here means holding two thoughts at once: the text says what it says, and the concerns it leaves unanswered are real.

The through line of the ledger is textual. The two words that structure nearly every public argument about this provision, neutral and platform, appear nowhere in it. Reading the twenty-six words resolves more disputes than any amount of argument about what the provision was meant to do. The full treatment of the language itself appears in the provision’s text explained. What follows applies that discipline claim by claim, with each verdict tied to the words, the cases, or the constitutional holdings, and with the record corrected at equal length in every direction.

How many claims does the ledger test, and how do the verdicts fall?

The ledger tests eight claims. Seven carry the verdict false: the neutrality condition, the publisher-platform sort, repeal ending censorship, protection only for large companies, immunity for illegal material, the provision struck down with the decency title, and a Supreme Court upholding. One claim, that repeal would improve moderation, is graded contested.

What three kinds of authority does the ledger test each claim against?

Each claim is tested against three kinds of authority: the statutory text itself, the case law construing that text, and the constitutional doctrine operating alongside it. No verdict stands without a named authority beside it, quoted or paraphrased closely enough for a reader to check the work.

Why does the ledger state each claim in its strongest form?

Each claim appears in the strongest form its holders would recognize, as a rule of method. A weakened version proves nothing when it falls, because its holders disown the weakened version. The strongest version is the one that has survived contact with objections, and it is the one worth testing.

What does a false verdict in this ledger not settle?

A false verdict settles only the reading of the text. The concerns behind a claim can be genuine even when the claim fails: the immunity shields conduct many people find indefensible, and pointing at statutory language does not answer that unease. The ledger treats the textual question and the grievance as separate questions.

What does subsection (c)(2)(B) protect?

Subsection (c)(2)(B) protects any action taken to enable or make available to others the technical means to restrict access to objectionable material. It covers the makers and distributors of filtering software and parental control tools, extending the Good Samaritan protection beyond the services that remove content to the people who build the instruments of removal.

The claim ledger

Claim	Verdict	Deciding text, case, or doctrine	Full treatment in this series
Section 230 requires platforms to be neutral	FALSE	The unconditional language of 47 U.S.C. 230(c)(1) and the express removal protection of 230(c)(2)(A); the Stratton Oakmont reversal confirmed in Zeran v. AOL, 129 F.3d 327 (4th Cir. 1997)	Section 230 Explained: The Provision’s Text, Clause by Clause
The law sorts services into publishers and platforms	FALSE	The word platform appears nowhere in the provision; the operative defined terms are interactive computer service and information content provider, 47 U.S.C. 230(f)(2)-(3)	Section 230 Explained: The Provision’s Text, Clause by Clause
Repealing Section 230 would end censorship	FALSE	Moody v. NetChoice, 603 U.S. 707 (2024): a platform’s curation of displayed content is expressive activity protected by the First Amendment, so repeal restores liability without compelling carriage	Internet Law Supreme Court Cases: Reno to Moody
Repealing Section 230 would improve moderation	CONTESTED	Two mechanisms pull in opposite directions, liability incentivizing removal-on-complaint versus liability incentivizing investment in professional standards, with no repealed baseline to test either	Section 230 vs. the DMCA Safe Harbors: Two Liability Regimes
Section 230 protects only large technology companies	FALSE	The size-blind definitions of 47 U.S.C. 230(f)(2)-(3); coverage extends to small services, non-profit entities, internet access providers, and ordinary users under 230(c)(1)	Section 230 Explained: The Provision’s Text, Clause by Clause
Section 230 immunizes illegal material	FALSE	Express carve-outs: 230(e)(1) preserves federal criminal law, 230(e)(2) preserves intellectual property claims, 230(e)(5) preserves specified sex-trafficking claims added by the 2018 FOSTA amendment	FOSTA and the 2018 Amendment to Section 230
Section 230 was struck down with the rest of the decency title	FALSE	Reno v. American Civil Liberties Union, 521 U.S. 844 (1997), invalidated only the indecency provisions of 47 U.S.C. 223; the immunity was never challenged	The Communications Decency Act of 1996: A Legislative History
The Supreme Court has upheld Section 230	FALSE	The Court has never construed the provision in a majority opinion; Gonzalez v. Google LLC, 598 U.S. 617 (2023), was resolved on the Taamneh aiding-and-abetting question without reaching immunity	Internet Law Supreme Court Cases: Reno to Moody

Claim: Section 230 requires platforms to be neutral

The provision was a bargain, and the bargain had terms. That is the claim in its strongest form. Congress extended an extraordinary shield, freedom from liability for the speech of others, to a young medium on an understanding: the services receiving the shield would function as open conduits rather than as editors with agendas. When a service ranks, promotes, demotes, or removes expression according to its own view of what deserves an audience, it performs the work of a publisher. Publishers have always answered for what they publish. The claim holds that the shield was never meant to cover that editorial work, and that a service which moderates with a point of view has surrendered the neutrality that justified its protection. On this reading the protection is conditional. Leave expression alone and keep the immunity. Shape expression to a viewpoint and answer as a publisher answers.

The textual hooks for this reading are not invented. Subsection (c)(1) says no provider shall be treated as the publisher of another’s information, which suggests to holders of the claim that the shield attaches to non-publisher behavior and falls away when publisher behavior begins. Subsection (c)(2) protects only action taken in good faith, language that reads like a standard the service must satisfy rather than an open grant. And the history points the same way. Congress acted after a New York court treated a service that exercised editorial control as a publisher, a result the provision was written to reverse, which implies the provision cares about the difference between moderating and not moderating. The grievance underneath is not abstract. A small number of services set the terms on which vast amounts of expression reach an audience, and the claim gives that grievance a legal form: neutrality was the price of the shield, and the price has not been paid. The bargain reading also draws force from ordinary ideas of fairness. A speaker that selects, frames, and amplifies expression while disclaiming responsibility for it seems to want the authority of a publisher with the accountability of a conduit. The claim insists the law cannot have meant to grant both at once, and that the good-faith language in (c)(2) is where Congress drew the line between protected housekeeping and unprotected viewpoint control.

Verdict: false. No neutrality condition appears anywhere in the provision.

The analysis begins with subsection (c)(1). Its twenty-six words read, in full: “No provider or user of an interactive computer service shall be treated as the publisher or speaker of any information provided by another information content provider.” The sentence contains no condition about viewpoint, balance, evenhandedness, or editorial policy. Protection attaches to two facts: the defendant is a provider or user of an interactive computer service, and the content at issue was provided by another information content provider. Nothing in the sentence asks how the service moderates, whether it moderates at all, or what viewpoint its moderation reflects. A service that removes nothing and a service that removes aggressively satisfy the same unconditional language, because the language states no behavioral test.

Subsection (c)(2) strengthens the point. It protects “any action voluntarily taken in good faith to restrict access to or availability of material that the provider or user considers to be obscene, lewd, lascivious, filthy, excessively violent, harassing, or otherwise objectionable, whether or not such material is constitutionally protected.” This is an express protection for removal decisions, including removal of material the provider itself considers objectionable. The operative phrase is “considers to be”: the judgment is the provider’s own, exercised according to its own lights. A neutrality requirement would penalize the very decisions this subsection protects. Had Congress meant to condition the shield on evenhandedness, it would not have written a subsection whose entire function is to protect non-neutral curation. The single stated condition is good faith, which qualifies the voluntary action to restrict access; it asks whether the restriction was genuinely aimed at material the provider considers objectionable, not whether the provider’s overall environment of expression is balanced. The language does not convert good faith into neutrality.

The companion clause, (c)(2)(B), points the same way. It protects providing others with the technical means to restrict access to objectionable material, which is to say it protects the distribution of filtering tools whose entire purpose is viewpoint-laden selection by the end user. A provision hostile to non-neutral curation would not protect the instruments of curation.

The history confirms the reading. The provision was enacted to reverse Stratton Oakmont, Inc. v. Prodigy Services Co., 1995 WL 323710 (N.Y. Sup. Ct. 1995), where a New York court held a service to the publisher standard precisely because it exercised editorial control over its forums. The claim that editorial control forfeits protection is the Stratton Oakmont rule restated. Congress rejected that rule, replacing a common-law regime under which screening content invited publisher liability with a regime under which screening content is expressly protected. Reading a neutrality condition back into the provision restores the regime the provision was written to displace.

The courts have applied the text as written. In Zeran v. AOL, 129 F.3d 327 (4th Cir. 1997), the first federal appellate decision construing the provision, the Fourth Circuit held that (c)(1) bars distributor liability as well as publisher liability and rejected the argument that notice of defamatory content creates a duty to remove it. The opinion treats the protection as categorical: once the content comes from another information content provider, the service is not treated as its publisher, and no inquiry into the service’s editorial posture follows. The court explained the purpose in the provision’s own terms, that it was meant to protect services attempting to screen offensive material and to avoid a rule that would punish the effort to moderate. A neutrality condition would invert that purpose, penalizing the screening the provision was written to encourage.

Nothing in the definitions helps the claim either. Subsection (f) defines the operative categories without any reference to viewpoint or balance, and subsection (e) lists the provision’s express carve-outs, none of which concerns viewpoint or neutrality. When a provision lists its exceptions and neutrality is not among them, the omission is part of the meaning. The text knows how to state conditions; it states none about viewpoint.

The source test is the whole of (c)(1)’s gatekeeping. The sentence protects the provider or user of an interactive computer service with respect to information provided by another information content provider, and the word another does the quiet work: where the provider is itself the information content provider for the material at issue, the sentence does not apply. That is a complete account of when protection attaches and when it falls away, and it leaves no room for a further inquiry into whether the service’s moderation was balanced. A court applying the sentence asks two questions, whether the defendant is a covered provider or user, and whether the content came from another. Viewpoint enters neither question.

The bargain metaphor from the claim’s strongest form deserves a direct answer, because it is doing the real work. The provision’s protections were indeed extended for a reason, but the reason stated in the text and recognized in the case law was to encourage services to screen offensive material without fear that the screening itself would create liability. That purpose points away from neutrality, not toward it. A service deciding what counts as excessively violent or otherwise objectionable is exercising judgment, and the provision protects the exercise. The quid pro quo the claim imagines, immunity in exchange for evenhandedness, has no anchor in any clause. The exchange the text actually records is different: protection conditioned on nothing except the service’s status and the content’s source, plus, in (c)(2), protection for the good-faith effort to keep objectionable material off the service.

The claim persists because it addresses a real question to the wrong text. The concentration of moderation power in a small number of services is a genuine subject of concern: private ordering governs the visibility of expression on a scale no prior medium approached, under standards the public does not write. The provision does not address that concern, because it was written to address a different one. The question before Congress was who should bear liability for third-party speech in a medium where intermediaries could not review everything they carried. The claim asks a governance question, how large services should moderate, and demands the answer from a liability provision. The provision is silent on the governance question, and silence is not a hidden neutrality rule. Conflating the two questions produces a reading the text cannot sustain, however strongly the governance question deserves an answer.

Part of the persistence is verbal. The word publisher in (c)(1) looks like a description of behavior, and in ordinary speech a publisher is an entity that exercises editorial judgment. In the provision, publisher is a term of liability, naming the legal treatment the sentence forbids: the provider shall not be treated as the publisher of another’s information. Reading behavioral conditions into a liability shield reverses the sentence. The provision does not say that providers behaving like publishers lose protection. It says providers shall not receive publisher treatment for content they did not create. The gap between the ordinary sense of the word and its legal function in the sentence accounts for much of the claim’s staying power.

What the claim gets right is the underlying unease. Immunity does shield moderation decisions that many people find indefensible, and the absence of a neutrality condition means the provision supplies no lever against viewpoint-based curation. That is a true statement about the law’s limits, stated plainly. The correction is narrower than the grievance: the grievance concerns how services should moderate, and its remedies lie in contract, in competition, or in new legislation addressed to moderation directly. The provision answers who pays for the speech of others. It does not answer, and was not written to answer, how the speech of others should be ordered.

