Section 230 of the Communications Act carries more public argument per word than almost any other provision in federal law. The core of the provision, the sentence that does most of the work, runs to twenty-six words: 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. Around those words has grown a debate conducted largely without them. The provision is described as a single immunity, as a special privilege for large technology companies, as an absolute shield against liability for harmful content, and as the legal basis for every content moderation decision. None of those descriptions survives contact with the text. The statute contains two distinct protections rather than one. It names individual users alongside providers in both. It conditions the take-down protection on good faith while imposing no motive requirement on the leave-up protection. It defines the two terms that decide most cases, and it lists the categories of law it does not touch.

This article reads the provision the way a researcher reads a statute: in order, with the words doing the work. The first task is placement. The provision did not arrive as a freestanding declaration about the internet. It arrived as section 509 of the Telecommunications Act of 1996, inside a title of that act that mostly restricted speech, and its survival when the rest of that title fell tells something important about what it is. The second task is the twenty-six words themselves, read clause by clause, because the disputes that sound the most dramatic usually turn on the plainest terms. The third is the distributor question, the early interpretive fight that fixed the breadth of the leave-up protection. The fourth is the pair of definitions in subsection (f) that determine when the protection is lost.

Then come the second protection, the exclusions, and the user half of the statute that public argument tends to miss. A sorting discipline closes the analytical work. Presented with any dispute, the equipped reader asks first whether the complaint targets content the defendant left up or content the defendant took down, and then reads the matching subsection. The provision is broad rather than absolute, and the difference is written into its own words.

Congress wrote the provision to solve a specific problem. In the first half of the 1990s, two courts reached opposite conclusions about online services that hosted user content. The service that did not moderate was treated as a mere distributor with light liability, while the service that screened its forums was treated as a publisher with heavy liability. The law rewarded passivity and punished cleanup. The amendment that became section 230 broke that incentive by protecting services both for what their users post and for what the services take down. The two purposes sit side by side in the statute’s own policy statement, and the two operative subsections give each purpose its own legal form. Most error about the law comes from confusing the two.

Section 230: Where the Provision Sits in the Statute

Section 230 did not arrive as a standalone declaration about the internet. It arrived as the answer to a specific legal trap, and the shape of that trap explains why the provision contains two different protections instead of one. In 1991, a federal court in Cubby, Inc. v. CompuServe, Inc. treated an online service that did not moderate its forums as a mere distributor, liable for defamatory user postings only if it knew or had reason to know of them, the same standard applied to bookstores and newsstands. Four years later, a New York trial court in Stratton Oakmont, Inc. v. Prodigy Services Co. reached the opposite result for a service that did moderate: because Prodigy exercised editorial control over its bulletin boards, deleting messages it considered offensive, the court held Prodigy to the standard of a publisher, liable for user postings regardless of knowledge. Together the two decisions created what commentators called the moderator’s dilemma. A service that left everything standing was treated gently, while a service that tried to clean up its forums took on the heaviest possible liability for whatever slipped through. The incentive ran exactly backward: the law rewarded hands-off operation and punished the attempt to moderate.

Representatives Christopher Cox and Ron Wyden designed their amendment to break that incentive. Cox spoke on the House floor on August 4, 1995, describing a provision that would protect services that screened offensive material for their customers, so that acting as a good Samaritan would no longer carry the price of publisher liability. The amendment was titled “Protection for Private Blocking and Screening of Offensive Material,” and that title survived into the statute, where it still sits as the heading of section 230. Congress enacted the language as section 509 of the Telecommunications Act of 1996, Public Law 104-104, which President Clinton signed on February 8, 1996. Section 501 of that act gives the short title of its Title V as the Communications Decency Act of 1996, which is why the provision is often described as part of the Communications Decency Act even though its legislative home is section 509 of the larger telecommunications bill. The provision was codified at 47 U.S.C. 230, in the Communications Act of 1934.

That placement matters because Title V was overwhelmingly a speech-restricting title, and section 230 was the exception that pointed the other way. Most of the Communications Decency Act consisted of criminal prohibitions, principally amendments to 47 U.S.C. 223, making it unlawful to transmit obscene or indecent material to minors over interactive computer services. In Reno v. ACLU, 521 U.S. 844 (1997), decided June 26, 1997, the Supreme Court struck down the act’s indecency provisions as facially overbroad under the First Amendment, holding that the statute’s imprecision chilled a large amount of protected adult speech in the course of shielding minors. The Court severed the unconstitutional “indecency” language and left the rest of the title standing where it could stand. Section 230 survived that wreckage untouched, and the reason is structural rather than accidental. The provisions that fell were restrictions on speech backed by criminal penalties. Section 230 is not a restriction on anything. It is an allocation of liability, a rule about who can be sued for whose words, and it imposes no duty on any speaker or service to say or to refrain from saying anything at all. A liability shield does not trigger the First Amendment analysis that destroyed the criminal provisions, so the shield endured while the surrounding title collapsed. The history of that title and the court’s 1997 decision are traced in the series article on the Communications Decency Act of 1996.

Congress prefaced the operative language with two framing subsections that announce what the provision is for, and they deserve attention because they hold the statute’s two purposes in a single view. Section 230(a) sets out five congressional findings: that the growing array of interactive computer services represents an extraordinary advance in educational and informational resources; that these services give users a great degree of control over the information they receive; that they offer a forum for a true diversity of political discourse and wide opportunities for cultural development and intellectual activity; that they have flourished to the benefit of Americans with a minimum of government regulation; and that Americans increasingly rely on interactive media for political, educational, cultural, and entertainment purposes. The findings paint the services as valuable, user-driven, and best left largely unregulated, which is the posture behind the leave-up protection. Section 230(b) then states the policy of the United States in five paragraphs (a sixth was added in 2018): to promote the continued development of the internet and other interactive computer services; to preserve the vibrant and competitive free market for those services, unfettered by federal or state regulation; to encourage the development of technologies that maximize user control over received information; to remove disincentives for the development and use of blocking and filtering technologies that let parents restrict children’s access to objectionable material; and to ensure vigorous enforcement of federal criminal laws against obscenity, stalking, and harassment by computer. The fourth paragraph is the one that points at the take-down protection: Congress wanted services to build and use filtering tools without fear that the very act of filtering would convert them into publishers. The two purposes sit side by side in the same subsection, and the two operative provisions of subsection (c) give each purpose its own legal form.

The statute also marks its own boundaries. Section 230(e), titled “Effect on other laws,” withholds the section from federal criminal law, intellectual property law, communications privacy law, and inconsistent state law claims, and a 2018 amendment added a fifth exclusion for specified sex trafficking claims. These exclusions are genuine limits, and they are examined in full in the section on what the provision does not touch. A service facing a federal criminal prosecution, an intellectual property claim, or a qualifying sex trafficking action cannot invoke section 230 as a defense, no matter how broadly the immunity provisions are read. The shield is broad rather than absolute, and subsection (e) is where Congress wrote the edges.

Two further features of the statutory setting complete the picture. Subsection (d) imposes a modest affirmative obligation that is easy to overlook: a provider of interactive computer service must notify its customers, at the time of entering a service agreement and in a manner the provider deems appropriate, that parental control protections such as hardware, software, or filtering services are commercially available to limit minors’ access to harmful material, and must identify or point to current providers of such protections. This is a disclosure duty rather than a liability rule, and it underscores the statute’s user-empowerment theme: Congress paired immunity for private blocking with an expectation that providers would tell customers the tools exist. The legislative history confirms the design. The conference report on the Telecommunications Act of 1996 described one of the specific purposes of the section as overruling Stratton Oakmont and similar decisions that had treated providers and users as publishers of content that was not their own because they had restricted access to objectionable material, and it tied the provision to the federal policy of empowering parents to determine the content of communications their children receive through interactive computer services. The “Good Samaritan” caption of subsection (c) is therefore not ornament. It names the conduct Congress most wanted to encourage: voluntary private action to screen offensive material, undertaken without the fear that the act of screening would itself create liability.

Which two decisions created the moderator’s dilemma that section 230 answered?

In Cubby, Inc. v. CompuServe, Inc. (1991), a federal court treated an unmoderated service as a distributor with light liability. In Stratton Oakmont, Inc. v. Prodigy Services Co. (1995), a New York court treated a moderated service as a publisher with heavy liability. Together they rewarded services for leaving content alone and punished services for screening it.

Who wrote the amendment that became section 230?

Representatives Christopher Cox and Ron Wyden drafted the amendment in the House. Cox defended it on the floor on August 4, 1995, arguing that services which screened offensive material for customers should not pay publisher liability for acting as good Samaritans. Its title, Protection for Private Blocking and Screening of Offensive Material, survives as the heading of section 230.

Why did section 230 survive when the rest of Title V fell?

In Reno v. ACLU (1997), the Supreme Court struck down the Communications Decency Act’s indecency provisions as overbroad and severed them. Section 230 survived because it restricts nothing: it allocates liability for other people’s words instead of penalizing speech, so the First Amendment analysis that destroyed the criminal provisions never reached it.

What do the findings in subsection (a) declare?

Subsection (a) records five findings: interactive computer services are an extraordinary educational and informational resource, they give users unusual control over what they receive, they host a true diversity of political discourse, they have flourished under minimal government regulation, and Americans increasingly rely on them for political, educational, cultural, and entertainment purposes.

What duty does subsection (d) impose on providers?

Subsection (d) requires a provider, when a customer enters a service agreement, to notify that customer that parental control protections, including hardware, software, and filtering services, are commercially available to limit minors’ access to harmful material, and to identify or point to providers of such protections.

The Twenty-Six Words

Few sentences in federal law carry the weight of the one at the center of this article. The provision that generates the most argument in American technology law does the overwhelming majority of its work in twenty-six words, and precision about those words is the entire subject of this section. The disputes that sound the most dramatic usually turn on the plainest of the terms, so the sentence deserves to be read slowly, clause by clause, before any argument about what it means can be evaluated.

The text, in full, reads: “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.”

Every word in the sentence does work, and the sentence contains no surplus. The first point worth registering is the pair of beneficiaries. “No provider or user” extends the protection to two categories at once. Providers are the companies and organizations that operate online services, and the word reaches them without regard to size, from the largest platform to the smallest forum. Users are the individuals who use those services, and their inclusion is one of the most consequential and least appreciated features of the sentence. An individual who forwards, reposts, or otherwise republishes content created by someone else is a user of an interactive computer service, and the sentence bars treating that individual as the publisher or speaker of the reposted content. The protection therefore covers ordinary internet users in their daily activity, not only the companies that operate the services. A reader who assumes the provision is a special shield for technology companies has missed half of its beneficiaries.

The next phrase, “of an interactive computer service,” identifies the category of service whose providers and users receive the protection. Congress defined the term separately, in subsection (f), and the definition is deliberately broad, reaching information services, systems, and access software providers that give multiple users access to a computer server, including services that provide access to the internet itself. The breadth of the definition means that the sentence applies across the full range of online activity, from internet access providers to discussion forums to websites of every size. The definitions are examined in detail in the section on the two defined terms, because the way the two definitions interact is where most of the difficult cases are decided.