The full treatment of the provision’s language belongs to the companion article in this series, which takes up the twenty-six words clause by clause, the structure of subsections (c)(1) and (c)(2), the defined terms in subsection (f), and the carve-outs in subsection (e). Readers seeking the complete textual foundation will find it there. This claim needed only the central point: the text conditions nothing on neutrality, and the subsection most often cited for a condition is the one that protects removal. The ledger entry reads: false, decided by the unconditional language of (c)(1) and the express removal protection of (c)(2)(A), confirmed by the Stratton Oakmont reversal and the categorical reading in Zeran. The words resolve the dispute; the argument about what the provision was meant to do cannot unsettle them.

The discipline applied here is the ledger’s discipline throughout: the words first, the purposes second, and every verdict naming the language that decided it. Where the text is silent, the ledger records silence rather than inferring a condition, and where the text speaks, as it does in (c)(2), the ledger follows it even when the result disappoints the claim. The neutrality condition is not hidden in the provision, waiting for a sympathetic reader. It is absent, and the absence is the verdict.

Claim: The law distinguishes publishers from platforms

The law built two boxes, and the immunity lives in only one of them. That is the claim in its strongest form. In one box sit publishers: entities that choose, arrange, promote, and suppress expression, exercising editorial judgment over what reaches an audience, and answering for what they publish. In the other box sit platforms: neutral conduits that merely carry the speech of others without shaping it. The claim holds that the provision’s protection was designed for the conduit box. A service that curates feeds, promotes some expression, and buries other expression has climbed out of the conduit box into the publisher box, and once there it should be judged by publisher rules. The vocabulary is everywhere in the public argument: publisher against platform, speaker against carrier, editor against bulletin board.

Holders of the claim argue the distinction is the load-bearing structure of the law. Without it, the immunity has no limiting principle: every act of curation a service performs would be shielded, and the protection would cover the very editorial conduct that publisher liability was meant to govern. That cannot have been the design, on this reading. The provision must sort services by function, protecting the passive carrier while exposing the active editor, or the shield swallows the law of defamation whole. The strongest version concedes that line-drawing is hard at the margins, then insists the line is real: a service that exercises editorial judgment over third-party expression functions as a publisher of that expression, and the law should treat function as decisive rather than letting labels do the work. The intuition draws on a genuine legal memory. Before the provision existed, courts did sort online services by function, treating some as distributors and others as publishers based on how much control they exercised. The claim presents itself as the continuation of that tradition, with Congress refining the common law rather than abandoning it. Holders also point to the word publisher inside (c)(1) itself as evidence that the drafters had the sorting in mind: why name publishers, the argument runs, unless the provision means to distinguish them from the conduits it protects. On this account the two boxes are not an invention of commentary but the architecture of the enactment, and the task of courts is to police the boundary between them.

Verdict: false. The word platform appears nowhere in the provision.

That single textual fact decides more than it might seem. The vocabulary that structures nearly every public argument about the law, platform against publisher, neutral conduit against active editor, is absent from the law itself. The provision does not sort services into publishers and platforms because it does not contain one of the two categories. The distinction is a rhetorical construct, serviceable in argument and empty in the text.

The categories the provision does contain are defined terms. An interactive computer service, defined in subsection (f)(2), is a broad category covering the services, systems, and access software providers that enable users to reach computer servers and related facilities, reaching services of every size and description. An information content provider, defined in subsection (f)(3), means anyone responsible in whole or in part for the creation or development of information. The line the provision draws runs between these two defined categories, not between publisher and platform. A service keeps the (c)(1) protection for content it did not create or develop; it crosses the line only by becoming an information content provider, responsible in whole or in part for the creation or development of the content at issue. Editorial judgment, curation, ranking, and removal do not appear in the definition. The test is creation or development, in whole or in part.

The word publisher does appear in the provision, but its function there is the opposite of what the claim assumes. Subsection (c)(1) provides that no provider or user of an interactive computer service shall be treated as the publisher or speaker of information provided by another information content provider. Publisher here names the liability the sentence takes away, not a status the sentence polices. The sentence does not say that a provider which acts like a publisher loses protection. It says a provider shall not receive publisher treatment for third-party content. The claim reads the sentence backwards: it treats publisher as a behavioral box the service must stay out of, when the sentence uses publisher as the legal consequence the service is spared. The common law knew three tiers, publisher, distributor, and mere conduit, each with its own standard of fault. Subsection (c)(1) collapses all three for the provider of third-party content: the sentence forbids treating the provider as a publisher, and Zeran holds that distributor treatment is forbidden with it, so the sorting the claim wants courts to perform is the sorting the text abolished.

The history explains why the confusion is natural and why the text rejects it. Before the provision, courts applied the common law of defamation to online services and sorted them by function. In Cubby, Inc. v. CompuServe, 776 F. Supp. 135 (S.D.N.Y. 1991), the court treated CompuServe as a distributor, akin to a bookstore or library, liable only with knowledge of the defamatory content. In Stratton Oakmont, Inc. v. Prodigy Services Co., 1995 WL 323710 (N.Y. Sup. Ct. 1995), a New York court reached the opposite result for Prodigy, holding it to the publisher standard because it exercised editorial control over its forums. The two decisions together created the moderator’s dilemma: screen content and risk publisher liability, or screen nothing and remain a distributor. Subsection (c)(1) was written to eliminate that dilemma. It provides that exercising editorial functions does not convert the service into a publisher for liability purposes. The claim that curation converts a service into a publisher is the Stratton Oakmont holding restated in new vocabulary, and it is the holding Congress reversed.

The first federal appellate construction confirmed the categorical reading. In Zeran v. AOL, 129 F.3d 327 (4th Cir. 1997), the Fourth Circuit held that (c)(1) bars distributor liability as well as publisher liability, rejecting the argument that notice creates a duty to act. The protection does not vary with the service’s editorial posture, because the text makes it unconditional on anything except the source of the content. A service that moderates heavily and a service that moderates not at all stand in the same position under the sentence, provided neither created or developed the content.

None of this means the protection lacks a boundary. It has one, drawn by the defined term information content provider. A service that is responsible in whole or in part for the creation or development of the content at issue steps outside the (c)(1) protection for that content. The boundary turns on authorship and development, not on curation. That is the limiting principle the text actually contains, and it is sharper than the publisher-platform sorting because it is defined rather than intuited.

The distinction feels real because it was once the law. Stratton Oakmont was a genuine holding, and the intuition that active curation differs from passive carriage is not foolish; it was the common-law rule for online services before 1996, and courts apply it in areas of law the provision does not touch. Readers who sense that a service shaping expression should answer for the shape are sensing something the law once recognized. Congress made a policy choice to displace that recognition for the categories the provision covers, for reasons the Zeran court set out: to protect services that attempt to screen offensive material, and to avoid imposing on intermediaries a burden of review the medium could not bear. The intuition survived the displacement because the result can feel counterintuitive, and because the commercial vocabulary of platforms supplied a new name for an old distinction.

What the claim gets right is that the shield has boundaries and that function matters. The provision does not protect everything a service does; it protects a service from being treated as the publisher or speaker of content provided by another. Where the service itself becomes responsible in whole or in part for the creation or development of information, the defined term information content provider removes it from the protection for that information. The instinct that the immunity must stop somewhere is correct. The correction is only about where the text draws the stop: at creation and development, not at curation. A vocabulary built on publisher against platform keeps generating the wrong boundary, because it measures the wrong thing. The provision measures authorship.

The persistence also owes something to the usefulness of the word platform outside the law. It is a serviceable commercial term and a vivid term of argument, and claims conducted in it feel concrete. But concreteness in argument is not presence in text, and the provision’s silence on the word is not an oversight for intuition to fill. The drafters wrote defined categories, and the defined categories do the work. A reader who sets aside the publisher-platform sorting and reads the definitions finds a sharper instrument: not two boxes for services, but a single question about each item of content, namely who is responsible for its creation or development. The sorting the claim seeks is not absent because the drafters overlooked it. It is absent because they replaced it, deliberately, with a question about authorship that needs no boxes at all.

The companion article in this series carries the full treatment of the defined terms, the pre-provision case law, and the moderator’s dilemma that Cubby and Stratton Oakmont created. Readers seeking the complete account of how the common-law sorting worked, why Congress displaced it, and where the information-content-provider boundary runs will find it there. This claim needed only the central point: the provision contains no platform, draws no publisher-against-platform line, and conditions protection on the source of the content rather than on the service’s editorial posture.

The ledger entry reads: false, decided by the absence of the word from the text, the defined terms in subsection (f), and the Stratton Oakmont reversal confirmed in Zeran. The twenty-six words, read with the definitions beside them, resolve the dispute without resort to the vocabulary the dispute prefers. That is the discipline the ledger applies throughout: the words first, the argument second, and every verdict naming the language that decided it. A claim that cannot find its central term in the provision is not interpreting the provision. It is arguing with a different text.

Claim: Repealing Section 230 Would End Censorship

In its strongest form, the claim runs as follows. A twenty-six-word passage in a 1996 federal law created the permission structure for private moderation. The passage provides, in full: “No provider or user of an interactive computer service shall be treated as the publisher or speaker of any information provided by another information content provider.” Because services cannot be held liable as publishers of third-party material, they face no legal cost when they remove expression they disfavor, and they face little legal cost when they leave expression in place. The result, in this telling, is censorship made costless: services may delete lawful posts, demote unpopular positions, and enforce house standards that answer to no public rule, all under the cover of the provision.

Repeal the provision, the argument continues, and the cover disappears. Without the immunity, a service that removes one user’s lawful expression while hosting another’s would face the same exposure as a newspaper that prints one letter and spikes another. Courts would treat the service as a publisher, and publisher liability would punish viewpoint-based removal. Services would respond the way rational publishers respond to liability: they would stop deleting lawful expression, stop enforcing ideological house rules, and restore the open exchange that the provision suppressed. Censorship would end not because anyone ordered it to end, but because the economics of liability would make it too expensive to continue.

Put at its strongest, then, the claim is a causal story: the statute enabled the moderation regime, removal of the statute disables it, and the forum returns to something like its pre-1996 openness. Holders of this view often add that the provision’s second subsection, which expressly protects good-faith removal of objectionable material, is the specific mechanism of the harm, because it invites services to take down expression and promises them legal safety for doing so.

The argument draws additional force from a common-law pairing that the statute severed. At common law, editorial control and publisher liability traveled together: the party that selected, edited, and arranged material answered for it. The provision broke the pairing for online services, letting them exercise editorial control through removal while answering for nothing they hosted. Holders of the claim treat repeal as the restoration of the natural order, in which control implies responsibility, and they predict that responsibility will discipline control. The causal story is therefore not only about cost but about coherence: liability and editorial power belonged together before 1996, and reuniting them would end the anomaly of power without responsibility.

The verdict on this claim is false, and the reason is constitutional rather than statutory. Repealing the provision would change the liability rules for hosted content. It would not change the rule that governs a service’s own curation, because that rule comes from the First Amendment, and the First Amendment does not depend on the statute.

In Moody v. NetChoice, 603 U.S. 707 (2024), the Supreme Court held that a platform’s curation of the content it displays is expressive activity protected by the First Amendment. Selecting, ordering, excluding, and labeling third-party material is itself expression, and government compulsion of that expression triggers the same constitutional scrutiny as compulsion of any other editorial choice. The Court vacated the lower-court judgments on procedural grounds, in the posture of a facial challenge, without reaching final merits, but the expressive-activity holding is the governing statement of doctrine.