The operative command of the sentence is “shall be treated as.” Congress did not write “shall not be liable,” and it did not create an affirmative privilege in so many words. Instead it directed courts not to apply a particular legal characterization: the defendant shall not be treated as the publisher or speaker. The formulation matters because it operates at the level of the theory of liability rather than at the level of a defense to be proved. A claim that requires the defendant to be the publisher or speaker of the content fails at the threshold, because the statute forbids the court to make the characterization the claim requires. This is the textual foundation for the way courts have applied the provision early in litigation, often on a motion to dismiss, rather than treating it as a defense to be developed through discovery. The phrasing also explains the reach of the provision beyond defamation. Any cause of action that depends on treating the defendant as the publisher or speaker of third-party content is barred, regardless of the label the plaintiff attaches to the claim, which is why courts have applied the sentence to negligence claims, housing discrimination claims, and other theories that have nothing to do with libel.

“The publisher or speaker” names the two roles that the provision takes off the table. A publisher is the party legally responsible for originating and disseminating content, and a speaker is the party who utters or expresses it. By naming both, Congress covered the two ways a defendant might be characterized as the source of the words at issue. The pairing also set up the question that dominated the first decade of litigation under the provision, which is whether a third category, the distributor, survived outside the sentence. At common law, distributors such as bookstores and libraries faced liability only when they knew or had reason to know of unlawful content, a lighter standard than the strict liability borne by publishers. The sentence says nothing about distributors in so many words, and the question whether the omission was deliberate became the central interpretive dispute. The Fourth Circuit answered it in 1997, in the decision examined in the next section, holding that the sentence bars distributor liability as well, because distributor liability is a subset of publisher liability and because a notice-based regime would defeat the purposes Congress set out in the statute.

“Of any information” states the scope of the content covered, and the word “any” does the work. The sentence contains no limitation to particular categories of content, no exception for especially harmful material, and no distinction between text, images, video, or other forms of information. The breadth is not accidental. Congress wrote the protection to cover information generally, and then handled the categories of law it wished to preserve through separate exclusions in subsection (e), examined in the section on what the provision does not touch. The structure is worth noting: the sentence itself is stated without qualification, and the qualifications live elsewhere in the section. A reader who looks only at the sentence will conclude the protection is absolute, and a reader who reads the whole section will see that it is broad rather than absolute.

Two further word choices in the sentence repay attention. Congress wrote “publisher or speaker” rather than “publisher” alone, and the addition of “speaker” extends the bar beyond the print metaphor. A speaker is the party that utters content, and the pairing covers spoken and broadcast forms, live audio, and any other mode of expression in which the originator is better described as speaking than publishing. Congress likewise wrote “information” rather than “speech” or “expression.” The choice avoids importing the doctrinal baggage of the First Amendment’s speech categories and keeps the provision focused on the functional question of who supplied the content. The sentence does not ask whether the material is political commentary, commercial advertising, or something else entirely. It asks who provided it, and the breadth of “any information” is what lets the same twenty-six words govern defamation claims, negligence claims, and the other theories courts have considered.

The final phrase, “provided by another information content provider,” contains the sentence’s most important limitation. The word “another” means that the protection applies only when the content originated with someone other than the defendant. If the defendant is itself the information content provider of the material at issue, the sentence does not apply, because the information was not provided by another. The term “information content provider” is defined in subsection (f) as any person or entity responsible, in whole or in part, for the creation or development of information, and the phrase “in whole or in part” is where services lose the protection. A service that merely transmits content created entirely by its users is not the information content provider of that content. A service that participates in creating or developing the content, even partially, is. The boundary between transmitting and developing is the most litigated question in the provision’s history after the distributor question, and it is examined in the section on the two definitions.

One further feature of the sentence deserves emphasis, which is what it omits. The sentence contains no requirement of knowledge, no notice procedure, no duty to investigate, and no good faith qualifier. A defendant that has been notified of allegedly unlawful content receives the same protection as one that has never heard of it, so long as the other elements are satisfied. The omission is deliberate, and it distinguishes the sentence from the neighboring subsection (c)(2), which expressly requires good faith for the separate protection it gives to voluntary content moderation. The asymmetry reflects a legislative judgment. Congress wanted services to be free to leave content up without fear that knowledge would convert them into liable parties, and it wanted them to be free to take content down without fear that the act of moderating would convert them into publishers. The first freedom lives in the twenty-six words. The second lives in the subsection examined later in this article, and confusing the two is the error that the closing of this article addresses.

Taken together, the sentence establishes a simple framework with difficult edges. The simple framework is that providers and users of interactive computer services are not to be treated as the publishers or speakers of content provided by someone else. The difficult edges are the meaning of “treated as,” the fate of distributor liability, and the line between a provider that transmits content and one that develops it. The next sections take those edges in order.

Publisher or Speaker, and the Distributor Question

To understand what the twenty-six words accomplished, it helps to know the common law categories they displaced. For centuries, the law of defamation sorted participants in the spread of information into publishers and distributors, and the two categories carried very different consequences. A publisher was the party that originated and disseminated content: the newspaper, the book publisher, the broadcaster. Publishers faced strict liability for defamatory statements they published, meaning they could be held liable even without knowledge of the defamation. A distributor was the party that delivered or transmitted the work of another: the bookstore, the library, the newsstand. Distributors faced liability only if they knew or had reason to know of the defamatory character of the material, a fault-based standard far lighter than the one applied to publishers. The distinction made sense in a world of physical media, where a bookstore could not plausibly read every book on its shelves but a newspaper could review every article before printing it.

The early internet forced courts to fit online services into these inherited categories, and the results were the two inconsistent decisions described in the preceding section: the service that did not moderate kept the lighter distributor standard, while the service that exercised editorial control was held to the publisher standard. The twenty-six words answered that pair of decisions by forbidding courts to impose the publisher characterization at all, and the distributor question then became the central interpretive dispute: did the distributor category survive outside the sentence, since the sentence speaks only of publishers and speakers?

The dispute that tested the design arrived within a year of the statute’s enactment. On April 25, 1995, an unidentified person posted a message on an America Online bulletin board advertising T-shirts and other merchandise commemorating the April 19, 1995 bombing of the Alfred P. Murrah Federal Building in Oklahoma City. The advertisement listed slogans mocking the attack, including one reading “Visit Oklahoma … It’s a BLAST!!!”, and it directed interested buyers to call “Ken” at a Seattle telephone number. The number belonged to Kenneth Zeran, a Seattle resident who had nothing to do with the posting and who learned of it only when strangers began calling him. Zeran called America Online the same day to report the hoax. The company removed the posting but declined his request to publish a retraction. The next day, another anonymous poster placed a similar advertisement, with slogans the trial court described as at least as vulgar and offensive as the first, including lines reading “Forget the rescue, let the maggots take over (Oklahoma 1995)” and “Finally a day care center that keeps the kids quiet (Oklahoma 1995).” Further postings followed. Five days after the first advertisement appeared, Zeran was receiving abusive calls every two minutes, including death threats, and he could not change his telephone number because he needed it for his business. He sued America Online for negligence.

Zeran’s legal theory was careful. He did not argue that America Online was a publisher of the hoax advertisements. He argued that the company was a distributor, like a bookstore, and that as a distributor it owed a duty once notified: a duty to remove the defamatory postings promptly, to notify its subscribers of the hoax, and to screen for similar postings in the future. His argument had two doctrinal legs. First, distributor liability and publisher liability were distinct categories at common law, and section 230(c)(1) spoke only of publishers and speakers, so distributor liability survived the statute. Second, statutes that displace common law principles are traditionally construed narrowly, which counseled against reading the provision to extinguish the distributor category. The district court in the Eastern District of Virginia rejected the theory and dismissed the case. Zeran appealed to the United States Court of Appeals for the Fourth Circuit.

The Fourth Circuit affirmed in Zeran v. America Online, Inc., 129 F.3d 327, decided November 12, 1997, in an opinion by Judge J. Harvie Wilkinson III. The court’s reasoning proceeded in three steps. First, it held that distributor liability is a subset of publisher liability rather than a separate category that the statute left intact. Publication, in the legal sense, includes communicating defamatory matter to third parties, and a distributor that transmits content to readers engages in publication. The common law distinguished publishers from distributors for the purpose of assigning different fault standards, but both were species of publisher liability. Congress’s direction that providers not be treated as publishers or speakers therefore covered distributors as well. As the court put it, the simple fact of notice could not transform a publisher into a distributor, because the statute forbade the publisher characterization in the first place.

Second, the court held that the statute bars lawsuits seeking to hold a service provider liable for exercising a publisher’s traditional editorial functions, such as deciding whether to publish, withdraw, postpone, or alter content. This formulation became the working test that courts applied for years afterward. The question was not the label the plaintiff chose for the claim but the nature of the activity the claim would punish. If liability would attach because the provider published third-party content, or declined to withdraw it, or failed to screen it, the claim treated the provider as a publisher and the statute barred it. Zeran’s negligence claim failed this test because each of the duties he asserted, the duty to remove promptly, the duty to notify subscribers, the duty to screen future postings, would have made the provider liable for its editorial choices about third-party content.

Third, the court held that a notice-based liability regime would defeat the statute’s purposes. Congress had identified two goals: avoiding an unwarranted chilling of online speech, and removing the disincentive to self-regulation that the Stratton Oakmont decision had created. Liability upon notice would have undermined both. A service facing millions of postings could not investigate every complaint of defamation, and the rational response to notice-based liability would have been either to remove content automatically upon complaint, chilling lawful speech through a heckler’s veto, or to abandon moderation entirely to avoid acquiring the knowledge that triggered the duty. Either outcome contradicted what Congress had set out to achieve. The court also rejected the narrow-construction argument, holding that Congress’s intent to displace the common law framework was plain on the face of the statute.

The holding was sweeping. Section 230, the court wrote, creates a federal immunity to any cause of action that would make service providers liable for information originating with a third-party user of the service. The word immunity is the court’s, and it reflects the practical effect of the “shall not be treated as” formulation: claims that depend on the forbidden characterization cannot proceed. The court also noted the preemption provision in subsection (e)(3), which provides that no cause of action may be brought and no liability may be imposed under any state or local law inconsistent with the section, confirming that Congress meant the protection to override the state common law theories on which Zeran had relied. Zeran sought review in the Supreme Court, which declined to hear the case in 1998, leaving the Fourth Circuit’s reading in place.

The Zeran reading became the foundation on which the rest of the case law was built. Other federal circuits adopted it in the years that followed, applying it to dismiss defamation, negligence, and related claims against service providers in cases including Batzel v. Smith in the Ninth Circuit in 2003 and Green v. America Online in the Third Circuit in 2003, and district courts followed the same path. The “traditional editorial functions” formulation gave lower courts a workable test, and the rejection of notice-based liability gave the provision its distinctive breadth: a service remains protected even after being told that unlawful content sits on its service, so long as the service did not create or develop the content itself.

That breadth has drawn criticism from the beginning, and the criticism deserves to be stated with the same care as the holding. Some commentators have argued that the Zeran court read the statute more broadly than Congress intended, pointing to the text’s reference to publishers and speakers and to the common law distinction the court collapsed. Others have argued on policy grounds that barring distributor liability removes the incentive for providers to act against content they know to be unlawful, allowing a provider that has been notified of defamation or worse to do nothing without legal consequence. Defenders of the reading respond that the text supports it, that the dual purposes Congress stated require it, and that a notice-based regime would produce the chilling effects and the abandonment of moderation that the court described. The competing assessments turn on contested judgments about legislative intent and about how providers would behave under alternative rules, and they are taken up directly in the series article on section 230 myths. What the text article must record is that the 1997 reading is the one the courts adopted, that it has anchored the case law ever since, and that its central move was to treat the distributor question as answered by the publisher language rather than left open by it.

The Two Definitions That Decide Everything

The twenty-six words turn on two defined terms, and Congress placed the definitions in subsection (f) of the same section. The first definition, “interactive computer service,” determines who can claim the protection. The second, “information content provider,” determines when the protection is lost. In practice the second definition does far more work than the first, because Congress defined the first category broadly and the second as the boundary. Nearly every close case under the provision is a fight about whether the defendant, in addition to providing the service, also helped create or develop the content at issue.

Subsection (f)(2) defines “interactive computer service” to mean any information service, system, or access software provider that provides or enables computer access by multiple users to a computer server, including specifically a service or system that provides access to the internet and such systems operated or services offered by libraries or educational institutions. The definition is deliberately capacious. It reaches internet access providers, the companies that sell connectivity to homes and businesses. It reaches platforms and forums of every kind, from the largest social networks to small discussion boards. It reaches websites of every size that let multiple users reach a server through their computers, which in the decades after 1996 came to mean nearly every interactive site. The specific mention of libraries and educational institutions shows that Congress was thinking beyond commercial providers and intended the category to be functional rather than tied to a business model. Courts have applied the definition generously, treating it as a threshold that almost any online service clears. The generosity is purposeful: Congress wanted the protection to turn on what the defendant did with the content, not on what kind of service the defendant happened to be.

Subsection (f)(3) defines “information content provider” to mean any person or entity that is responsible, in whole or in part, for the creation or development of information provided through the internet or any other interactive computer service. This is the definition that decides cases, and each of its phrases carries weight. “Any person or entity” means the category is not limited to individuals or to companies; a service provider can itself be an information content provider with respect to particular content, and often the question in litigation is whether the defendant wore both hats at once. “Responsible, in whole or in part” means that partial responsibility suffices. A defendant need not be the sole author of the content to lose the protection; contributing to its creation or development, even alongside the user who supplied it, is enough. “Creation or development” reaches beyond literal authorship. Creation covers writing or producing the content outright. Development covers shaping, soliciting, or materially contributing to content that someone else supplied, and it is this word that has generated the most litigation. The definition contains no requirement that the development be intentional wrongdoing; the question is responsibility for the content’s creation or development, not the defendant’s state of mind about its lawfulness.

The leading decision on the boundary is Fair Housing Council of San Fernando Valley v. Roommates.com, LLC, 521 F.3d 1157, decided en banc by the United States Court of Appeals for the Ninth Circuit on April 3, 2008, in a majority opinion by Chief Judge Alex Kozinski, with a dissent by Judge M. Margaret McKeown. The case concerned a roommate-matching website operated by Roommate.com, LLC. As a condition of using the service, each new subscriber had to create a profile by answering a questionnaire. The questionnaire did not invite free-form answers to its central questions. Instead it required subscribers to choose from pre-populated drop-down menus, and the questions solicited the subscriber’s sex, sexual orientation, and family status, along with the subscriber’s preferences regarding the sex, sexual orientation, and family status of a future roommate. The site then assembled each subscriber’s answers into a profile page, used the answers to match subscribers with one another, and sent subscribers emails about housing opportunities matching their stated preferences. The site also offered an “Additional Comments” section where subscribers could write open-ended text about themselves and the roommate they sought.

Two fair housing councils sued, alleging that the questionnaire violated the federal Fair Housing Act and parallel California laws, which prohibit discriminatory statements in connection with housing. The district court granted summary judgment for the website on section 230 grounds, treating the site as a provider of an interactive computer service that merely published information supplied by its users. The Ninth Circuit, sitting en banc, reversed in the part that mattered. The court held that by requiring subscribers to answer questions from pre-populated menus that solicited protected characteristics and discriminatory preferences, the website was responsible, in whole or in part, for the creation or development of that information. The site had not merely transmitted what users volunteered. It had designed the questions, supplied the limited set of answers, made answering mandatory, and then assembled and distributed the resulting profiles. In the court’s analysis, the website was a co-developer of the questionnaire content, and therefore an information content provider of that content, and the twenty-six words did not protect an information content provider from claims based on its own content. The information was not “provided by another” when the provider itself had shaped it.

The court was careful to limit the reach of its holding, and the limits are as instructive as the holding. The “Additional Comments” section of the site, the open-ended text box where subscribers wrote in their own words, remained protected. The site gave users no guidance steering them toward unlawful content in that box, and the court treated the provision of a blank field as a neutral tool rather than development. The opinion drew a general principle from the contrast: providing neutral tools to carry out what may be unlawful or illicit searches does not amount to development for purposes of the provision, while substantial affirmative conduct that promotes the unlawful use of those tools can cross the line. A search engine that filters content only by user-supplied criteria, for example, remains protected, because the engine does not contribute to the content’s character. A site that requires users to express illegal preferences as the price of admission does not.

The opinion also addressed the kind of ordinary editorial activity that does not convert a service into a content provider. A website operator that edits user content in the usual ways, such as correcting spelling, removing obscenity, or trimming submissions for length, does not become responsible for the creation or development of the content. The distinction the court drew was between activity that shapes the message’s unlawfulness and activity that merely presents it. The questionnaire in Roommates materially contributed to the alleged unlawfulness of the resulting profiles, because the site itself had solicited the discriminatory content through its mandatory questions and constrained answers. That material-contribution analysis has become the standard formulation: a service loses the protection when it contributes materially to the alleged unlawfulness of the content, not when it merely hosts, organizes, or lightly edits what users supply. The housing law context of the underlying claims is covered in the series guide to the Fair Housing Act of 1968.

The contrast with an earlier Ninth Circuit decision sharpens the point. In Carafano v. Metrosplash.com, Inc., decided in 2003, the court protected an online dating service that used a detailed questionnaire with dozens of questions and pre-prepared responses, even though a third party had used the questionnaire to post a false and harmful profile of the plaintiff. The difference was that the dating site’s classifications did nothing to enhance the defamatory sting of the message or to encourage the wrongdoing; the site’s structure was neutral with respect to the unlawfulness at issue. In Roommates, by contrast, the site’s questions were designed to elicit precisely the discriminatory preferences that the housing laws prohibited. The two decisions together establish that questionnaires and structured inputs are not automatically fatal to the protection and not automatically safe. The question is always what the structure contributed to the specific information at issue.

The user side of the protection received its own leading decision in Batzel v. Smith, 333 F.3d 1018, decided by the Ninth Circuit in 2003. The defendant, Ton Cremers, operated a website and email list devoted to museum security and stolen art. He received an email from Bob Smith accusing Ellen Batzel of possessing paintings looted during the Second World War, selected the message for distribution, made minor edits, and sent it to his list subscribers. Batzel sued both Smith and Cremers for defamation. The court held that Cremers, as a user of an interactive computer service distributing content provided by another, was protected, and that selecting the message for publication and editing it lightly did not make him responsible for its creation or development. The decision added an important refinement for cases where the author’s intent is disputed. Smith claimed he had not meant the email for public distribution, and the court held that the question for the distributor’s protection was not the author’s private intent but the reasonable perception of the provider or user: immunity applies where a reasonable person in the distributor’s position would conclude the information was provided for publication on the internet or another interactive computer service. The ruling illustrates the users point in concrete form. An individual who passes along another person’s content, even after choosing it and touching it up, stands in the same protected position as the largest platform, provided the content remains another’s.

The Roommates facts also show how assembly itself can count as development. The website did not merely host the answers subscribers selected. It required the answers as a condition of use, constrained them to pre-populated menus of its own design, assembled them into profile pages displaying each subscriber’s pseudonym, description, and preferences, and then distributed those pages to other subscribers and used them to generate housing-opportunity emails. Each step compounded the site’s responsibility for the resulting information. The court’s analysis treated the questionnaire, the constrained answers, and the assembled profiles as a single course of conduct in which the site participated in developing the content at every stage.

Two further features of the Roommates decision deserve notice. First, the court analyzed the protection content by content rather than site by site. The same website was an information content provider with respect to the questionnaire answers and a protected provider with respect to the open-ended comments. Immunity under the provision depends on the pedigree of the particular information at issue, so a defendant can win as to one portion of its service and lose as to another. Second, the decision resolved only the section 230 question, not the underlying housing law claims. The court remanded for determination whether the questionnaire actually violated the Fair Housing Act, and in later proceedings the courts concluded that it did not. The section 230 holding stands independently of that merits outcome: the case is authority for when a service becomes a content provider, not for when housing questionnaires violate the housing laws. The dissent, meanwhile, warned that the majority’s approach narrowed the provision’s protection and risked exposing services to liability for the ordinary structuring of user submissions, a disagreement that reflects the genuine difficulty of drawing the development line.

The practical guidance that emerges is straightforward to state and demanding to apply. A service crosses into content creation when it solicits specific unlawful content, when it structures user inputs to elicit such content, or when it materially shapes content in ways that contribute to its unlawfulness. A service stays on the protected side when it provides neutral tools, blank fields, and organizational features that leave the substance of user expression to the users. Between those poles lies the territory where most litigation happens, and the Roommates framework, co-development plus material contribution, is the map courts use to navigate it.

With the text, the distributor question, and the definitions in place, the provision’s structure comes into focus as two protections rather than one. Subsection (c)(1), the twenty-six words, addresses liability for content the provider leaves up: it bars treating the provider or user as the publisher or speaker of information provided by someone else. Subsection (c)(2) addresses liability for content the provider takes down. Its first part shields good faith voluntary action to restrict access to objectionable material, whether or not the material is constitutionally protected. Its second part shields providing others with the technical means to restrict access to such material. The two subsections ask different questions and impose different requirements. Under (c)(1), the question is who created or developed the content, and the answer turns on the information content provider definition. Under (c)(2), the question is whether the removal was voluntary, in good faith, and directed at material within the objectionable categories the statute lists. A dispute about a platform leaving content up is a (c)(1) dispute, to be analyzed through authorship and the distributor analysis. A dispute about a platform taking content down is a (c)(2) dispute, to be analyzed through good faith and the objectionable-material categories. Much public argument about the statute proceeds as though there were a single provision doing a single job, and the argument that platforms hide behind the statute both to leave content up and to take content down treats as one thing what Congress wrote as two. A reader who can say which subsection is at issue in a given dispute, and what each subsection requires, is equipped to evaluate any claim about the law. A reader who cannot is not.

The Second Protection

Much public argument about section 230 proceeds as though the statute contains a single immunity. It contains two. Subsection (c)(1) addresses whether a provider or user of an interactive computer service may be treated as the publisher or speaker of information supplied by someone else. Subsection (c)(2) addresses something else entirely: whether a provider or user may be held liable for restricting access to material, or for giving others the technical means to restrict access. The first provision concerns disputes about content left up. The second concerns disputes about content taken down. Confusing them produces a large share of the muddled commentary about the law, because each paragraph carries its own triggers and its own limits, and the requirements of one do not apply to the other.