That holding decides the claim. A legislature that repealed the twenty-six words could restore publisher and distributor liability for user-supplied material. It could not, by the same act, compel a private service to carry expression the service does not wish to display, because such compulsion would regulate the service’s own protected expression. The First Amendment protects the decision to exclude as fully as it protects the decision to include. Repeal would therefore do exactly one of the two things the claim predicts, and it would do the one the claim does not want: it would expose services to liability for what they host, while leaving their constitutional right to refuse to host anything untouched.

The repeal scenario that follows is the opposite of the one the claim imagines. Liability for hosted content gives every service a new reason to remove material, and the removal incentive falls hardest on expression near the legal line. Defamation suits, not viewpoint rules, would drive the next round of takedowns. A service weighing whether to keep a contested post would face a one-sided calculation: leaving it up risks a lawsuit, while taking it down risks only a complaint. Rational services resolve one-sided calculations the same way, and the resolution is more removal, not less.

This is not speculation about motive. It is the documented structure of distributor liability, the body of law the statute displaced. In Zeran v. AOL, 129 F.3d 327 (4th Cir. 1997), the Fourth Circuit held that the provision bars distributor liability as well as publisher liability. Distributor liability is the rule that a party with notice of defamatory material that fails to remove it answers for it. Repeal would revive that rule. Under it, a service that receives a complaint about a post and leaves the post standing accepts legal exposure for the post’s contents. The cheapest compliance strategy is removal on complaint, and the history of distributor-liability regimes is a history of exactly that strategy.

The severed pairing deserves a closer look, because it is the strongest part of the claim’s logic and the point where the constitutional answer bites hardest. Repeal would indeed rejoin editorial control and liability for hosted content: a service that curates aggressively would face greater exposure than one that curates lightly, because curation creates knowledge, and knowledge triggers distributor liability. But the rejoining stops at the service’s own expression. The First Amendment does not permit the government to resolve the pairing by ordering the service to carry material against its judgment, because the judgment itself is protected speech. The common-law coherence the claim seeks is therefore achievable only on one side of the line: liability for what is carried, never compulsion to carry.

Packingham v. North Carolina, 582 U.S. 98 (2017), sometimes enters this argument on the claim’s side, because the Court described social media services as among the most important places for the exchange of views. The description is accurate, and the holding is real: the Court struck down a state law that barred certain offenders from accessing such services. But the holding bound the state, not private services. Nothing in Packingham converts a private service’s curation into state action, and nothing in it obliges a service to carry expression. The case protects access against government restriction; it does not create a private right to be hosted.

There is a further asymmetry worth stating plainly. The First Amendment protects editorial choice in both directions: a service may exclude expression it dislikes, and a service may include expression it favors, and the Constitution protects both decisions. Repeal would not create a must-carry rule, because must-carry rules imposed on private speakers face the same constitutional barrier. The only regime in which services must carry expression they would rather exclude is a common-carrier regime, which is a different legal theory from repeal, imposed by different legislation and defended on different constitutional grounds. Repeal alone produces liability without obligation: exposure for what is hosted, freedom as to what is excluded.

So the ledger on the claim reads as follows: repeal would restore liability for hosting, revive distributor liability with its removal-on-complaint incentive, and leave the First Amendment’s protection of editorial choice fully intact. The predicted outcome, an end to viewpoint-based removal, is the one outcome repeal cannot produce, because the Constitution, not the statute, is what makes a service’s curation its own speech.

False verdicts need explanation, because this claim persists among serious observers, and it persists for reasons that deserve respect. The underlying concern is real: services make removal decisions that are effectively unreviewable, enforce standards that shift without public input, and answer complaints with procedures the public cannot see. The provision’s second subsection expressly protects good-faith removal, which means the removal decision itself is doubly insulated, first by the statute and then by the constitutional protection of editorial choice. A reader who watches a lawful post disappear with no explanation is watching something the law genuinely permits, and frustration with that permission is not a misunderstanding.

What the claim gets right, then, is the diagnosis of unaccountability. The immunity does shield conduct that many people find indefensible: removals without stated reasons, inconsistent enforcement across similar posts, and house rules that track no published standard. The gap between the power services exercise and the accountability they face is a genuine feature of the legal regime, not an invention of the claim’s holders. Any treatment that dismisses the grievance as mere confusion will lose the readers who most need the textual precision.

The claim also draws strength from a textual hook that is easy to misread. The provision’s second subsection expressly protects good-faith action to restrict access to objectionable material, and a plain reader can be forgiven for concluding that the statute is a censorship charter: it shields the hosting of third-party material in one breath and shields the removal of third-party material in the next. Add the publisher-versus-platform confusion addressed elsewhere in this article, which makes repeal sound like a return to a simple publisher standard, and the claim assembles itself from the text. The misreading is understandable, which is why the correction has to be patient: the text protects removal, the Constitution protects the choice behind the removal, and repeal touches only the first.

Where the claim goes wrong is the remedy. Repeal targets the statute, but the conduct the claim objects to is protected by a second, independent layer of law that repeal would not touch. The accurate version of the grievance is narrower and harder: the problem is not one provision but the combination of a statutory shield and a constitutional shield, and removing only the statutory one leaves the constitutional one standing. Precision about the text does not require indifference to the concern; it requires directing the concern at the layer of law that actually does the work.

The full constitutional treatment belongs to the series article on Supreme Court internet-law cases, which traces the editorial-choice holdings from the print era through Moody and explains why curation counts as expression. Readers who want the doctrine in its complete form, including the procedural posture that left Moody’s merits unresolved, will find it in the series survey of the Court’s internet-law decisions. The survey also carries the Packingham discussion in full, so readers can see exactly what the access holding decided and what it did not decide. The text-level treatment of the twenty-six words sits in the series explainer on the provision, and the history of the decency title that produced them sits in the series history of the 1996 act. No other article in the series takes up the constitutional question at the same length, which is why the verdicts on both repeal claims in this article rest on it. Readers who work through the survey with the twenty-six words in hand will see why the repeal predictions in this article divide the way they do: one false, one contested, and both decided by doctrine rather than by preference.

Claim: Repealing Section 230 Would Improve Moderation

In its strongest form, this claim is a story about incentives. The provision, on this account, removed the ordinary legal pressure that makes publishers careful. A newspaper checks facts before printing because a false statement costs money in court. A service protected by the twenty-six words faces no equivalent cost, so it underinvests in review, writes vague standards, and staffs moderation as a cost center rather than a craft. The result is arbitrary enforcement: identical posts treated differently, appeals that go nowhere, and rules that change without notice.

Holders of the claim are usually specific about what improvement would look like. Better moderation, in their account, means standards written clearly enough to be followed and defended, enforcement consistent enough that similar posts receive similar treatment, appeals that produce written reasons rather than silence, and investment in review proportional to the scale of what is hosted. None of these, they argue, requires heroism; all of them are ordinary features of careful publishing, and all of them are optional under the immunity because the law asks nothing. Liability would make them mandatory, and mandatory care is what the claim means by improvement.

Repeal the provision, and the ordinary pressure returns. Services that host third-party material would answer for what they host, which would make carelessness expensive. Rational services would respond the way careful publishers respond to liability: they would hire trained reviewers, build documented appeals processes, write precise public standards that can survive scrutiny, and invest in systems that catch harmful material before it spreads. Moderation would improve because the law would finally price the harm of bad moderation.

At its strongest, the claim does not promise perfection. It promises direction: liability points the incentives toward investment, and investment points moderation toward consistency. Holders of this view often concede that some services would remove more material and that some smaller services would struggle, but they treat those costs as the price of accountability, and they hold that the price is worth paying for a moderation regime that answers to rules instead of whim.

The claim also answers the objection that liability produces only more removal, which is the subject of the next section, by distinguishing quantity from quality. More removal is not the goal; better-governed removal is. A service that removes under clear rules, with documented reasons and a functioning appeal, is practicing better moderation even if it removes more than before. The holders’ measure is not the volume of takedowns but the discipline behind them.

The verdict on this claim is contested. Two mechanisms pull in opposite directions, each is grounded in how liability shapes behavior, and serious analysts endorse each. The honest treatment states both at full strength and declines to declare a winner, because the evidence available cannot settle the dispute.

Mechanism one holds that increased liability exposure produces more removal and less tolerance for edge content. The logic is the economics of the moderator’s position. Review is expensive, lawsuits are more expensive, and the two kinds of moderation error carry wildly different price tags. Leaving up a defamatory post risks a judgment. Taking down a lawful post risks a user complaint, and user complaints carry no legal price. When the costs are one-sided, rational services resolve the one-sidedness by removing. The predicted result is not better moderation but broader moderation: aggressive takedowns, hair-trigger responses to complaints, and a shrinking margin for expression near the legal line.

The mechanism has a second-order effect that its proponents stress. Large services can afford review teams and legal departments; small services, non-profit reference sites, and individual hosts of comment sections cannot. For those smaller actors, the rational response to publisher liability is not better review but no user content at all: close the comments, drop the forum, and publish only material the service itself produces. The long tail of participatory spaces, which the provision’s broad definition of protected services was written to cover, would shorten. Moderation would not improve on those sites; it would disappear along with the sites themselves.

Mechanism one has a further dynamic that its proponents emphasize: notice-based liability hands leverage to whoever complains loudest. Under a revived distributor-liability rule, a complaint creates knowledge, and knowledge creates exposure, so the rational service removes first and investigates later, if at all. The cost of filing a complaint is near zero; the cost of reviewing one properly is substantial; the cost of guessing wrong about a defamation claim is a lawsuit. That arithmetic favors the complainant in every round. Organized campaigns that file complaints at scale can therefore move a service’s enforcement without persuading anyone of anything, and the predicted result is a moderation regime governed less by the service’s judgment than by the stamina of its complainants, which mechanism one’s proponents count as degradation rather than improvement.

Mechanism two holds that accountability pressure produces investment in better systems and clearer rules. The logic starts from the same premise, that liability changes incentives, and draws the opposite conclusion. When hosting and removal decisions carry legal consequences, services can no longer treat moderation as a cost center. They hire reviewers, train them, document decisions, build appeals that function, and write standards precise enough to defend in court. The predicted result is professionalization: moderation becomes a discipline with procedures, records, and accountability, closer to the editorial standards of traditional publishers than to the improvised enforcement the immunity permitted.

Proponents of this mechanism add that its critics underestimate how much of the moderation failure described in the commentary comes from the absence of pressure. Vague standards persist, they argue, because vagueness is cheap when no court will ever test the standard. Appeals go nowhere because there is no consequence for ignoring them. Attach a price to carelessness, and the vagueness and the indifference become liabilities on the balance sheet, which is exactly the kind of problem that well-funded services know how to solve. On this account, the immunity did not protect moderation; it protected bad moderation, and repeal would force the distinction.

Mechanism two’s proponents point to the model that traditional publishers provide. Newspapers operating under full publisher liability developed corrections policies, standards desks, libel review, and public editors, not out of generosity but because the law made carelessness expensive and the market rewarded the resulting trust. The claim’s holders argue that large services, competing for users and advertisers on the basis of reliability, would follow the same path: liability would convert trust from a marketing slogan into a legal asset, and the asset would be protected by the same professional apparatus that publishers built. On this account, the difference between a service and a publisher is not a difference in kind but a difference in legal exposure, and equalizing the exposure would equalize the care.

Both mechanisms are internally coherent, and both rest on the same uncontroversial premise: legal exposure changes behavior. They differ on which behavioral change dominates. Mechanism one bets that the cheapest compliance strategy, removal, wins. Mechanism two bets that the reputational and competitive value of a functioning forum wins, and that services will spend their way to quality rather than shrink their way to safety. Nothing in the logic of liability itself picks between the bets. The disagreement is therefore empirical rather than logical, a dispute about elasticities: how responsive removal volume is to liability risk, how responsive investment is to the same risk, and which response arrives first. Both sides agree that the short run belongs to removal, because removal is the faster strategy; they disagree about whether the long run belongs to investment. That timing asymmetry matters for grading, because a transition that lasts years is, for the services that fail during it, indistinguishable from a permanent state. Mechanism one’s warning about the long tail is therefore not answered by mechanism two’s promise about the long run; the two operate on different clocks, and the verdict stays contested on both.