Subsection (c) is titled “Protection for ‘Good Samaritan’ blocking and screening of offensive material,” and paragraph (2) within it is titled “Civil liability.” The text of subparagraph (A) reads in full:

“No provider or user of an interactive computer service shall be held liable on account of (A) 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;”

Every phrase in that sentence does work, and the sentence repays slow reading. “No provider or user of an interactive computer service shall be held liable on account of” sets the scope of actors and the kind of lawsuit barred. The actors are providers and users alike, the same pairing that appears in (c)(1), so the takedown protection belongs to individuals as much as to companies. The lawsuits barred are those that seek to hold the actor liable because of the described restriction: a poster suing a forum for deleting a thread, a business suing a review site for removing a listing, a user suing a service for suspending an account. The paragraph does not immunize everything a service does. It immunizes liability imposed on account of the specific actions the subparagraphs describe.

“Any action voluntarily taken in good faith to restrict access to or availability of material” contains the paragraph’s operative conditions. “Voluntarily” marks the action as the actor’s own initiative, the kind of editorial choice a publisher makes when it decides what to carry. “In good faith” is the paragraph’s genuine limiting principle and the reason (c)(2) cannot be described as unconditional. A removal carried out honestly, to keep material the actor finds objectionable off the service, satisfies the requirement. A removal carried out as a pretext, for instance to sabotage a competitor while invoking a content policy as cover, would not. Good faith turns on honesty of purpose, not on whether a court agrees with the underlying judgment about the material. The statute asks whether the actor acted sincerely, not whether it acted correctly.

The next phrase sets the standard for what may be restricted: “material that the provider or user considers to be obscene, lewd, lascivious, filthy, excessively violent, harassing, or otherwise objectionable.” The operative verb is “considers.” The standard is subjective. A court applying (c)(2)(A) does not ask whether the material was in fact obscene, lewd, or harassing. It asks whether the provider or user considered it so. A federal court stated the point directly: the paragraph does not require that the material actually be objectionable, but rather protects blocking material the provider or user considers objectionable. That subjectivity is deliberate. It allows services and individuals to moderate without first winning an argument, in court, about the true character of the content. The alternative would make every moderation decision the opening of a trial about the content itself.

“Whether or not such material is constitutionally protected” extends that logic one step further. The actor need not establish that the removed material falls outside the protection of the First Amendment. The shield applies even to speech the Constitution would protect against government suppression. This clause severs the moderation decision from constitutional litigation. A service does not have to prove the content was unprotected speech to be shielded for removing it, and a plaintiff cannot defeat the protection by showing the removed material was constitutionally protected expression. The paragraph thus gives moderators certainty without requiring them to become constitutional scholars.

The list of categories ends with a catch-all: “or otherwise objectionable.” The enumerated categories (obscene, lewd, lascivious, filthy, excessively violent, harassing) give way to a general phrase that reaches whatever else the actor considers objectionable. Courts have generally read the catch-all broadly, treating it as covering material beyond the listed examples. Commentators have debated whether the phrase should be narrowed by the interpretive canon under which general words take color from the specific words around them, and the Ninth Circuit has expressed doubt about applying that canon to this paragraph. The debate over the catch-all’s outer boundary belongs to the courts and the commentary; what the text establishes beyond dispute is that Congress wrote an open-ended category rather than a closed list, and paired it with the subjective “considers” standard.

Subparagraph (B) protects a different act. Its text reads:

“or (B) any action taken to enable or make available to information content providers or others the technical means to restrict access to material described in paragraph (1).”

Where (A) shields the actor’s own restriction decisions, (B) shields the act of giving others the tools to restrict. Parental control software, spam filters offered to account holders, blocklists a service distributes, user-operated mute and block functions, and third-party filtering products all fall within enabling or making available the technical means of restriction. The phrase “material described in paragraph (1)” ties the clause to information provided by another information content provider, the same third-party content at the heart of (c)(1). And “or others” extends the protection beyond content providers to anyone who receives the tools.

The textual differences between (A) and (B) matter. Subparagraph (B) contains no “good faith” qualifier, no “voluntarily,” and no list of objectionable categories. Congress wrote distinct requirements for distinct acts: the actor who restricts on its own judgment must act in good faith toward material it considers objectionable, while the actor who supplies filtering tools is protected for the supplying. Section 230(d) reinforces this design by obliging providers, when entering service agreements with customers, to notify them that parental control protections are commercially available and to identify current providers of such protections. The statute contemplates an ecosystem of filtering tools, and (c)(2)(B) shields those who build and distribute them.

The practical upshot is a clean division of labor between the two paragraphs. A dispute about a service removing posts, suspending accounts, limiting distribution, or otherwise restricting access is a (c)(2) dispute. The questions it raises are the ones (c)(2) asks: was the action voluntary, was it taken in good faith, did the actor consider the material objectionable, or did the actor supply the technical means of restriction. A dispute about a service leaving posts up is a (c)(1) dispute, which asks none of those questions and instead asks whether the defendant is being treated as the publisher or speaker of someone else’s content. A reader who cannot say which paragraph is at issue in a given controversy is not equipped to evaluate any claim about what the law permits, because the two paragraphs protect different conduct under different conditions.

The title Congress gave the subsection explains the problem it solves. A “Good Samaritan” provision protects the actor who intervenes to reduce harm. Before the statute, the law created a perverse incentive: a service that screened content risked being treated as the publisher of everything it carried, while a service that screened nothing could claim the lighter treatment of a mere distributor. Rational operators faced a choice between moderating and thereby assuming broad liability, or refusing to moderate at all. Subsection (c)(2) breaks that dilemma by protecting the intervention itself. A service that takes down objectionable material does not thereby forfeit the (c)(1) protection for what remains, and it gains an independent shield for the takedown. The two paragraphs work as a pair: one protects hosting, the other protects cleaning up.

The verbs of (c)(2)(A) are deliberately broad: “restrict access to or availability of.” The paragraph does not enumerate technical methods, and its protection is not limited to outright deletion. Removing a post, suspending an account, limiting the distribution of content, placing material behind an age gate, or quarantining suspected spam all restrict access to or availability of material within the ordinary meaning of the words. What matters is the function, not the mechanism. A plaintiff cannot evade the paragraph by arguing that the service demoted rather than deleted, or suspended rather than removed, because each of those actions restricts access or availability on the paragraph’s own terms.

The asymmetry between the two paragraphs deserves emphasis because it is the textual proof of the article’s central claim. Subsection (c)(1) asks no questions about motive. It does not require good faith, voluntariness, or any judgment about the content; it asks only whether the defendant is being treated as the publisher of another’s information. Subparagraph (c)(2)(A), by contrast, conditions its protection on good faith and on the actor’s own assessment of the material. This difference means a takedown challenged as pretextual receives a good-faith inquiry that a hosting decision never faces, while a hosting decision challenged as careless receives no motive inquiry at all. Readers who apply one paragraph’s requirements to the other’s disputes are not reading strictly; they are reading the wrong paragraph.

What the Section Does Not Touch

Subsection (e), titled “Effect on other laws,” lists the bodies of law that section 230 leaves alone. The list is the clearest textual evidence that the protection, though broad, is not absolute. Each exclusion preserves a category of claims that proceeds exactly as it would have without the statute, and each has practical consequences for litigants who assume the section answers every question about online liability. The five exclusions are numbered (e)(1) through (e)(5): federal criminal law, intellectual property law, state law, communications privacy law, and, since a 2018 amendment, specified sex trafficking claims.

The first exclusion, (e)(1), preserves federal criminal law. Its text reads: “Nothing in this section shall be construed to impair the enforcement of section 223 or 231 of this title, chapter 71 (relating to obscenity) or 110 (relating to sexual exploitation of children) of title 18, or any other Federal criminal statute.” The named provisions cover federal obscenity offenses and the sexual exploitation of children, and the closing phrase, “any other Federal criminal statute,” sweeps in the rest of federal criminal law. The references to sections 223 and 231 of “this title” point back into the Communications Act itself, the title where section 230 is codified, so the exclusion preserves enforcement of the neighboring criminal provisions of the same act alongside the title 18 offenses. The consequence is straightforward: section 230 is no defense to a federal criminal prosecution. A service whose operators distribute child sexual abuse material cannot invoke the statute against charges under chapter 110 of title 18. A prosecution for federal obscenity offenses under chapter 71 proceeds untouched. Most courts to consider the question have held that this exclusion does not preserve civil lawsuits merely because the underlying conduct also violates a criminal law; the carve-out is for criminal enforcement itself, not for private claims dressed in criminal-law language. The distinction matters because it keeps the exclusion from swallowing the protection: if every civil claim alleging criminal-like conduct escaped section 230, little of the immunity would remain.

The criminal-law exclusion is one-directional. “Nothing in this section shall be construed to impair the enforcement” preserves existing criminal enforcement; it does not create new offenses or expand prosecutorial authority. Section 230 takes nothing away from the criminal law and adds nothing to it. Relatedly, most courts to consider the question have held that the exclusion does not allow civil plaintiffs to evade the statute by framing their claims as violations of criminal law. A private lawsuit alleging conduct that also happens to be criminal remains subject to (c)(1) if it treats the defendant as the publisher of third-party content. Without that limitation, the exclusion would swallow the protection, since much tortious conduct can be described in criminal terms.

The second exclusion, (e)(2), preserves intellectual property law: “Nothing in this section shall be construed to limit or expand any law pertaining to intellectual property.” Courts have read “intellectual property” to encompass copyright and trademark claims, among others. This is why copyright disputes over user-uploaded material proceed under the Digital Millennium Copyright Act rather than under section 230. The DMCA, at 17 U.S.C. 512, establishes its own safe harbors for online service providers, conditioned on requirements section 230 never imposes: designating an agent to receive infringement notices, acting expeditiously to remove material upon proper notice, and maintaining a policy for terminating repeat infringers. A copyright owner whose film appears in full on a video-sharing site sends a takedown notice under the DMCA framework, and the resulting dispute turns on compliance with that framework, not on section 230. Section 230 simply does not enter the analysis. The same holds for trademark claims. A brand owner challenging listings for counterfeit goods under trademark law litigates that claim on trademark principles; section 230 neither blocks the claim nor strengthens it. The exclusion reflects a legislative choice to leave the intellectual property system, with its own carefully balanced immunities and liabilities, exactly where it stood. The contrast between the two intermediary-liability regimes is compared in detail in the series article on section 230 and the DMCA.

The intellectual property exclusion is neutral in the same way the criminal exclusion is one-directional. “Nothing in this section shall be construed to limit or expand any law pertaining to intellectual property” neither creates intellectual property liability nor reduces it. A copyright or trademark claim stands or falls under the intellectual property laws as they exist, with their own immunities, safe harbors, and defenses. The trademark example illustrates the point: a brand owner alleging that a marketplace hosts listings for counterfeit goods litigates under trademark principles, including any contributory liability doctrines trademark law provides. Section 230 neither blocks that claim nor strengthens it. The provision simply steps aside, leaving the intellectual property system, with its own balances between owners and intermediaries, exactly where Congress left it in other statutes.

The third exclusion, (e)(3), addresses state law: “Nothing in this section shall be construed to prevent any State from enforcing any State law that is consistent with this section. No cause of action may be brought and no liability may be imposed under any State or local law that is inconsistent with this section.” The two sentences work together. A state law that is consistent with section 230 remains fully enforceable; a state or local law that is inconsistent with it is preempted. The practical line runs through the nature of the claim. A state consumer-protection or contract claim that targets a service’s own promises or its own conduct, without treating the service as the publisher of third-party content, is consistent with the section and may proceed. A state defamation or negligence claim that seeks to hold the service liable for hosting or failing to remove a user’s post is inconsistent with the section and is barred. The paragraph thus preserves state regulatory authority while preventing states from achieving through local law what the federal statute forbids: publisher liability for third-party content.

The mechanics of that preemption deserve a final note. The second sentence of the paragraph, barring any cause of action or liability under state or local law inconsistent with the section, functions as an express federal ceiling. Courts applying it focus on the consistency question the text poses: does the state claim, as pleaded, treat the defendant as the publisher or speaker of third-party content, or does it target the defendant’s own conduct. Claims framed under state law but functionally identical to defamation, such as negligence claims alleging a failure to remove a user’s post, are treated as inconsistent and barred. Claims that genuinely sound in the defendant’s own promises or misrepresentations, such as a promissory estoppel claim based on a service’s specific assurance, have sometimes survived, because they do not ask the court to treat the service as anyone’s publisher. The line is drawn claim by claim, from the pleadings, which is why litigants on both sides draft their complaints and motions with this paragraph in mind.

The fourth exclusion, (e)(4), preserves communications privacy law, naming “the Electronic Communications Privacy Act of 1986 or any of the amendments made by such Act, or any similar State law.” That 1986 Act is the federal framework for the privacy of electronic communications. It includes the prohibitions on the interception of wire, oral, and electronic communications, codified at 18 U.S.C. 2511; the stored-communications provisions, at 18 U.S.C. 2701 and following, governing unauthorized access to stored electronic messages; and the pen-register and trap-and-trace provisions that followed. The exclusion means a claim that a service unlawfully intercepted or disclosed users’ private communications cannot be answered with section 230. Suppose a messaging service were accused of reading and disseminating the contents of users’ private messages in violation of the wiretapping prohibitions. The service could not defeat that claim by arguing it was merely hosting third-party content, because the claim would not treat it as a publisher at all; it would treat it as an interceptor. The exclusion confirms that result in the text. Privacy claims of this kind live in a different statutory regime, with their own elements and defenses, and section 230 neither helps nor hinders them. The paragraph also preserves similar state communications-privacy laws, so a state wiretapping claim proceeds on the same footing.

The fifth exclusion arrived by amendment. The Allow States and Victims to Fight Online Sex Trafficking Act of 2017, Public Law 115-164, enacted April 11, 2018, added paragraph (e)(5) providing that, apart from the good-faith removal protection of (c)(2)(A), nothing in the section shall be construed to impair or limit three categories of claims: “(A) any claim in a civil action brought under section 1595 of title 18, if the conduct underlying the claim constitutes a violation of section 1591 of that title; (B) any charge in a criminal prosecution brought under State law if the conduct underlying the charge would constitute a violation of section 1591 of title 18; or (C) any charge in a criminal prosecution brought under State law if the conduct underlying the charge would constitute a violation of section 2421A of title 18, and promotion or facilitation of prostitution is illegal in the jurisdiction where the defendant’s promotion or facilitation of prostitution was targeted.” Section 1591 of title 18 is the federal sex trafficking statute, covering trafficking by force, fraud, or coercion and trafficking of minors; section 1595 gives victims a civil remedy. Section 2421A, created by the same 2018 law, targets owning, managing, or operating an interactive computer service with the intent to promote or facilitate prostitution. The amendment applies regardless of whether the underlying conduct occurred before, on, or after its enactment date.

A concrete example shows the exclusion at work. A trafficking victim brings a civil action under 18 U.S.C. 1595 against a website, alleging the site’s operators knowingly facilitated the trafficking operation in violation of section 1591. Before 2018, the site would have argued that section 230 barred the claim because it rested on third-party postings. After the amendment, that argument fails for claims within the three enumerated categories: the statute expressly withholds its protection. The parenthetical reservation, “other than subsection (c)(2)(A),” preserves one piece of the shield: a provider that in good faith removes suspected trafficking content remains protected for the removal itself. Congress thus withdrew immunity from the facilitation claims while keeping the incentive to take the material down. The amendment’s full legislative history and its effects are taken up in the series article on FOSTA and section 230.

The 2018 amendment carries its own statement of purpose. In the sense-of-Congress section, Congress declared that section 230 was never intended to provide legal protection to websites that unlawfully promote and facilitate prostitution or to websites that facilitate traffickers in advertising the sale of unlawful sex acts with trafficking victims. The operative change then carved the three categories of sex trafficking claims out of the immunity, while the new section 2421A of title 18 created a federal offense for owning, managing, or operating an interactive computer service with the intent to promote or facilitate prostitution. The amendment’s effective-date provision made the carve-out applicable regardless of whether the underlying conduct occurred before, on, or after enactment. Supporters of the change argued it closed a gap that had allowed facilitation to shelter behind the publisher provision; opponents of the change warned it would push platforms toward over-removal of lawful speech about sex work and make trafficking harder to detect. Those competing predictions are contested claims about consequences, and they belong to the myths article. The text establishes the legal change itself: specified trafficking-related claims proceed without regard to section 230, while good-faith removal of suspected trafficking content remains protected under (c)(2)(A).

Taken together, the exclusions draw a defined perimeter around the protection. Federal crimes, wiretapping and privacy violations, copyright and trademark infringement, consistent state laws, and specified sex trafficking claims all proceed without regard to section 230. A reader who absorbs this list understands the statute’s scale: it displaces the publisher-liability theories that would otherwise attach to hosting third-party content, and it leaves everything else to the laws that already govern those wrongs. The protection is broad within its domain and absent outside it.

The Protection Covers Users Too

The twenty-six words of (c)(1) protect “provider or user” of an interactive computer service, and the pairing is not decorative. An individual who forwards, reposts, shares, or links to another person’s content is covered by the same sentence that protects the largest platforms. The statute draws no distinction between a company operating a service with millions of accounts and a person operating a small discussion forum, a group chat, or a neighborhood mailing list. If the content was provided by another information content provider, and the claim would treat the forwarder as its publisher or speaker, the protection applies regardless of the forwarder’s size or commercial status.

The practical reach of this point is easy to underestimate. Ordinary internet use consists largely of passing along other people’s expression: forwarding an email thread, reposting a restaurant review, sharing a neighbor’s post about a contractor, quoting a comment into a new discussion, linking to an article with a few words of endorsement. Under traditional defamation principles, every repetition of a defamatory statement counts as a publication, so each act of forwarding or reposting would create fresh exposure. Without the user half of (c)(1), the internet’s most basic function, sharing, would be a liability trap for everyone who uses it. The statute prevents that outcome by extending the same publisher protection to the individual sharer that it extends to the platform host. A person sued for reposting someone else’s allegedly defamatory review can invoke the same twenty-six words as the service that hosted the original. The Batzel decision discussed earlier, in which the Ninth Circuit protected an individual who selected another person’s email for distribution to a mailing list, is the leading illustration.

The point extends to the small-scale operators who sit between pure users and large providers. A hobbyist who runs a forum for a local sports league, a volunteer who administers a discussion group, a blogger who hosts comments: each is a provider or user of an interactive computer service, and each receives the statute’s protection for third-party content appearing in the space they maintain. When such an operator removes a disruptive post, (c)(2) covers the removal as well, since that paragraph also names providers and users. The statute’s design thus protects the full stack of online participation, from the person who clicks share to the person who runs the server, without requiring any of them to be a particular kind of business.

Why is this widely missed in public argument? The debate about section 230 is conducted almost entirely as a debate about large platforms. Headlines describe the law as platform immunity, commentary weighs the responsibilities of major technology companies, and reform proposals are framed around the conduct of the biggest services. In that framing, the user half of the phrase quietly drops out, and the statute comes to sound like a special dispensation for corporations. The text never limits itself that way. Every dispute framed as being about what platforms may do is also, by the statute’s terms, about what individuals may do when they handle other people’s content.

The limits that apply to providers apply to users as well. The protection does not cover a user’s own authored words; the person who writes the defamatory post is its information content provider and cannot invoke the statute against claims arising from that authorship. Nor does it cover the user who materially contributes to the unlawful content, since anyone responsible in whole or in part for the creation or development of information falls within the statutory definition of an information content provider for that content. A user who edits another’s post to add a false accusation, or who solicits and shapes unlawful material, has crossed from protected forwarder to content creator. The boundary is functional, not titular: what matters is whether the person contributed to the creation or development of the content at issue, not whether the person is called a user or a provider.

The stakes for ordinary users are the reason the point belongs in any precise account of the law. Reform proposals that would narrow or condition the protection are often debated as though only companies would feel the effects. Because the text protects users equally, changes to the publisher provision would alter the legal position of every person who forwards, reposts, or shares. A reader who understands that the statute shields the individual sharer as well as the platform host is equipped to ask, of any proposed change, whose conduct it would actually reach.

Concrete situations show how far the user coverage reaches. An employee forwards a message containing a false claim about a colleague’s misconduct; a neighbor shares a post accusing a local business of fraud; a volunteer administrator of a fifty-person chat group pins another member’s allegation for all to see; a mailing-list operator distributes a subscriber’s submission to thousands of recipients. In each case the forwarder is a user of an interactive computer service handling information provided by another, and each can invoke the same provision as the service hosting the original. The administrator who then removes a disruptive member from the group is likewise covered, this time by (c)(2), which names users alongside providers. The statute protects the entire ordinary repertoire of online participation: sharing, forwarding, hosting comments, curating a small community.

The reasons this coverage escapes public notice run deeper than headline habits. User-side invocations of the statute rarely generate news because they typically succeed quietly, ending lawsuits at the pleading stage without the drama of a major platform dispute. Press and commentary gravitate toward conflicts involving large companies, where the stakes are visible and the parties are famous. Over time, the accumulation of platform-centered coverage trains even careful readers to hear “section 230” as “the platform law,” and the user half of the statutory phrase fades from the mental model. Recovering it requires returning to the words Congress wrote, which name users in both paragraphs, and which condition nothing on corporate size, revenue, or number of accounts.

Reading a Dispute Through the Two Provisions

The most useful habit a reader can develop with this statute is a sorting habit. Nearly every public argument about section 230 runs the two operative provisions together, and the arguments only become legible once they are separated. The discipline has two steps. Step one: ask whether the complaint targets content the defendant left up, content the defendant took down, or a moderation tool the defendant supplied to others. Step two: match the answer to the subsection that governs it. Content left standing implicates subsection (c)(1). Content removed, restricted, or filtered implicates subsection (c)(2). A moderation tool supplied for others to use also implicates subsection (c)(2). When the answer is unclear, the analysis is unclear, and that is usually a sign that the person making the argument has not identified the provision at issue.

Consider first the most familiar pattern: a defamation claim over a user comment that the service left visible. A person writes a false and damaging statement about someone else in a comment thread, a review, or a forum post, and the subject of the statement sues the service that hosted the comment. This is the core case for subsection (c)(1). The twenty-six words do the entire work: 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 lawsuit asks the court to treat the service as the publisher of the comment, which is precisely what the sentence forbids. The identity of the actual author is what matters, not the service’s conduct after the fact. A frequent misunderstanding holds that the service must have remained ignorant of the comment to keep the protection, or that notifying the service of the defamation strips the shield away. The text contains no such condition. Subsection (c)(1) asks one question about authorship: did the information come from another information content provider? If it did, the service is not the publisher or speaker of it, and the claim fails at the threshold. The Zeran decision examined earlier made this reading the foundation of the doctrine, holding that the provision bars distributor liability as well as publisher liability, so that notice of unlawful content does not create a duty to remove it.