Each side’s account carries a concession that explains why the dispute persists. The proponents of mechanism one concede that their prediction describes a transition, not a permanent state: after an initial wave of over-removal, services might learn to calibrate, and the long-run equilibrium could look better than the short-run panic. They also concede that the heaviest costs fall on small services, which means their prediction is really two predictions, one for large services that can afford review and one for small ones that cannot.

The proponents of mechanism two concede the mirror image: investment follows only where the returns justify it, and there is no guarantee that the returns on careful moderation exceed the returns on aggressive removal. A service that discovers over-removal is cheaper than good review will choose over-removal, and liability alone does not punish that choice. They also concede that professionalization takes time, during which the removal-heavy strategy of mechanism one would dominate.

Both sides also draw on the same history and draw opposite lessons from it, which is a further reason the dispute resists settlement. The distributor-liability era before Zeran is mechanism one’s exhibit: notice-based exposure produced removal on complaint, exactly as predicted. The publisher-liability tradition of the press is mechanism two’s exhibit: full exposure produced professional standards, exactly as predicted. Each side can point to a real regime that behaved the way its mechanism predicts, and neither regime is the repealed provision, because the repealed provision has no history. The evidence underdetermines the choice because the closest analogues point in opposite directions. A further difficulty is selection: the services that survived the distributor-liability era and the publishers that thrived under publisher liability are the ones whose behavior the record preserves, while the ones that closed or never launched left no data. Both mechanisms may therefore be describing survivors’ strategies rather than universal ones.

The deeper reason the dispute persists is evidential. The provision has applied in substantially the same form since its enactment in 1996, which means there is no large-scale repealed baseline against which to test either prediction. Analysts argue from adjacent regimes, from the history of distributor liability, and from the economics of publishing, but adjacent is not identical, and the economics cut both ways. Neither mechanism can be falsified by the record, because the record contains no repeal. The claim is contested not because the analysts are careless but because the question is genuinely hard, and hard questions do not yield to the volume of argument.

The series carries the full treatment of the provision’s text in its explainer article and the full treatment of reform proposals in its comparison of the provision with the copyright liability regime. Readers who want to test either mechanism against the actual statutory language should start with the text, and readers who want to see what reform proposals actually propose, as distinct from what repeal would do, should start with the comparison. The explainer also covers the breadth of the protected-services definition, which is the factual basis for mechanism one’s warning about the long tail. Both articles keep the same discipline this claim requires: state the mechanism, label the prediction, and refuse to grade what cannot be tested. Together the three treatments give the reader the words, the proposals, and the honest limits of prediction. Readers who want the constitutional holdings behind the repeal analysis will find them in the series survey of the Court’s internet-law decisions, which carries the Moody expressive-activity holding and the Packingham access holding in full. The claim ledger at the article’s close records this claim as contested, with each mechanism named, so that the verdict stated here matches the verdict entered there.

Claim: Section 230 protects only large technology companies

The claim in its strongest form runs something like this. Section 230 was written in 1996, long before the internet was dominated by a handful of giant firms, and it was meant to protect the fledgling services of that era. What it actually protects in practice are the largest and most powerful technology companies on earth, businesses with market capitalizations in the trillions, legal departments numbering in the thousands, and lobbying operations that span the globe. The immunity shield lets these firms host billions of user posts, moderate them according to private rules, and sell the advertising around them, all without facing the liability that a newspaper or broadcaster would face for distributing the same material. A newspaper that prints a defamatory letter can be sued for libel. A platform that algorithmically recommends a defamatory post to millions of readers cannot, and the difference in treatment is the statute at work. Ordinary people get no comparable benefit. A small business owner who posts a review, a neighborhood forum with a dozen members, a volunteer-run reference site, none of these actors has a compliance apparatus, and none of them is the reason the provision exists. The statute is, in this telling, a subsidy for Big Tech, drafted when the internet was young and captured by the industry it was meant to nurture. When people say Section 230 protects only large technology companies, they mean that the real-world distribution of its benefits is concentrated at the top, and that any honest description of the provision has to start with that concentration. The claim does not necessarily say that the text names large companies. It says that the mechanism of the text serves large companies, that the concentration is structural rather than accidental, and that the rest of the coverage is incidental.

The verdict on this claim, as stated, is false. The text decides it, and it decides it at the definitional level. Subsection 230(f)(2) defines an interactive computer service as any information service, system, or access software provider that provides or enables computer access by multiple users to a computer server. The definition contains no size threshold, no revenue floor, no employee count, and no market-share test. A dial-up internet access provider with a few hundred subscribers is an interactive computer service under this definition. A hobbyist forum with a hundred members is an interactive computer service. A non-profit reference site maintained by volunteers is an interactive computer service. A university hosting a student discussion board is an interactive computer service. The definition reaches every provider that enables multi-user access to a server, from the smallest bulletin board to the largest social network, because the operative fact is the provision of access, not the scale of the enterprise. Note how the definition is structured. It names three categories, information service, system, and access software provider, and it keys the definition to the function performed: providing or enabling computer access by multiple users to a computer server. Function, not identity. A provider qualifies because of what it does, and nothing in the subsection asks how many users it serves, how much revenue it earns, or how large its workforce is. Congress knew how to write size thresholds when it wanted them; employment statutes, securities regulations, and small-business programs are full of numerical cutoffs. This subsection contains none.

The functional character of the definition matters for another reason. When Congress revisited the provision in later years, it amended the exceptions without narrowing the definitions. The 2018 FOSTA amendment added subsection 230(e)(5) while leaving the (f)(2) definition exactly as it was. If the legislative understanding had been that the definitions needed to be drawn more tightly, whether by size or by any other criterion, that amendment would have been the natural place to do it. The definitions have remained stable across decades of controversy, and their stability is itself evidence that the breadth was deliberate, not accidental.

The companion definition in subsection 230(f)(3) defines an information content provider as anyone responsible in whole or in part for the creation or development of information. That definition, too, is size-blind. The dividing line the provision draws is between those who create content and those who provide access to it, and that line is drawn by conduct, not by headcount. A sole proprietor who writes the words is an information content provider. A multinational corporation that merely hosts them is not. The categories sort by behavior, which means that the large firm and the small firm are treated identically when they engage in the same behavior. And the operative immunity in subsection 230(c)(1) is granted to a provider or user of an interactive computer service. The statute protects the user alongside the provider, which means that an individual who forwards or reposts another person’s content is covered. A person who shares a defamatory post written by someone else cannot be treated as the publisher of that post for the purposes of civil liability, because the text says provider or user, and a user is a person. This matters because the most numerous class of beneficiaries under the provision is not the large firms at all. It is ordinary users, the millions of individuals who repost, forward, quote, and comment on material they did not create. Every time a person shares someone else’s review, retweets someone else’s allegation, or pastes someone else’s claim into a group chat, the user-side coverage is the thing standing between that person and a defamation suit. The claim that the provision protects only large companies ignores this entire half of the immunity, and it ignores the half of the immunity that covers the most people.

The case law confirms the breadth of the definitions in practice. Courts have applied the provision to small and mid-sized services with no national profile: local newspaper comment sections, municipal websites, consumer complaint sites, personal blogs, niche forums, dating sites, job boards, and apartment-listing pages. The protection has been extended to non-profit entities, including educational and charitable organizations that host user comments or community content. Internet access providers, the kind of companies that simply sell a connection and host no content of their own, have invoked the provision successfully. And the user-side coverage has mattered in cases where an individual defendant, with no corporate affiliation at all, faced defamation claims over reposted material and was shielded by the same twenty-six words that shield the largest firms. None of these outcomes required the court to stretch the text. The definitions were written broadly enough to cover them on their own terms, and the courts have read them that way for decades. The consistent judicial reading is that the subsection describes a category of activity, providing interactive access to third-party content, and that anyone engaged in that activity qualifies. Parties have occasionally urged courts to read a size distinction into the provision, arguing that the policy rationale for immunity applies differently to large commercial services than to small community ones. The courts have declined, and they have declined on textual grounds: the definitions do not contain the distinction, and a court cannot insert what the statute omits. This is not judicial indifference to the policy concern. It is the ordinary consequence of a definition written in functional terms. A rule that asks what a defendant does, rather than how large the defendant is, produces the same answer for defendants of different sizes who do the same thing.

There is a further precision worth making, because the strongest version of the claim sometimes shifts from coverage to benefit. Suppose the claim is revised to say that even if the text covers small services, the practical benefit flows mostly to large firms. The honest answer is that this is partly true, and it is true for reasons that have nothing to do with the text. Large firms host more third-party content, face more lawsuits, and have more to lose from liability, so any immunity from that liability is worth more to them in absolute terms. That is a fact about scale, not a fact about the provision. The same is true of any general legal rule. A broad evidentiary privilege is worth more to a company that is sued often. A favorable tax provision is worth more to a company with more income. The distribution of the benefit follows the distribution of the underlying exposure, and the underlying exposure is larger for larger firms. Nothing in that observation narrows the coverage of the text.

Why, then, does the claim persist, and what does it get right? The largest beneficiaries are the most visible ones. When a major firm invokes the provision in a high-profile dispute, the filing is reported, the arguments are quoted, and the outcome is discussed. When a small forum invokes it to dismiss a defamation claim over a comment thread, no one writes about it. The visibility of the large companies creates an availability illusion, and the illusion is reinforced by the fact that the provision does disproportionately matter to large firms in absolute terms, as just described. There is also a legitimate policy kernel inside the claim that deserves to be separated from the textual error. The argument that a valuable legal benefit should not flow most heavily to the most powerful firms is an argument about fairness, and it does not depend on a misreading of the definitions. A reader can accept that the text is size-blind and still hold that the text is badly designed, that the immunity should be conditioned on something, or that the distribution of its benefits reveals a flaw in its design. The claim goes wrong only when it converts that policy objection into a description of the provision, asserting that the provision protects only large firms when the text protects everyone engaged in the covered activity. The distinction between what the law is and what the law should be is the distinction the claim collapses, and the textual correction restores it. The claim also gets something right about the moral terrain. The immunity does shield conduct that many people find indefensible, and the largest firms provide the largest venues on which such conduct occurs, so the concentration of the benefit at the top is not merely an artifact of reporting but a real feature of how the provision operates at scale. The correction is narrower than it first appears. Visibility is not coverage, and the magnitude of a benefit is not its definition. The text draws no line by size, the courts have applied it across the whole range of services the definitions describe, and the user-side half of the immunity protects individuals with no corporate affiliation at all. A reader who wants to argue that the benefit should be narrowed, or that it should be conditioned on something, is making a policy argument about what the law ought to say. That is a legitimate argument. But it is an argument about reform, not a description of the provision as it exists.

For the full treatment of the definitions and the scope of the provider-or-user language, see the earlier article in this series devoted to the statutory text and its construction.