Consider second a claim over removed posts. A user sues a service because the service deleted the user’s posts, suspended the user’s account, or demoted the user’s material in a feed, alleging breach of contract, unfair business practices, or some other theory. This dispute belongs to subsection (c)(2)(A), a different provision with different requirements. That subparagraph provides that no provider or user of an interactive computer service shall be held liable on account of 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. Three features of this language deserve emphasis. First, good faith appears here and nowhere in subsection (c)(1): the take-down protection is conditional in a way the leave-up protection is not. Second, the standard is subjective, turning on what the provider or user considers objectionable, and the statute expressly adds that the material need not be constitutionally unprotected. Third, the list of categories is illustrative rather than exhaustive, ending in the open phrase “otherwise objectionable,” which is the phrase that gives the subparagraph its breadth. A reader evaluating a take-down dispute should therefore ask whether the service acted in good faith to restrict material it considered objectionable. A reader who instead asks whether the service was the publisher of the material is applying the wrong subsection to the wrong dispute.

Consider third a claim over the platform’s own editorial writing. A service publishes an article written by its own staff, an editorial explaining its policies, or a promotional description of its own product, and someone sues over the content of that material. Neither subsection protects the service here, and the reason lies in the definition that does the quiet work throughout the statute. Subsection (c)(1) shields a provider only with respect to information provided by another information content provider, and section 230(f)(3) defines an information content provider as any person or entity responsible, in whole or in part, for the creation or development of information. A service that writes its own material is the information content provider of that material, so the leave-up protection has nothing to attach to, and subsection (c)(2) is equally unavailable because the dispute concerns the service’s own speech rather than action taken to restrict someone else’s. The statute protects intermediaries for others’ words; it does not protect speakers for their own.

Consider fourth a spam-filter or antivirus false-positive claim. A company makes filtering software that classifies another company’s program as adware or blocks a category of messages as spam, and the blocked company sues for tortious interference, trade libel, or unfair competition. This pattern implicates subsection (c)(2), and often specifically subparagraph (B), which covers any action taken to enable or make available to information content providers or others the technical means to restrict access to the described material. The distinction between the two subparagraphs of (c)(2) carries a consequence that careful readers should register: subparagraph (A) contains the good-faith requirement, while subparagraph (B) does not, a difference courts have treated as deliberate. In litigation over filtering software, courts have held that the statute does not require the blocked material to be objectionable in fact, only that the provider or user consider it objectionable, and that supplying users with the technical means to restrict such material falls within subparagraph (B). The pattern shows the statute reaching beyond hosting platforms to the makers of the filtering tools themselves, which is exactly what the fourth paragraph of the policy statement, on removing disincentives for blocking and filtering technologies, anticipated.

Consider fifth the pattern in which the service shapes the content itself. A site requires users to answer questions with prepopulated choices as a condition of using the service, and the resulting profiles contain the allegedly unlawful material. Whether subsection (c)(1) applies turns on the information-content-provider definition: if the service was responsible, in whole or in part, for the creation or development of the information at issue, the service is not dealing with information provided by another information content provider, and the leave-up protection falls away. Courts have drawn this line at material contribution, the point where a service stops being a passive host and becomes a co-developer of the content. The Roommates.com decision examined earlier is the leading illustration. The important point for the sorting habit is that this is still a (c)(1) question, answered through the definition of information content provider, and it has nothing to do with good faith or with subsection (c)(2).

Consider sixth a defamation claim against an individual user rather than a company. A person reposts another user’s false statement to a wider audience, or forwards a defamatory message to a mailing list, and the subject of the statement sues the reposter. This pattern belongs to subsection (c)(1) just as fully as the platform case does, because the provision shields no provider or user of an interactive computer service. The reposter is a user, the statement is information provided by another information content provider, and the lawsuit asks that the reposter be treated as the publisher or speaker of it. The text draws no distinction between institutional and individual defendants at this step. The protection is widely missed in public discussion, which tends to frame the statute as a corporate immunity, but the words cover the ordinary internet user who shares what someone else wrote. The same authorship question decides the case: if the reposter added original defamatory content of their own, the reposter is the information content provider of that added material and the shield does not cover it, but the mere act of passing along another’s words is within the provision’s terms.

Two common errors show why the sorting habit matters. One error reads a good-faith requirement into subsection (c)(1), as though a service that behaves badly loses the leave-up protection. The text contains no such requirement, and courts have not imposed one. The other error treats subsection (c)(2) as the provision that allows services to leave content up, when it is the provision that protects them for taking content down. These are mirror-image mistakes, and both dissolve once the reader asks the step-one question: left up or taken down. The statute’s two provisions answer two different complaints, and a reader who cannot say which subsection is at issue in a given dispute is not equipped to evaluate any claim about the law.

What the Case Law Settled, and What It Did Not

The Zeran construction examined earlier fixed the outer boundary of subsection (c)(1): a service is protected even after being notified of unlawful third-party content, because notice cannot convert the service into the publisher or speaker of someone else’s words. The Supreme Court denied certiorari in 1998, and courts across the circuits followed Zeran’s construction for the next two decades, dismissing defamation, negligence, and related claims against providers that had hosted but not authored the content at issue.

The Zeran construction also shaped how the immunity is litigated, not only what it covers. Because subsection (c)(1) turns on a threshold question of authorship rather than on the defendant’s conduct or state of mind, courts have repeatedly resolved the immunity question at the outset of litigation, on motions to dismiss or for summary judgment, rather than treating it as a defense to be proved at trial. The reasoning is practical: if the complaint’s own allegations show that the defendant is being asked to answer for information provided by another information content provider, no discovery into the defendant’s knowledge, motives, or moderation practices can change the statutory answer. The Zango litigation illustrates the point, with the district court granting summary judgment and declining to order discovery into the defendant’s alleged bad faith. This procedural posture magnified the doctrine’s effect, since early resolution spares providers the cost of litigating to judgment over content they did not create, while concentrating the remaining disputes on the authorship question that the statute makes dispositive.

If Zeran fixed the outer boundary of subsection (c)(1), the Roommates.com decision examined earlier fixed its inner boundary, the point where a service stops being a protected intermediary and becomes a co-author. The Ninth Circuit’s material-contribution analysis remains the principal textual limit on subsection (c)(1): structuring, soliciting, or channeling content in a way that develops the unlawful information defeats the shield, while merely providing the forum does not.

The boundary the Ninth Circuit drew becomes sharper when set against the decision the Roommates.com majority distinguished. In Chicago Lawyers’ Committee for Civil Rights Under Law, Inc. v. craigslist, Inc., decided by the Seventh Circuit in March 2008, the same fair housing organization sued the classified-advertising site over allegedly discriminatory housing postings submitted by users. The Seventh Circuit held craigslist immune, reasoning that the site had done nothing to induce the specific content of the postings: it provided a forum and a set of optional blank fields, but it did not require users to disclose protected characteristics, supply prepopulated answers, or channel listings according to the answers given. The contrast with Roommates.com is instructive rather than contradictory. A passive forum that hosts whatever users submit, even where the submissions are unlawful, remains within subsection (c)(1); a service that designs the submission process to elicit and structure the unlawful content crosses into development. Together the two 2008 decisions map the territory between hosting and authoring, and they show that the material-contribution inquiry is fact-specific: it asks what the service actually did to shape the information at issue, not how large the service is or how much content it carries.

Courts have also given content to the good-faith requirement of subsection (c)(2), though the treatment has been generous to defendants and, in one respect, structurally asymmetric. The leading illustration is Zango, Inc. v. Kaspersky Lab, Inc., 568 F.3d 1169 (9th Cir. 2009), decided June 25, 2009, in which a distributor of ad-supported software sued a maker of anti-malware filtering software that had classified its programs as adware and blocked them. The Ninth Circuit held the filtering-software maker immune under subsection (c)(2)(B), the subparagraph covering the provision of technical means to restrict access. Two holdings from that line of cases matter for reading the statute. First, the good-faith condition appears in subparagraph (A), which covers actions voluntarily taken to restrict access to material, but not in subparagraph (B), which covers supplying the technical means of restriction to others; courts have treated that difference as deliberate, so a filtering-tool maker does not face the good-faith inquiry at all. Second, where the good-faith requirement does apply, courts have read it through the statute’s subjective language: the material need only be what the provider or user considers objectionable, not what is objectionable in fact, and the phrase “whether or not such material is constitutionally protected” removes any requirement that the blocked material be unlawful. Conclusory allegations of bad faith, without more, have been held insufficient to defeat the shield at summary judgment. A concurring judge in Zango worried aloud that this construction could protect anticompetitive blocking by filtering-software vendors, a concern the majority answered by pointing to the statute’s text and purpose. The net effect is that good faith under subsection (c)(2)(A) remains a real textual requirement, and a service acting in demonstrable bad faith to restrict material is not what Congress described, but courts have set the evidentiary bar so that the requirement functions as a backstop against abuse rather than as a general license to relitigate moderation decisions.

What the case law has not settled is nearly as instructive as what it has. The Supreme Court has never construed the scope of section 230 in a majority opinion. The closest it came was in May 2023, when it decided two terrorism-related cases together: Twitter, Inc. v. Taamneh, 143 S. Ct. 1206 (2023), held unanimously that social media platforms were not liable for aiding and abetting terrorism under the Anti-Terrorism Act, and Gonzalez v. Google LLC, 598 U.S. 617 (2023), which had presented the question whether section 230 protects algorithmic recommendation of user content, was resolved in a brief per curiam opinion that expressly declined to address the application of section 230, vacating and remanding in light of Taamneh. The Court was offered a vehicle to reinterpret the provision and chose not to take it, leaving the lower-court doctrine intact. Open questions therefore persist at the margins: how much algorithmic structuring of third-party content counts as development under the Roommates.com standard, how the subjective good-faith inquiry of subsection (c)(2)(A) applies to large-scale automated moderation systems, and how far the material-contribution limit reaches when a service’s design choices shape what users see and say. These are live disputes in the lower courts, and they turn on the statutory text rather than on slogans about the provision’s purpose.

Separate from those doctrinal questions are the debates about what should be done with the statute, and those belong elsewhere in this series. Proposals to repeal the provision, to condition the shield on practices described as reasonable moderation or viewpoint neutrality, to carve out additional categories of claims, or to rewrite the definition of information content provider are all contested, with supporters and critics advancing incompatible accounts of the consequences. This article does not adjudicate those proposals. The companion myths article takes them up directly. The point for this text-centered article is narrower: whatever Congress or the courts do next will be done to this text, and the text’s two provisions, its definitions, and its exclusions are the terrain on which every reform argument must be fought.

Broad, Not Absolute

The limits surveyed across these sections combine into a single picture: the protection section 230 offers is broad within its domain and absent outside it. Four boundaries define that domain. The good-faith requirement in (c)(2)(A) conditions the takedown shield on honesty of purpose. The definition of information content provider withholds the (c)(1) shield from anyone who materially contributes to the creation or development of the content. The exclusions in subsection (e) remove federal criminal law, intellectual property law, communications privacy law, state law inconsistent with the section, and specified sex trafficking claims from the statute’s reach. And nothing in the section shields a service’s or user’s own original content, which was never third-party information to begin with. Each boundary is stated in the text. Together they answer the assumption that the protection is unconditional in every respect.