Claim: Section 230 immunizes illegal material

The claim in its strongest form runs something like this. Section 230 gives internet services a blanket immunity from liability for the content they host, and the content they host includes material that is illegal. Exploitation imagery, drug trafficking, fraud, defamation, threats, harassment, the sale of counterfeit goods, services host all of it in the torrent of user posts, and the provision means that no one can sue the service for it. The effect is a zone of lawlessness. A victim of illegal content posted by a stranger cannot recover damages from the only party that had the power to remove the material before it spread, because the provision blocks the claim at the door. The usual counter that the service is not the wrongdoer misses the point, because the immunity does not merely allocate responsibility to the poster, it eliminates the only practical remedy when the poster is anonymous, judgment-proof, or beyond the reach of the courts. Consider the victim’s position concretely. A person defamed by an anonymous account has no defendant except the service that published the defamation to thousands of readers. A small business destroyed by fraudulent reviews posted from throwaway accounts has no defendant except the site that hosted and ranked those reviews. The poster is a ghost, and the provision protects the one party that is not. When people say Section 230 immunizes illegal material, they mean that the practical consequence of the provision is that unlawful content circulates with no legal accountability for the entity that made its circulation possible. The claim is not about the narrow question of criminal prosecution. It is about the lived reality of victims who find that the courthouse door is closed to them, and it draws its force from that reality. The claim deserves to be stated at this strength, because anything weaker would be a straw version of a serious objection. Holders of the claim often refine it further, and the refinement deserves stating too. The argument is not always that the provision shields illegality from the first moment of posting. It is that the shield should end where knowledge begins: once a service has been notified that specific content on its servers is unlawful, or once it actually knows, the continued hosting is a choice, and a provision that protects that choice is protecting illegal material. In this refined form, the claim targets not the immunity as such but the absence of a knowledge-based cutoff, the absence of any point at which the service must act or answer. The refined version also reaches the service’s own design choices. A service that builds its product to amplify engagement may find that unlawful content is among the most engaging material it carries, and a provision that shields the resulting distribution is, in the holders’ view, shielding a business model that profits from illegality. These refinements are the strongest form of the claim, and the verdict below addresses them, because a verdict that only answers the crudest version would not answer the one that moves people.

The verdict on this claim, as stated, is false. The text decides it, and it decides it in the exceptions. Subsection 230(e)(1) provides that nothing in the provision shall be construed to impair the enforcement of federal criminal law, including the obscenity and sexual-exploitation provisions. A service that knowingly hosts material constituting a federal crime remains exposed to federal prosecution. The provision does not touch that exposure at all. The word nothing is doing real work in the subsection. It is a categorical preservation, not a balancing test, and it applies to the entire enforcement apparatus of federal criminal law, from investigation through prosecution. The exceptions continue. Subsection 230(e)(2) preserves intellectual property claims, so the copyright and trademark frameworks operate independently of the provision. A service facing a copyright claim answers under the copyright laws, not under Section 230, and the provision provides no defense there. Subsection 230(e)(3) preserves state laws consistent with the provision. Subsection 230(e)(4) preserves communications privacy and electronic communications privacy claims. And subsection 230(e)(5) preserves specified sex-trafficking claims, an exception added by the Allow States and Victims to Fight Online Sex Trafficking Act, Public Law 115-164, enacted April 11, 2018, which amended the provision to carve out civil and criminal actions arising from the knowing facilitation of sex trafficking. The full history of that amendment, and what the carve-out does and does not reach, is treated at length in this series’ account of the 2018 legislation, available at /2017/11/01/fosta-sesta-2018-section-230/.

The precision that matters here concerns what the shield actually covers. The provision bars treating a service as the publisher or speaker of third-party content in civil claims arising from that content. That is a narrower shield than the claim imagines. It does not purport to legalize any material. It does not extinguish the underlying unlawfulness of the content. A defamatory post remains defamatory, a fraudulent listing remains fraudulent, and the person who created the content remains liable for it. What the provision does is channel the civil claim away from the intermediary and toward the author, on the theory that the intermediary did not create the content and should not be treated as its publisher. This is a liability-allocation rule, not a legalization rule. The distinction is the difference between saying that no law was broken and saying that the wrong defendant is being sued. Federal prosecutors retain their tools under subsection 230(e)(1). Intellectual property owners retain their remedies under subsection 230(e)(2). And since 2018, trafficking-related claims have their own express carve-out under subsection 230(e)(5). The claim that the provision immunizes illegal material as a category collapses against this list of exceptions, because the category of illegal material overlaps heavily with the categories the exceptions preserve. The most serious forms of illegality, federal crimes, trafficking, and intellectual property violations, are precisely the forms that the exceptions name.

A word on each of the two most consequential exceptions, because their scope is often underestimated. Subsection 230(e)(1) is not a narrow carve-out for a handful of offenses. It preserves the enforcement of federal criminal law as a whole, and it names obscenity and sexual exploitation as examples, not as an exhaustive list. Fraud, extortion, threats transmitted in interstate commerce, the distribution of controlled substances, and the full range of federal criminal provisions remain enforceable against services without any interference from the provision. The preservation is addressed to enforcement, which means investigation and prosecution by the government, and it means that the provision cannot be raised as a defense in a criminal case. Separately, subsection 230(e)(2) keeps intellectual property entirely outside the provision’s reach. Copyright owners pursue their claims under the copyright statutes, trademark owners under trademark law, and the provision offers no shelter in either forum. The intellectual property exception is worth emphasizing because infringement is one of the most common forms of unlawful content online, and it is one of the forms most clearly excluded from the shield. A reader who believes the provision protects piracy, counterfeiting, or the misappropriation of creative work is mistaken about the text in a way that the subsection resolves directly.

The knowledge-based refinement of the claim raises its own textual question, and it, too, is answered by the provision. Nothing in the text conditions the shield on the service’s knowledge. The operative language bars treating the service as the publisher of third-party content, and it does so without a scienter requirement, without a notice-and-takedown procedure, and without a duty to investigate. A reader who holds that knowledge should matter is making a judgment about what the provision should say, and that judgment is the basis of several reform proposals discussed elsewhere in this series. But as a description of the provision, the claim that knowledge ends the immunity is false. The text draws no such line.

There is a second precision, concerning the word immunizes. Immunity in this provision means immunity from being treated as the publisher of someone else’s words. It does not mean immunity from the service’s own conduct. A service that creates or develops the content itself falls under the information content provider definition in subsection 230(f)(3) and loses the shield for that content. A service that conspires with a poster, solicits the unlawful material, or materially contributes to its creation is not protected as to that material. The courts have drawn this line repeatedly: the shield covers the passive host of third-party content, and it stops where the service’s own involvement in the content begins. A reader who believes the provision gives services a free pass for their own wrongdoing is reading a broader immunity than the text grants.

Why, then, does the claim persist, and what does it get right? The genuine cost the claim is pointing at is real. The civil bar does leave some victims of unlawful third-party content without a remedy against the service, and for those victims the distinction between the service and the author is cold comfort. When the author is anonymous, insolvent, or located beyond the reach of the courts, the provision’s channeling of liability toward the author functions, for practical purposes, as a channeling of liability toward no one. The victim’s loss is real, the service’s role in amplifying the loss is real, and the provision’s answer, that the service is not the publisher, does not undo either reality. The claim also gets something right about the limits of the criminal carve-out. Subsection 230(e)(1) preserves enforcement of federal criminal law, but criminal prosecution is a government decision, governed by resource constraints and prosecutorial discretion, and it is not a remedy available to the victim. A victim who cannot interest a prosecutor in the case is left with a civil system that the provision has closed off against the service. The victim cannot compel a prosecution, cannot direct investigative resources, and cannot recover damages through a process that the government alone controls. The honest treatment of the claim acknowledges this. The provision does not immunize illegal material as a category, and the exceptions are express and consequential. But the civil bar it erects has a real cost, and that cost falls on the people with the least ability to bear it. The claim is wrong about the text and right about the wound. An article that treated every critic as mistaken would miss the distinction, and would lose the readers who most need the textual precision to land.

The remedy gap has a mechanism worth describing, because it explains why the cost is not theoretical. Identifying an anonymous poster requires legal process, often a subpoena to the service, and sometimes the poster used no identifying information at all, or used the identifying information of someone else. Even when the poster is identified, the poster may be judgment-proof, lacking the assets to satisfy a damages award. And even when the poster has assets, the poster may be located in a jurisdiction where a judgment cannot be enforced. At each of these steps, the practical remedy narrows, until the only party left standing is the service, which the provision removes from the case. The result is not that the content was lawful. The result is that the unlawfulness had no defendant. The claim’s force comes from this sequence, and any honest treatment of the provision has to sit with it. The text is what it is: a liability-allocation rule with express exceptions, not a blanket immunization of illegal material. The wound is what it is: a set of victims for whom the allocation leaves nothing. Both can be true, and a reader who understands both is the reader this article is written for.

For the full treatment of the 2018 amendment and the trafficking carve-out, see the earlier article in this series on the FOSTA legislation.

Claim: Section 230 was struck down with the rest of the decency title

In its strongest form, the claim runs as follows. Congress enacted the Communications Decency Act in 1996 as Title V of the Telecommunications Act, a comprehensive effort to impose decency standards on the new medium of the internet. The Supreme Court held the heart of that enactment facially unconstitutional in Reno v. American Civil Liberties Union, 521 U.S. 844 (1997), rejecting the criminalization of indecent and patently offensive online expression. Because Section 230 arrived inside that same title, enacted by the same Congress for the same avowed purpose of cleaning up online discourse, it cannot be disentangled from the scheme the justices rejected. The immunity, on this account, was one instrument of a regulatory project found unconstitutional, and it fell, or should be understood to have fallen, with the rest.

A more careful version concedes that courts speak of severability but insists the concession changes little. The indecency prohibitions were the title’s reason for being; the immunity was a secondary accommodation tucked into a criminal statute. When the primary purpose failed constitutional review, the secondary accommodation lost its foundation. Holders of this view sometimes add that placement matters: the provision does not sit among the liability rules of the Communications Act, where an immunity would naturally belong, but inside a title devoted to obscenity and violence, a placement that marks it as part of the decency project rather than as an independent enactment.

The claim draws force from a simple and widely repeated premise: the Supreme Court struck down the Communications Decency Act. That premise circulates in classrooms, news coverage and public argument as a settled fact. From it, the conclusion follows by ordinary logic. If the Court invalidated the CDA, and Section 230 was part of the CDA, then the Court invalidated Section 230. The argument needs no theory of severability and no close reading of the opinion, because it treats the title as a single object that the Court either sustained or destroyed. Its strength lies in that simplicity: one famous case, one famous result, one inference.

Holders sometimes sharpen the claim into a charge of incoherence: a decency statute cannot coherently contain an immunity that frustrates decency enforcement, so the immunity must have been swept away with the prohibitions, whether or not any opinion says so expressly. On this telling, the absence of an express invalidation is a technicality, and the honest reading treats the title’s fate as single and indivisible.

The verdict is false, and the precision that decides it is a matter of public record. Reno v. American Civil Liberties Union, 521 U.S. 844 (1997), invalidated two provisions: 47 U.S.C. 223(a)(1)(B), which criminalized the knowing transmission of obscene or indecent messages to recipients under eighteen, and 47 U.S.C. 223(d), which criminalized the knowing display of patently offensive material to minors. The Court held those prohibitions facially unconstitutional under the First Amendment and, in the course of doing so, held that expression on the internet receives the full constitutional protection accorded to the printed page. The opinion said nothing about Section 230, because Section 230 was not before the Court. The plaintiffs, led by the American Civil Liberties Union, challenged the criminal prohibitions; no party asked the justices to pass on the immunity, and the judgment vacated only the provisions the complaint put at issue.

Section 230 reached the statute books by a different route than the claim supposes. Congress added it to the Communications Act through section 509 of the Telecommunications Act of 1996, Public Law 104-104, placing it inside Title V alongside the indecency provisions. The two halves of that title pulled in opposite directions. The indecency provisions imposed criminal liability for transmitting certain content; Section 230 shielded intermediaries from civil liability for content supplied by others. One half punished the movement of expression across networks, while the other protected the networks across which expression moved. The title’s shorthand name, the Communications Decency Act, attached to the whole package, but the package contained instruments with contrary functions.

That contrary function explains why the immunity survived. The challengers in Reno sought to enlarge the space for online expression by striking down speech restrictions; Section 230 was itself a speech-protective instrument, and no litigant had reason to attack it. The judgment therefore left it intact, and it has operated without interruption in the years after 1996. The supposed choice between the title’s two halves was never presented to a tribunal in a single case, because only the criminal half was ever challenged. The judiciary struck down what the plaintiffs challenged and left alone what nobody challenged, which is the ordinary operation of adversarial litigation rather than a paradox.

A further precision concerns what “the rest of the decency title” denotes. Title V of the Telecommunications Act bore the heading “Obscenity and Violence,” and its criminal provisions addressed obscenity, indecency and the protection of minors. Section 230, by contrast, announced findings about the internet as a forum for intellectual and political activity and declared a national policy of preserving a vibrant, competitive free market unfettered by regulation. The immunity answered an early state-court decision that had treated a service’s screening of user content as the mark of a publisher; Congress meant to remove the disincentive that decision created for voluntary moderation. In substance, the title contained a criminal code for indecent transmission and, a few pages later, a liability shield designed to encourage private screening of indecent material. The two were neighbors, not partners.

The opinion’s reasoning sharpens the point. The Court found the indecency provisions vague and overbroad, chilling protected expression among adults in the name of shielding minors, and it refused to extend the reduced First Amendment protection of broadcasting to the internet. The medium, the Court held, warranted the full protection of print. That holding concerned government-imposed criminal punishment of speech. Section 230 concerns the opposite question: whether private intermediaries face civil liability for speech supplied by others. The two questions share a constitutional neighborhood but involve different actors, different remedies and different doctrinal tests, which is why the invalidation of one never implied the invalidation of the other.

Nothing in the Reno opinion purported to vacate Title V wholesale. The decree addressed the challenged provisions, and the unchallenged remainder of the title, including Section 230, continued in force by the ordinary rule that a judgment binds the provisions before the court. Later courts have applied the immunity without pausing over its pedigree, because its pedigree presents no difficulty: Congress enacted it, the President signed it, no litigant challenged it and no court invalidated it. There is a final irony worth stating precisely, because it dissolves the claim’s intuitive appeal. Section 230(c)(2) protects providers that voluntarily restrict access to material they consider objectionable, which was Congress’s chosen mechanism for encouraging private screening. The title’s two instruments embodied two theories of achieving decency online: punishment by the state and encouragement of private action. The Court rejected the first theory as unconstitutional; the second has operated ever since. To say the immunity was struck down with the rest of the title is therefore wrong twice over: the rest of the title was not struck down in its entirety, and the immunity was the half most compatible with the constitutional order the Court described.

Congressional practice after Reno supplies a final confirmation. Congress has returned to the provision more than once, amending its terms and legislating around its edges, most notably through the 2018 amendments addressing sex trafficking. A legislature does not amend a dead letter. Each amendment presupposed a live, operative immunity and adjusted its boundaries, which is conduct inconsistent with the idea that the provision perished in 1997. The claim requires believing that Congress, the executive and the entire federal judiciary spent decades operating an invalidated statute without anyone noticing; the record shows the opposite, a provision continuously in force and repeatedly revisited.

The claim persists because the shorthand is nearly irresistible. The Communications Decency Act is remembered as a single famous failure: Congress tried to censor the internet, and the Supreme Court said no in Reno. That compressed memory treats the title as one object with one fate, leaving no room for the unglamorous truth that omnibus titles contain unrelated instruments. The indecency half was the famous half, litigated by the nation’s leading civil liberties organization in a landmark First Amendment case; the immunity half was, in 1997, an obscure provision noticed mainly by communications lawyers. Public memory kept the headline and discarded the carve-outs.

Legal commentary has sometimes abetted the compression. Casebooks and survey articles introduce the provision as “part of the Communications Decency Act” and then, pages later, recount Reno’s invalidation of “the Act,” and the hurried reader merges the two. The merger is understandable: few writers pause to enumerate which sections of a long title survived, and the immunity’s section number gives no hint of its distinct history.

The claim also gets something right, and the correction should acknowledge it. The pairing was genuinely odd. A statute avowedly devoted to decency ended up sheltering, from civil liability, the transmission of the same material the statute’s criminal half targeted. The surviving provision is invoked to explain why objectionable material remains accessible, which is the opposite of decency enforcement. That tension is real, and it fuels the suspicion that something must have gone wrong in the telling, that the immunity cannot have survived the wreck of the scheme around it. The suspicion is understandable; it is also misplaced. The immunity survived because it was never under attack. It was built for a different purpose, and the litigation that sank the criminal half never took aim at it.

A related confusion concerns timing. Many accounts describe Section 230 as “part of the CDA” and then describe the CDA as “struck down in Reno,” and the conclusion writes itself. Each step of that compressed history contains a grain of truth: the provision was enacted inside Title V, and Reno did strike down the title’s most prominent provisions. The error lies in the compression, which drops the distinction between a title and its provisions, between the challenged and the unchallenged, between a criminal prohibition and a civil immunity. Uncompressed, the history is straightforward: Congress criminalized indecent transmission and immunized intermediaries in the same title; the courts invalidated the criminalization and left the immunity unquestioned. Precision, rather than argument, is what the claim needs.

The full treatment of the title’s two halves, the litigation that divided them and the provisions that survived belongs to the companion history of the Communications Decency Act. That account traces Title V from its drafting through the Reno litigation, lays out which provisions the Court invalidated and which it left standing, and explains how an immunity designed to encourage voluntary screening came to sit inside a criminal statute. It carries the complete treatment of the pre-enactment case law that prompted Section 509, the congressional findings that accompanied the immunity, and the later history of the title’s surviving provisions. Readers who want the complete record of what the Court struck down, what it left alone and why the two halves shared a title will find it in that history, which also examines the broader pattern of which early internet statutes survived constitutional review and which did not.

Claim: The Supreme Court has upheld Section 230

In its strongest form, the claim treats the provision’s litigation record as the functional equivalent of Supreme Court approval. Section 230 has been construed by the federal courts of appeals in a broad and consistent manner since the Fourth Circuit’s decision in Zeran; district courts have applied that construction in case after case; and the Supreme Court, presented with repeated opportunities to narrow or revisit the immunity, has left the consensus undisturbed. The Court denied certiorari in the cases that would have tested the broad reading, and when it finally took a case squarely presenting a Section 230 question, Gonzalez v. Google LLC, 598 U.S. 617 (2023), it resolved the dispute without narrowing the immunity by a syllable. On this account, a doctrine that has survived every challenge, been applied uniformly across the circuits and emerged unscathed from the one Supreme Court case to present it has been upheld in every sense that matters.

A subtler version of the claim points to engagement rather than outcome. Justice Thomas’s extended discussions of the provision, in statements respecting the denial of certiorari in Malwarebytes, Inc. v. Enigma Software Group USA, LLC, 141 S. Ct. 13 (2020), and in Doe v. Facebook, Inc., 595 U.S. (2022), show the Court attending closely to the text and its history. Holders of this view read that attention as evidence that the immunity has been considered at the highest level and found sound; the broad reading, they argue, would not have endured repeated denials of certiorari if a majority of the justices regarded it as mistaken.

The claim draws additional force from practical certainty. A litigant invoking the immunity in federal court faces long odds of losing; the lower-court consensus operates, for all practical purposes, as settled law. To insist that the provision has never been “upheld” can sound like a technicality, a lawyer’s distinction without a difference. The strongest form of the claim therefore frames the technicality as the error: when every court to consider a question answers it the same way and the highest court declines every invitation to disagree, the result is an endorsement, whether or not it arrives in the form of a majority opinion.

The verdict is false. The Supreme Court has never construed Section 230 in a majority opinion, and “upheld” describes a merits holding the Court has never issued.

Gonzalez v. Google LLC, 598 U.S. 617 (2023), is the decision most often cited for the contrary proposition, and its disposition proves the point. The Court granted certiorari to consider whether Section 230 shielded algorithmic recommendations of third-party content, the first time the justices had agreed to address the provision’s scope directly. It did not answer the question. The companion case, Twitter, Inc. v. Taamneh, presented claims under the Anti-Terrorism Act’s aiding-and-abetting provision, and the Court held in Taamneh that the plaintiffs’ allegations failed to state a claim under that law. Having resolved Gonzalez’s claims on that independent ground, the Court vacated the Ninth Circuit’s judgment and remanded for reconsideration in light of Taamneh, declining to reach the Section 230 question it had granted certiorari to decide. A case the Court deliberately left undecided cannot be cited as a decision upholding the provision; the per curiam disposition contains no construction of the text, no holding about its scope and no endorsement of the lower-court consensus.

A genuine upholding would bear no resemblance to this record. It would be a majority opinion reaching the merits of a challenge to the provision, construing its terms and sustaining it against attack, in the manner of the Court’s decisions sustaining other federal enactments against constitutional challenge. No such opinion exists for Section 230, because no such challenge has ever reached the Court. The docket contains denials, a deliberate non-decision in Gonzalez and separate writings; it contains no holding, and holdings are the only currency in which “upheld” can be paid.

The separate writings sometimes invoked fare no better as authority. Justice Thomas filed statements respecting the denial of certiorari in Malwarebytes, Inc. v. Enigma Software Group USA, LLC, 141 S. Ct. 13 (2020), and in Doe v. Facebook, Inc., 595 U.S. (2022), each discussing Section 230 at length and questioning whether the lower courts had read the immunity too broadly. Those statements were separate writings by a single justice, attached to orders denying review; they were not majority opinions, and they bind no court. A denial of certiorari is not a ruling on the merits under the Court’s own practice; it expresses no view of the correctness of the decision below. To treat a statement respecting such a denial as an upholding of the provision inverts its significance: the orders in Malwarebytes and Doe left the lower-court judgments standing without endorsing them, which is the precise opposite of a merits affirmance.

The consequence is that the broad reading associated with Zeran rests on lower-court consensus rather than Supreme Court authority. That consensus is genuine and deep: every regional circuit to address the provision adopted the expansive construction, and district courts have applied it in case after case. But consensus among the circuits, however uniform, is not a holding of the Supreme Court, and the distinction carries practical weight. A future majority could construe the text differently without overruling any precedent of its own, because it has set none. The claim mistakes durability for ratification.

The distinction between survival and endorsement deserves emphasis, because it is the engine of the confusion. A provision survives when no court strikes it down; it is upheld when a court with authority to do so sustains it against a challenge on the merits. Section 230 has enjoyed the first fate in abundance. Its constitutionality has never been seriously questioned in litigation, no plaintiff has mounted a successful facial or as-applied attack, and Congress has legislated on the assumption of its validity, including through later amendments. None of that amounts to the second fate. The Court’s docket management, its resolution of Gonzalez on alternative grounds and its members’ separate writings are all consistent with a tribunal that has not spoken, not with one that has approved.

The Court’s practice gives the point its full weight. Certiorari is denied for reasons that have nothing to do with agreement: an unsuitable vehicle, a desire for further percolation among the circuits, or the press of a crowded docket. The rule that a denial imports no view of the merits is among the oldest housekeeping doctrines of the Court’s discretionary jurisdiction. Reading a string of denials as tacit approval therefore asks the orders to carry meaning the Court itself disclaims. The broad construction may be correct; the lower courts may have read the twenty-six words exactly as Congress wrote them. But correctness as a matter of first impression is not the same as endorsement by the highest court, and the claim asserts the second.

The open texture of the question cuts in both directions, which is why the falsehood matters beyond pedantry. Defenders of the broad reading cannot cite a Supreme Court holding when the consensus is challenged; critics of the broad reading cannot cite a Supreme Court doubt, since the Thomas statements are the views of one justice in non-precedential writings. Both sides argue from the same silence, and the silence belongs to neither. An honest account of the provision’s status therefore describes a deep lower-court consensus awaiting its first authoritative construction, a description that credits the consensus without promoting it to a rank it has not attained.