The breadth of the protection within those boundaries was established early. In 1997, the Fourth Circuit decided Zeran v. America Online, 129 F.3d 327, the foundational construction examined earlier in this article. Under that reading, notice of the unlawful nature of third-party content does not defeat the protection: a service is shielded whether or not someone has told it about the offending material, because the claim still seeks to treat the service as the publisher of someone else’s content. Later courts adopted the reasoning widely, and it remains the assumed starting point for (c)(1) disputes.

That breadth is sometimes mistaken for boundlessness. The mistake overlooks how the statute’s own terms cabin it. Consider a service that does more than host: it edits user submissions to add false statements of fact, or it solicits unlawful material and shapes its presentation. At that point the service has become responsible in part for the creation or development of the information, fitting the statutory definition of an information content provider for that content, and the (c)(1) protection for third-party material no longer covers it. Consider a service sued for its own statements, its own product design representations, or its own contractual promises. Those claims do not treat the service as the publisher of another’s content, so (c)(1) does not address them; they rise or fall on their own legal merits. And consider the exclusions: a copyright owner proceeding under the DMCA, a trafficking victim proceeding under 18 U.S.C. 1595 for conduct violating section 1591, a federal prosecutor bringing criminal charges. Each proceeds without encountering the statute at all.

The moderation-encouraging rationale and the accountability concerns both deserve their due in describing this design. The statute’s defenders argue that the two-paragraph structure reflects a deliberate bargain: (c)(1) removes the penalty for hosting third-party speech, so that services need not choose between carrying everything and carrying nothing, while (c)(2) removes the penalty for moderating, so that services can remove objectionable material without thereby assuming liability for what remains. On this account, the Good Samaritan title is the point: Congress wanted to encourage voluntary cleanup of the network, and the subjective, good-faith standard in (c)(2)(A) gives moderators room to act without litigating each decision. Critics of the breadth argue that the combination shields too much, leaving victims of harmful third-party content without recourse against the services that profit from its distribution, and that the good-faith and material-contribution limits have proven difficult to enforce in practice. Both positions are arguments about what the law should be. The text itself is neutral ground: it states the protections, states their conditions, and states the exclusions, and any evaluation of reform proposals must start from what those words actually cover.

That is where the article’s central claim returns. Two provisions, not one: the argument that platforms hide behind the statute to leave content up and the argument that they hide behind it to take content down concern different subsections with different requirements. Leaving-up disputes are (c)(1) disputes. They ask whether the defendant is being treated as the publisher or speaker of information provided by another, whether the defendant materially contributed to the content, and whether the claim falls within an exclusion. Taking-down disputes are (c)(2) disputes. They ask whether the restriction was voluntary and in good faith, whether the actor considered the material objectionable, or whether the actor supplied the technical means of restriction. Importing (c)(2)’s good-faith inquiry into a leave-up dispute, or (c)(1)’s publisher analysis into a takedown dispute, misstates the law in both directions. Contested claims about reform, repeal, and the statute’s real-world effects belong to the myths article, where they are addressed directly. The work of this article is narrower and, for that reason, more durable: to fix the words in place, so that every argument about the most debated provision in technology law can at least begin from the provision itself.

Both the moderation-encouraging rationale and the accountability critique read the same text and draw opposite lessons from it, which is why precision about the words matters more than allegiance to either camp. Defenders point to the Good Samaritan structure and argue that without (c)(1), services would either over-remove to avoid liability or abandon moderation entirely, and without (c)(2), each act of moderation would expand exposure. Critics point to the same paragraphs and argue that the combination of a motive-free hosting shield and a subjective takedown shield leaves too little room for redress, and that the statutory limits have not constrained the largest beneficiaries in practice. The text does not resolve that dispute. It defines the field on which the dispute is fought: two protections, stated conditions, stated exclusions, and a set of words that have anchored three decades of case law. Any argument about what the law should become must begin with an accurate account of what the law says, and that account starts, and ends, with the distinction between leaving content up and taking it down.

Why the Text Matters More Than the Argument

No provision in American technology law has generated more confident public argument from less textual study. Section 230 is invoked as though it were a single rule with a single purpose, a kind of all-purpose permission slip for the internet industry, and the debates that follow from that premise are fought almost entirely at the level of slogans. The text, read with care, supports none of that looseness. It contains two provisions that do completely different work, definitions that determine when each provision applies, a conditional good-faith requirement that attaches to one provision and not the other, and a set of express exclusions that mark where the protection stops. The series thesis for this cluster holds that the operative text is the location of the real dispute, and in this provision, where public argument has drifted furthest from the words, the thesis is at its strongest. The arguments about the statute cannot be evaluated without the statute.

The drift begins with the failure to distinguish the two subsections, and that failure produces the central confusion of the public debate. The argument that services hide behind the statute to leave harmful content up is an argument about subsection (c)(1): it concerns whether a provider can be treated as the publisher or speaker of information provided by another information content provider, and the defenses available to the service run through authorship and the definition of information content provider. The argument that services hide behind the statute to take content down, censoring viewpoints or silencing critics, is an argument about subsection (c)(2): it concerns whether action voluntarily taken in good faith to restrict material the provider considers objectionable is shielded from liability, and the defenses available to the service run through voluntariness, good faith, and the subjective standard of objectionability. These are not two angles on one rule. They are two rules with different triggers, different textual conditions, and different bodies of case law, addressing two different complaints. The namable claim of this article follows directly: a reader who cannot say which subsection is at issue in a given dispute is not equipped to evaluate any claim about the law. The test is practical rather than academic. Presented with a lawsuit over a service’s handling of third-party content, the equipped reader asks first whether the complaint targets what was left standing or what was taken away, and then reads the matching subsection. The unequipped reader reaches for a general impression of what the statute supposedly does, and the general impression is where the errors live.

The text also disciplines the two great competing accounts of what the statute is for, and it does so without choosing between them. On one side, supporters of the shield argue that it is what makes moderation possible at all: without subsection (c)(1), the reasoning goes, services would face the moderator’s dilemma that Stratton Oakmont created, and the rational response would be to stop moderating, since every act of editorial judgment would invite publisher liability for everything missed. Without subsection (c)(2), the argument continues, even services willing to moderate would hesitate, since removing one user’s material invites lawsuits from the removed user. On this account, the statute is the legal foundation of content moderation rather than its enemy, and narrowing the shield would produce either abandoned moderation or defensive over-removal, harming the very speech the critics want protected. On the other side, critics of the shield argue that the breadth of subsection (c)(1), particularly as the Zeran line of cases developed it, allows services to profit from the distribution of harmful third-party content while disclaiming responsibility for it, and that the subjective standard of subsection (c)(2) gives services cover to remove material on pretextual grounds with little prospect of judicial scrutiny. On this account, the statute externalizes the costs of harmful content onto its victims while privatizing the benefits of scale, and reform is a matter of accountability. Both accounts are advanced in good faith by serious participants, and this article does not resolve between them. The point is that each account maps onto different words in the statute, and the words constrain what each account can claim. The moderation-encouraging account must reckon with the fact that subsection (c)(1) contains no good-faith condition and that notice does not defeat it. The accountability account must reckon with the fact that subsection (c)(2) does contain a good-faith condition, that the information-content-provider definition excludes co-developed content from the leave-up shield, and that subsection (e) removes entire categories of law from the statute’s reach. Neither side gets to rewrite the text, and the text gives neither side everything it wants.

A final discipline concerns the persons the statute protects, because the protection is wider than the public debate usually assumes. Subsection (c)(1) shields no provider or user of an interactive computer service, and the inclusion of the user is not decorative: as the earlier section showed, the ordinary individual who forwards, reposts, or shares another person’s content stands in the same protected position as the largest platform. Any proposal to narrow the shield therefore implicates the legal position of every individual who shares, forwards, or curates the speech of others. The text does not limit itself to corporate defendants, and arguments that assume otherwise are arguments about a different statute.

The protection is not unconditional in every respect, and the conditions are textual: good faith for the take-down shield, the information-content-provider definition excluding co-developed content from the leave-up shield, and the exclusions of subsection (e). These limits do not make the protection narrow; the twenty-six words of subsection (c)(1) remain one of the broadest liability shields in American law, and the Zeran construction remains the foundation of the case law. But broad is not the same as absolute, and the difference is written into the statute for anyone who reads it.

Return, then, to the beginning. The statute sits in Title V of the Telecommunications Act of 1996 as the liability-shield survivor of a speech-restricting title. It was written to solve the moderator’s dilemma by protecting services both for what their users post and for what the services take down. It defines its own key terms, conditions one of its two protections on good faith, and carves out the categories of law it does not touch. The case law has given the leave-up provision a broad construction anchored in Zeran, drawn the material-contribution limit in Roommates.com, read the take-down provision’s good faith generously but not emptily, and left the Supreme Court’s own view of the statute unexpressed. That is the text, and the text is the dispute. Every claim about what section 230 does, whether it does too much or too little, whether it should be kept, narrowed, or repealed, is a claim about these subsections, these definitions, and these exclusions. The reader who can place a dispute in the right subsection, name the definition that decides it, and identify the exclusion that might remove it can evaluate any such claim. The reader who cannot is working from the argument rather than the law, and in this provision more than any other, the argument has drifted furthest from the words.

A companion study notebook for working through the provision subsection by subsection, with space to record the statutory text, the case holdings, and the exclusions behind each answer, is available through the series study tools (legislation study notebook). Use it to build the habit this article teaches: identify the subsection, check the definition, check the exclusions, and only then decide what the law does.

The Provision Map

Subsection	What it protects	Who it protects	Condition attached	Category of claim it does not reach
(c)(1)	Hosting third-party content: bars treating the actor as the publisher or speaker of information provided by another	Providers and users of an interactive computer service	The content must have been provided by another information content provider; no good-faith, voluntariness, or notice condition	Any claim that depends on treating the actor as the publisher or speaker of third-party content, regardless of the cause of action named
(c)(2)(A)	Restricting objectionable material: bars liability on account of voluntary good-faith action to restrict access to or availability of material	Providers and users of an interactive computer service	The action must be voluntary and in good faith, directed at material the actor considers obscene, lewd, lascivious, filthy, excessively violent, harassing, or otherwise objectionable	Takedown claims within the (e) exclusions; removals shown to be pretextual rather than in good faith
(c)(2)(B)	Supplying filtering tools: bars liability for enabling or making available the technical means to restrict access	Providers and users who supply the tools, for the benefit of information content providers or others	The tools must enable restriction of third-party material; the text imposes no good-faith qualifier here	Claims outside the tool-supplying act; the (e) exclusions where applicable
(e)(1)	Exclusion: federal criminal law preserved	Federal prosecutors and criminal enforcement	None; the section neither impairs nor expands criminal enforcement	Criminal prosecutions under chapter 71 or 110 of title 18, sections 223 or 231 of the Communications Act, or any other federal criminal statute
(e)(2)	Exclusion: intellectual property law preserved	Copyright and trademark owners and defendants under those laws	The claim must sound in intellectual property law, with its own immunities and defenses	Copyright and trademark claims, including DMCA safe-harbor disputes, which proceed without regard to section 230
(e)(3)	Exclusion and preemption: consistent state laws preserved, inconsistent ones barred	States enforcing consistent laws; defendants facing inconsistent ones	The state claim must not treat the actor as the publisher or speaker of third-party content	State-law claims inconsistent with the section, including state defamation and negligence theories targeting hosting or failure to remove
(e)(4)	Exclusion: communications privacy law preserved	Parties to wiretapping and stored-communications claims under ECPA or similar state laws	The claim must arise under the Electronic Communications Privacy Act of 1986, its amendments, or a similar state law	Interception and disclosure claims, which live in their own statutory regime with their own elements and defenses
(e)(5)	Exclusion: specified sex trafficking claims preserved	Trafficking victims under 18 U.S.C. 1595 and state criminal prosecutors	The conduct must violate 18 U.S.C. 1591, or the charge must rest on section 1591 or 2421A as the statute specifies	Civil trafficking claims and qualifying state prosecutions, which proceed without regard to section 230 while good-faith removal stays protected under (c)(2)(A)

Frequently Asked Questions

Q: What does section 230 actually say?