Precision about Gonzalez matters in particular, because the case dominates discussion. The petitioners asked the Court to hold that algorithmic amplification fell outside the immunity; the Court answered a different question entirely, about the pleading standard for aiding terrorism under a different statute. Observers who describe the outcome as a victory for the provision are describing a practical result, the immunity was not narrowed, rather than a legal one. The immunity emerged from the October 2022 term exactly as it entered it: construed broadly below, unconstrued above.

The claim persists because the provision’s record looks, from any practical vantage, like endorsement. Every serious challenge to the immunity’s existence has failed; the lower courts speak with one voice; the Supreme Court’s single encounter with the question left the consensus untouched. A reader who absorbs that record naturally concludes that the highest court has approved the doctrine, because in ordinary life an authority that repeatedly declines to intervene is understood to consent. The inference is natural, and it is wrong only in the technical sense that matters most in law: consent inferred from silence is not a holding.

What the claim gets right is the functional reality. For a litigant, a journalist or a legislator, the distinction between a unanimous circuit consensus and a Supreme Court holding rarely changes the advice. The immunity operates as settled law in the forums where cases are decided, and predictions about its application carry the confidence that only long uniformity can produce. The claim correctly perceives that stability; its error lies in assigning the stability to the wrong source, crediting the Supreme Court for work performed by the courts of appeals.

The absence of a construction also does argumentative work for both sides, which keeps the claim in circulation. Defenders invoke the Court’s inaction as quiet approval, a sign that the broad reading needs no correction. Critics invoke the same inaction as an open door, proof that the question remains live and the consensus vulnerable. Each side hears in the silence what its position requires, and each retelling hardens the impression that the Court has taken a view. The truth is less useful to everyone and more useful to the reader: the Court has taken no view, the consensus stands on its own foundations, and those foundations, strong as they are, have never been tested at the highest level.

There is a final, sympathetic reason the claim endures. “The Supreme Court has never ruled on it” sounds like an attack, as though the provision’s authority were somehow provisional or suspect. Holders of the claim resist that implication by upgrading survival into endorsement, defending the doctrine’s legitimacy against a perceived slight. The defense is unnecessary. A unanimous lower-court consensus, applied without interruption from 1996 forward and left undisturbed by the justices, is authority enough for every practical purpose. It is not a Supreme Court holding, and saying so diminishes nothing.

The complete treatment of the Court’s encounters with internet-liability questions belongs to the companion survey of internet-law Supreme Court cases. That account reconstructs the full procedural history of Gonzalez and its companion Taamneh, explains the certiorari practice that gives denials of review their limited significance, and traces how the lower-court consensus developed during the years the justices stayed their hand. It also examines the constitutional holdings that operate alongside the provision, the line of decisions addressing intermediary liability in other statutory contexts, and the reasons a future construction, when it comes, would write on a nearly blank slate at the Supreme Court level. Readers seeking the full docket history behind this claim’s verdict, and the doctrinal setting in which the next encounter will occur, will find it in that survey.

The text has no platform

The text has no platform. That sentence, offered as the namable claim of this article, repays a literal reading. The two words that structure nearly every public argument about the provision, “neutral” and “platform,” appear nowhere in it. The operative language is twenty-six words long: “No provider or user of an interactive computer service shall be treated as the publisher or speaker of any information provided by another information content provider.” Those words contain no neutrality condition, no distinction between publishers and platforms, and no vocabulary for the categories into which public argument insists on sorting intermediaries. The statute speaks of providers, users and information content providers; it assigns liability according to who supplied the content, not according to the posture in which the intermediary held itself out.

The correction matters because argument about intent cannot supply what the text omits. Debates over what Congress meant to accomplish in 1996, over whether the drafters envisioned neutrality or its opposite, proceed as though the answers were hidden in the legislative history and recoverable by sufficient research. The twenty-six words suggest a shorter path. A reader who has read them knows that no neutrality requirement can be extracted from a sentence that never mentions neutrality, and that no publisher-platform distinction can be enforced under a provision that never draws it. The most heated disputes in the field concern obligations the text never imposes and categories it never creates, which is why reading the provision resolves more of them than any amount of argument about its purposes.

That observation supplies the thesis of the series for this entry. Correcting the record is presented here as an exercise in competence rather than persuasion: a reader who can distinguish the text from the argument about the text is equipped for every subsequent dispute, including disputes whose outcomes this article does not predict. The field of internet-liability law is the field in which the gap between the words and the argument is widest, because the provision is short, the stakes are large and the vocabulary of public debate was invented long after the drafting. Competence in such a field consists largely of returning, claim by claim, to the words Congress enacted, and the ledger assembled across these sections is an instrument for that return.

The verdicts recorded here share a common shape. The neutrality condition is absent from the text. The publisher-platform distinction is a rhetorical construct without statutory footing. Repeal would encounter constitutional holdings that protect curation as expression, and the competing predictions about its effects remain contested because the mechanisms point in opposite directions. The immunity covers intermediaries of every size and description, not only the largest. Federal criminal law stands outside its scope by express carve-out. The provision survived Reno because it was never challenged there. And the Supreme Court has never construed it, which leaves the lower-court consensus authoritative in practice and unratified in precedent. Each verdict was reached the same way: by setting the claim beside the text, the cases or the doctrine, and recording what the comparison showed.

Two cautions accompany those verdicts, and the article’s framework requires them. The first is that several predictions about repeal are genuinely contested rather than false; the mechanisms each side invokes are real, and honesty requires presenting both rather than grading every claim a failure. The second is that the concerns motivating criticism of the provision are themselves real: the immunity does shield conduct that many find indefensible, and textual precision does not dissolve that discomfort. An article that treated every critic as mistaken would lose the readers who most need the precision, which is why the verdicts distinguish the false from the contested rather than grading every claim a failure.

For readers whose interest runs forward rather than backward, toward what might replace or revise the current regime, the companion comparison of Section 230 with the copyright safe harbors sets out what reform proposals actually propose. That treatment examines the mechanisms available to legislators, the models already operating in adjacent bodies of law and the trade-offs each model entails, without endorsing any of them. It is the natural next step for a reader who has finished correcting the record and wants to understand the menu of alternatives on its own terms.

The study of a short, consequential text rewards three habits: reading the words first, distinguishing holding from silence, and distrusting any claim that needs the text to say what it does not say. Those habits transfer to every statute the series examines. Readers who wish to continue the exercise will find the VaultBook legislation study notebook a suitable companion for working through the provisions, verdicts and cross-references collected here.

Frequently Asked Questions

Q: Does section 230 require platforms to be neutral?

No. Subsection (c)(1) contains no condition about viewpoint, balance, or editorial policy. Protection attaches to two facts: the defendant is a provider or user of an interactive computer service, and the content at issue came from another information content provider. Subsection (c)(2) strengthens the point by expressly protecting action voluntarily taken in good faith to restrict access to material the provider considers objectionable, which is the opposite of a neutrality mandate. A neutrality requirement would penalize the very removal decisions that subsection protects. The provision was enacted to reverse Stratton Oakmont, Inc. v. Prodigy Services Co., where a New York court imposed publisher liability because the service exercised editorial control; Congress replaced that regime with one that protects screening. The single stated condition is good faith, which asks whether the restriction genuinely targeted objectionable material, not whether the service’s overall environment is balanced. Courts have applied the protection categorically, without inquiring into editorial posture.

Q: Is there a publisher versus platform distinction in section 230?

No. The word platform appears nowhere in the provision, so the distinction has no statutory footing. The operative categories are defined terms: an interactive computer service under subsection (f)(2), and an information content provider under subsection (f)(3), meaning anyone responsible in whole or in part for the creation or development of information. The line the provision draws runs between those categories, not between publishers and platforms. A service keeps the (c)(1) protection for content it did not create or develop; it crosses the line only by becoming an information content provider for the material at issue. The word publisher does appear in (c)(1), but it names the liability the sentence removes, not a status the provision polices. The common law once sorted services by function, and the provision was written to eliminate that sorting. Curation, ranking, and removal do not convert a service into a publisher for liability purposes; creation or development does.

Q: Would repealing section 230 stop censorship?

No. Repeal would change the liability rules for hosted content, but it would not change the rule governing a service’s own curation, because that rule comes from the First Amendment. In Moody v. NetChoice, 603 U.S. 707 (2024), the Supreme Court held that a platform’s curation of the content it displays, selecting, ordering, excluding, and labeling third-party material, is expressive activity protected by the Constitution. A repeal could restore publisher and distributor liability for user-supplied material, but it could not compel a private service to carry expression it does not wish to display, because such compulsion would regulate the service’s own protected expression. The likely result is the opposite of what the claim predicts: liability for hosted content gives every service a new reason to remove material, and the incentive falls hardest on expression near the legal line. Under revived distributor liability, a complaint creates knowledge and knowledge creates exposure, so the cheapest compliance strategy is removal on complaint. Repeal produces liability without obligation.

Q: Does section 230 only protect big tech?

No. Subsection (f)(2) defines an interactive computer service by function, providing or enabling computer access by multiple users to a computer server, with no size threshold, revenue floor, employee count, or market-share test. Courts have applied the provision to local newspaper comment sections, municipal websites, consumer complaint sites, personal blogs, niche forums, dating sites, job boards, non-profit entities, and internet access providers. The operative immunity in (c)(1) also protects the user alongside the provider, so an individual who forwards or reposts another person’s content is covered, which makes ordinary users the most numerous class of beneficiaries. It is true that the practical benefit is worth more to large firms in absolute terms, because they host more content and face more lawsuits, but that is a fact about scale, not about the provision. A broad rule is always worth more to those with more exposure. The text draws no line by size, and courts have declined to insert one.

Q: Does section 230 immunize illegal content?

No. Subsection (e) contains express carve-outs. Subsection (e)(1) preserves the enforcement of federal criminal law as a whole, including the obscenity and sexual-exploitation provisions, so a service that knowingly hosts material constituting a federal crime remains exposed to prosecution. Subsection (e)(2) preserves intellectual property claims, leaving copyright and trademark actions to their own frameworks. Subsection (e)(5), added by the 2018 FOSTA amendment, preserves specified sex-trafficking claims. The provision is a liability-allocation rule, not a legalization rule: it bars treating a service as the publisher of third-party content in civil claims, while the underlying unlawfulness of the content and the liability of its author remain intact. It also draws no knowledge-based cutoff; Zeran v. AOL, 129 F.3d 327 (4th Cir. 1997), holds that notice of unlawful content creates no duty to remove it. And the shield stops where the service’s own involvement begins, since a service responsible in whole or in part for creating the content is an information content provider.

Q: Was section 230 passed to protect a young industry?

The provision was enacted in 1996, when the commercial internet was young, and Congress stated findings about the internet as a forum for intellectual and political activity and a national policy of preserving a vibrant competitive free market. But the text was written in size-blind, functional terms rather than as a temporary subsidy for an infant industry. The definitions in subsection (f) contain no sunset, no size threshold, and no limitation to services of a particular era, and Congress left those definitions untouched when it amended the exceptions in 2018. The motivating case was Stratton Oakmont, Inc. v. Prodigy Services Co., where a New York court imposed publisher liability on a service precisely because it screened content, creating a disincentive to moderate. Congress reversed that result to encourage voluntary screening without fear that screening would create liability. The protection therefore attaches to conduct, providing interactive access to third-party content, not to the age of the industry. A dial-up access provider with a few hundred subscribers qualifies on the same terms as the largest network.

Q: Was section 230 struck down along with the indecency provisions?

No. Reno v. American Civil Liberties Union, 521 U.S. 844 (1997), invalidated two provisions, 47 U.S.C. 223(a)(1)(B) and 223(d), which criminalized the knowing transmission and display of indecent or patently offensive material to minors. The opinion said nothing about Section 230 because no party challenged it; the plaintiffs sought to enlarge the space for online expression, and the immunity was itself a speech-protective instrument, so no litigant had reason to attack it. The judgment vacated only the provisions the complaint put at issue, and the unchallenged remainder of Title V, including Section 230, continued in force by the ordinary rule that a judgment binds the provisions before the court. The two halves of the title pulled in opposite directions: the criminal half punished the movement of expression across networks, while the immunity protected the networks carrying it. Congress later amended the provision, most notably in 2018, which a legislature does not do to a dead letter. The shorthand that the Court struck down the Communications Decency Act is accurate only for the indecency provisions.

Q: Is it a myth that the Supreme Court upheld section 230?

It is not a myth; the statement that the Court has never upheld the provision is accurate. No majority opinion has ever construed Section 230, because no challenge to it has reached the merits. In Gonzalez v. Google LLC, 598 U.S. 617 (2023), the Court granted certiorari on whether the immunity shielded algorithmic recommendations of third-party content, the first time it had agreed to address the provision’s scope directly, then resolved the case on the companion aiding-and-abetting question in Twitter, Inc. v. Taamneh and remanded without reaching the immunity issue. A case deliberately left undecided cannot be cited as an upholding. Denials of certiorari express no view of the merits under the Court’s own practice, and separate writings by a single justice, such as the statements respecting denial in Malwarebytes, Inc. v. Enigma Software Group USA, LLC, 141 S. Ct. 13 (2020), bind no court. The broad reading associated with Zeran v. AOL rests on a deep and uniform lower-court consensus, which operates as settled law in practice, but circuit consensus is not a Supreme Court holding.

Q: Would repealing section 230 improve content moderation?

Contested. Two mechanisms pull in opposite directions, and serious analysts endorse each. Mechanism one holds that liability produces more removal rather than better removal. Leaving up a defamatory post risks a judgment, while taking down a lawful post risks only a user complaint, so the one-sided arithmetic favors removal on complaint. Small services that cannot afford review teams would close comment sections and forums rather than improve them, and organized complainants could move enforcement without persuading anyone of anything. Mechanism two holds that liability produces investment. Vague standards and unanswered appeals persist because vagueness is cheap when no court tests it; attach a price to carelessness, and services hire reviewers, document decisions, and build functioning appeals, following the path traditional publishers took under full liability. Both mechanisms rest on the same premise, that legal exposure changes behavior, and differ on which change dominates. No repealed baseline exists to test either prediction, since the provision has applied in substantially the same form since 1996, so the dispute is empirical rather than logical.

Q: What does good faith mean under section 230(c)(2)?

Subsection (c)(2) protects any action voluntarily taken in good faith to restrict access to or availability of material that the provider or user considers to be obscene, lewd, lascivious, filthy, excessively violent, harassing, or otherwise objectionable, whether or not the material is constitutionally protected. Good faith is the single stated condition, and it qualifies the voluntary act of restriction. It asks whether the restriction was genuinely aimed at material the provider considers objectionable, not whether the service’s overall environment of expression is balanced. The operative phrase is considers to be: the judgment is the provider’s own, exercised according to its own lights. Good faith therefore does not convert into a neutrality requirement; it presupposes the opposite, that the provider is making its own judgments about what counts as objectionable. A companion clause, (c)(2)(B), protects providing others with the technical means to restrict access, which is to say it protects the distribution of filtering tools whose purpose is viewpoint-laden selection by the end user.

Q: Does section 230 protect recommendation algorithms?

The Supreme Court has never decided the question. In Gonzalez v. Google LLC, 598 U.S. 617 (2023), the Court granted certiorari on whether the immunity shields algorithmic recommendations of third-party content, the first time it had agreed to address the provision’s scope directly, then resolved the case on the companion aiding-and-abetting question in Twitter, Inc. v. Taamneh without reaching the immunity issue. Lower courts have generally treated the algorithmic arrangement of third-party content as within the protection, on the reasoning that ranking, promoting, or recommending material provided by another does not make the service the provider of that material. The operative question under (c)(1) remains the source of the content: where the information was provided by another information content provider, the service is not treated as its publisher or speaker. Curation does not convert the service into a content creator under the defined terms. That reading rests on lower-court consensus, however, not on Supreme Court authority, and a future construction could draw the line differently.

Q: Can a service lose section 230 protection after learning about unlawful content?

No. Nothing in the text conditions the shield on the service’s knowledge. The operative language bars treating the service as the publisher of third-party content without a scienter requirement, without a notice-and-takedown procedure, and without a duty to investigate. In Zeran v. AOL, 129 F.3d 327 (4th Cir. 1997), the Fourth Circuit held that the provision bars distributor liability as well as publisher liability, rejecting the argument that notice of defamatory content creates a duty to remove it. Distributor liability is the rule that a party with notice of defamatory material that fails to remove it answers for it, and the decision holds that rule inapplicable to covered services. The protection is categorical: once the content comes from another information content provider, the service is not treated as its publisher, and no inquiry into knowledge follows. A reader who holds that knowledge should end the immunity is making a judgment about what the provision should say, which is the basis of several reform proposals, but as a description of the provision the claim is false.

Q: When does a service become a content provider under section 230?

A service becomes an information content provider, and loses the (c)(1) protection for the material at issue, when it is responsible in whole or in part for the creation or development of that information, under subsection (f)(3). The boundary turns on authorship and development, not on curation. Ranking, promoting, demoting, removing, or arranging third-party expression does not cross the line, because none of those functions appears in the definition. The line is crossed where the service’s own involvement in the content begins: creating the material itself, soliciting its creation, or materially contributing to its development. Courts have drawn this boundary repeatedly, holding that the shield covers the host of third-party content and stops where the service becomes responsible for the content’s substance. Because the definition reaches responsibility in part as well as in whole, partial contribution can be enough. The test is applied item by item: the same service may be a protected intermediary for one post and an unprotected content provider for another.

Q: What did the 2018 FOSTA amendment change in section 230?

The Allow States and Victims to Fight Online Sex Trafficking Act, Public Law 115-164, enacted April 11, 2018, added subsection 230(e)(5), the first and only substantive narrowing of the immunity since 1996. The new exception provides that the immunity does not bar federal sex-trafficking prosecutions, state criminal charges under laws mirroring the federal offense, or civil claims by victims under the federal trafficking remedy where the underlying conduct violates the trafficking statute. The act also created a separate federal offense at 18 U.S.C. 2421A for owning or operating a service with intent to promote or facilitate prostitution, with enhanced penalties for aggravated conduct. The definitions in subsection (f) were left exactly as they were, so the breadth of the protected categories did not change. The amendment presupposed a live, operative immunity and adjusted its boundaries, which is consistent with the provision’s continuous force since 1996 and inconsistent with any notion that it had perished when the indecency provisions fell.

Q: Does section 230 bar distributor liability as well as publisher liability?

Yes. In Zeran v. AOL, 129 F.3d 327 (4th Cir. 1997), the first federal appellate decision construing the provision, the Fourth Circuit held that (c)(1) bars distributor liability as well as publisher liability and rejected the argument that notice of defamatory content creates a duty to remove it. At common law, distributors such as bookstores and libraries were liable for defamatory material only with knowledge of it, a lower standard than the publisher standard. The Zeran court held that the provision’s bar on treating a service as the publisher or speaker of third-party content forecloses the distributor theory too, treating the protection as categorical: once the content comes from another information content provider, the service is not treated as its publisher, and no inquiry into the service’s knowledge or editorial posture follows. The court explained that the provision was meant to protect services attempting to screen offensive material and to avoid a rule that would punish the effort to moderate. That reading has anchored the case law ever since.

Q: Why do people say platform when the statute never uses the word?

Several forces keep the word in circulation. It is a serviceable commercial term for the businesses the provision covers, and a vivid term of argument, so claims conducted in it feel concrete. It also carries a genuine legal memory: before the provision existed, courts sorted online services by function, treating CompuServe as a distributor in Cubby, Inc. v. CompuServe, 776 F. Supp. 135 (S.D.N.Y. 1991), and Prodigy as a publisher in Stratton Oakmont, Inc. v. Prodigy Services Co., because it exercised editorial control. The provision was written to eliminate that sorting, but the intuition that active curation differs from passive carriage survived the displacement, and platform supplied a new name for the old distinction. The word publisher inside (c)(1) reinforces the confusion, since readers take it as a behavioral description rather than as the liability treatment the sentence forbids. Concreteness in argument is not presence in text, however, and the provision’s silence on the word is not an oversight for intuition to fill.

Q: Does section 230 protect a site that edits user posts?

It depends on what the editing does. The (c)(1) test asks whether the information was provided by another information content provider. A site that corrects spelling, trims length, or formats a user’s post without becoming responsible in whole or in part for the creation or development of the information remains on the protected side of the line, because the content is still the user’s. The line is crossed where the site’s own involvement makes it responsible in whole or in part for the content’s creation or development under subsection (f)(3): writing the substance, soliciting the unlawful material, or materially contributing to its development. Courts have drawn the boundary at the service’s own authorship rather than at the fact of touching the material. The provision does not grant a blanket editing privilege, and it does not withdraw protection at the first alteration either. The question is always about the specific item of content: who is responsible for its creation or development, the user or the service.

Q: How does section 230 differ from the DMCA safe harbors?

The two regimes answer the same practical question, what a service must do about unlawful user content, with opposite designs. Section 230 is an unconditional immunity: it attaches automatically, with no procedural prerequisites, and it survives the service’s actual knowledge of unlawful content, as Zeran established. The DMCA safe harbors at 17 U.S.C. 512 are conditional: a service keeps the harbor only by registering a designated agent, responding expeditiously to proper notices, maintaining a repeat-infringer policy, and lacking disqualifying knowledge or financial benefit. Under Section 230 nothing is required of the service; under the DMCA, failing any condition removes the defense, though it does not itself create liability. The immunity covers essentially all civil claims arising from third-party content except the express carve-outs, while the safe harbor covers copyright alone. The pairing is complementary by design: the immunity’s intellectual-property exclusion in (e)(2) is precisely why copyright needed its own regime. Most serious reform proposals are, in substance, proposals to make the communications provision resemble the copyright one.

Q: If section 230 does not require neutrality, where does the neutrality claim come from?

Several sources feed it. The bargain metaphor does the most work: the immunity is described as a deal in which Congress gave services a shield in exchange for evenhandedness, though no clause states that exchange. The Stratton Oakmont memory supplies the legal form: a court once held that editorial control invited publisher liability, and the claim restates that rule as though the provision had preserved it rather than reversed it. The good-faith language in (c)(2) is misread as a behavioral standard the service must satisfy across its operations, when it qualifies only voluntary restriction decisions. And the governance concern is real: a small number of services set the terms on which vast amounts of expression reach an audience, under standards the public does not write. That is a genuine question about how large services should moderate, but it is addressed to the wrong text. The provision answers who pays for the speech of others, not how the speech of others should be ordered, and silence on the governance question is not a hidden neutrality rule.

Q: Has any court narrowed section 230 for bad-faith moderation?

The good-faith language appears only in subsection (c)(2), which protects action voluntarily taken in good faith to restrict access to objectionable material. It qualifies the voluntary restriction decision, not the (c)(1) shield, which contains no good-faith condition and attaches on the service’s status and the content’s source alone. Courts have applied (c)(1) categorically, without inquiring into the service’s editorial posture or its motives for hosting or removing material. No line of decisions has made the (c)(1) immunity turn on whether the service moderated in good faith; the provision’s text gives courts no such inquiry to conduct. Disputes about good faith arise at the margins of (c)(2), where the protection is expressly conditioned on it, not as a general limitation on the immunity. A service that removes material it does not genuinely consider objectionable may fall outside (c)(2)’s protection for that removal, but the (c)(1) bar on publisher treatment for third-party content stands on its own terms.