Section 230 is a provision of the Communications Act, at 47 U.S.C. 230, enacted as section 509 of the Telecommunications Act of 1996. Its core sentence, in subsection (c)(1), provides that 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. Subsection (c)(2) adds a second protection: providers and users cannot be held liable for good-faith voluntary action to restrict access to material they consider objectionable, or for supplying others with the technical means to do so. The section defines its two key terms in subsection (f) and lists five exclusions in subsection (e), covering federal criminal law, intellectual property, state law, communications privacy, and specified sex trafficking claims. It is two protections, not one, with stated conditions and stated limits.

Q: What are the 26 words of section 230?

The twenty-six words are the text of subsection (c)(1): “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 bars courts from applying the publisher or speaker characterization to providers and users for content that originated with someone else. It contains no requirement of knowledge, no notice procedure, no duty to investigate, and no good-faith qualifier. Its most important limitation is the final phrase: the protection applies only when the information was provided by another information content provider, so a defendant that created or developed the content itself cannot invoke the sentence for that content.

Q: What does section 230(c)(2) protect?

Subsection (c)(2) protects the take-down side of the statute’s design. Subparagraph (A) shields providers and users from liability for action voluntarily taken in good faith to restrict access to material they consider obscene, lewd, lascivious, filthy, excessively violent, harassing, or otherwise objectionable, whether or not the material is constitutionally protected. Subparagraph (B) shields the act of enabling or making available to others the technical means to restrict access to such material, such as parental control software, spam filters, and blocklists. Unlike (c)(1), subparagraph (A) conditions its protection on good faith and on the actor’s own assessment of the material, while subparagraph (B) contains no good-faith qualifier. The provision protects moderating, not hosting.

Q: Who counts as an interactive computer service under section 230?

Subsection (f)(2) defines “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, including services that provide access to the internet and systems operated by libraries or educational institutions. Courts have read the definition generously: internet access providers, platforms and forums of every size, and nearly every interactive website clear the threshold. The specific mention of libraries and educational institutions shows Congress intended a functional category rather than one tied to a business model. Because the definition is so broad, litigation rarely turns on it; the protection turns instead on what the defendant did with the content, governed by the second definition.

Q: What is an information content provider under section 230?

Subsection (f)(3) defines “information content provider” as any person or entity responsible, in whole or in part, for the creation or development of information provided through the internet or another interactive computer service. The definition decides when the (c)(1) protection is lost: the twenty-six words cover only information provided by another information content provider, so a defendant that is itself the information content provider of the disputed material cannot invoke them. Partial responsibility suffices, and “development” reaches beyond literal authorship to shaping, soliciting, or materially contributing to content someone else supplied. Courts apply a material-contribution test, asking whether the defendant contributed to the alleged unlawfulness of the content, as the Ninth Circuit did in the Roommates.com decision.

Q: What does section 230 not cover?

Section 230 does not cover five categories of law, listed in subsection (e): federal criminal prosecutions, intellectual property claims such as copyright and trademark, state-law claims that are consistent with the section or rather preempted only when inconsistent, communications privacy claims under the Electronic Communications Privacy Act, and specified sex trafficking civil claims and state prosecutions added by the 2018 amendment. It also does not cover a defendant’s own authored content, since the (c)(1) protection reaches only information provided by another information content provider. It does not impose a good-faith requirement on hosting decisions, because (c)(1) contains none, and it does not shield pretextual takedowns, because (c)(2)(A) requires good faith. The protection is broad within its domain and absent outside it.

Q: Does section 230 protect individual users?

Yes. Both operative provisions name “provider or user” of an interactive computer service, and the user half is not decorative. An individual who forwards, reposts, shares, or links to another person’s content is covered by the same twenty-six words that protect the largest platforms. Under traditional defamation principles every repetition of a defamatory statement counts as a publication, so without the user coverage ordinary sharing would be a liability trap. The Ninth Circuit confirmed the point in Batzel v. Smith in 2003, protecting an individual who selected another person’s email for distribution to a mailing list. The limits apply equally: the protection does not cover the user’s own authored words or content the user materially helped create, and (c)(2) likewise protects users who restrict objectionable material in good faith.

Q: What did Zeran v. America Online decide about section 230?

Zeran v. America Online, Inc., 129 F.3d 327, decided by the Fourth Circuit on November 12, 1997, is the foundational construction of section 230. Kenneth Zeran sued America Online for negligence after hoax postings about the Oklahoma City bombing listed his telephone number and he received threatening calls even after notifying the company. He argued that notice made AOL a distributor with a duty to remove the postings, since the statute spoke only of publishers and speakers. The court rejected the distinction, holding that distributor liability is a subset of publisher liability and therefore barred as well, that the statute blocks suits targeting a provider’s traditional editorial functions, and that a notice-based regime would defeat the statute’s purposes. The Supreme Court declined review in 1998, and the Zeran reading anchored the case law that followed.

Q: How does section 230 treat content left up differently from content taken down?

Content left up is governed by subsection (c)(1), which asks a single authorship question: is the defendant being treated as the publisher or speaker of information provided by another information content provider? If so, the claim fails, with no inquiry into motive, knowledge, or good faith. Content taken down is governed by subsection (c)(2), which asks different questions: was the restriction voluntary and in good faith, and did the actor consider the material objectionable, or did the actor supply the technical means of restriction? The two provisions have different triggers, different conditions, and different bodies of case law. Reading a good-faith requirement into a leave-up dispute, or a publisher analysis into a takedown dispute, applies the wrong subsection to the wrong complaint, and most public confusion about the statute comes from exactly that error.

Q: Does a service lose section 230 protection once it learns about unlawful content?

No. Subsection (c)(1) contains no requirement of knowledge and no notice procedure, and the Zeran decision held that notice does not defeat the protection. A service that has been told about allegedly unlawful third-party content receives the same shield as one that has never heard of it, so long as the service did not create or develop the content itself. The Fourth Circuit reasoned that distributor liability is a subset of publisher liability, so the statute’s bar on the publisher characterization covers notice-based claims too, and that a notice regime would force services either to remove content automatically upon complaint or to abandon moderation entirely. The protection turns on authorship, not awareness: what matters is who provided the content, not whether the defendant knew about it.

Q: What did Roommates.com decide about when a service becomes a content provider?

Fair Housing Council of San Fernando Valley v. Roommates.com, LLC, 521 F.3d 1157, decided en banc by the Ninth Circuit in 2008, is the leading decision on the boundary between hosting and developing content. The roommate-matching site required subscribers to answer mandatory questions from pre-populated menus soliciting sex, sexual orientation, and family status, then assembled the answers into profiles. The court held the site was responsible, in whole or in part, for the development of that information and therefore an information content provider as to it, so the (c)(1) shield did not apply to the questionnaire content. The open-ended comments section, a blank field with no steering, stayed protected. The decision established the material-contribution test: a service loses protection when it contributes materially to the alleged unlawfulness of the content.

Q: What does treated as the publisher or speaker mean in section 230?

The phrase “shall be treated as” in subsection (c)(1) directs courts not to apply a particular legal characterization to the defendant. Congress did not write “shall not be liable” as a defense to be proved; it forbade the characterization on which publisher-based claims depend. A publisher is the party legally responsible for originating and disseminating content, and a speaker is the party who utters it. Any cause of action that requires the defendant to occupy one of those roles for third-party content, whether labeled defamation, negligence, or something else, fails at the threshold because the statute bars the court from making the required characterization. This formulation is also why courts resolve (c)(1) disputes early, often on a motion to dismiss: if the complaint itself shows the defendant is being asked to answer for another’s information, no discovery into knowledge or motives can change the statutory answer.

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

Good faith under subparagraph (c)(2)(A) turns on honesty of purpose, not on whether a court agrees with the underlying judgment about the material. A removal carried out sincerely to keep material the actor finds objectionable off the service satisfies the requirement; a removal carried out as a pretext, such as sabotaging a competitor while invoking a content policy as cover, would not. The standard pairs with the subjective verb “considers”: the court asks whether the provider or user considered the material objectionable, not whether it was objectionable in fact, and the statute adds that the material need not be constitutionally unprotected. Courts have set the evidentiary bar so that conclusory allegations of bad faith, without more, do not defeat the shield, making good faith a backstop against abuse rather than a license to relitigate moderation decisions.

Q: Does section 230 protect filtering software and parental control tools?

Yes, through subparagraph (c)(2)(B), which shields any action taken to enable or make available to information content providers or others the technical means to restrict access to third-party material. Parental control software, spam filters, blocklists, and user-operated mute and block functions all fall within this clause. The text of (B) contains no good-faith qualifier and no list of objectionable categories, and courts have treated that difference from subparagraph (A) as deliberate, so a filtering-tool maker does not face the good-faith inquiry. Section 230(d) reinforces the design by requiring providers to notify customers that parental control protections are commercially available. In Zango, Inc. v. Kaspersky Lab, Inc., 568 F.3d 1169 (9th Cir. 2009), the Ninth Circuit held a maker of anti-malware filtering software immune under this subparagraph.

Q: What does otherwise objectionable mean in section 230(c)(2)?

The phrase “or otherwise objectionable” is the catch-all at the end of the (c)(2)(A) list of material a provider or user may restrict in good faith, following obscene, lewd, lascivious, filthy, excessively violent, and harassing. Courts have generally read it broadly, as covering material beyond the enumerated examples, and it is paired with the subjective verb “considers,” so the question is what the actor found objectionable rather than what a court would classify the material as. Commentators have debated whether the phrase should be narrowed by the canon under which general words take color from the specific words around them, and the Ninth Circuit has expressed doubt about applying that canon to this paragraph. What the text establishes beyond dispute is an open-ended category rather than a closed list, giving moderators room to act without litigating each decision’s characterization.

Q: Why is section 230(c) called the Good Samaritan provision?

Subsection (c) carries the heading “Protection for ‘Good Samaritan’ blocking and screening of offensive material,” and the name reflects the problem Congress set out to solve. Before the statute, the moderator’s dilemma punished the service that screened its forums with publisher liability while the service that did nothing kept the lighter distributor standard. The Good Samaritan title names the conduct Congress wanted to encourage: voluntary private action to screen offensive material, undertaken without fear that the act of screening would itself create liability. The two paragraphs of the subsection give the title its legal form: (c)(1) protects hosting third-party content, and (c)(2) protects the intervention itself, so a service that takes down objectionable material does not forfeit protection for what remains and gains an independent shield for the takedown.

Q: Can section 230 be used as a defense in a federal criminal case?

No. Subsection (e)(1) provides that nothing in the section shall be construed to impair the enforcement of specified federal criminal provisions or any other federal criminal statute. A service whose operators face prosecution under federal obscenity law, child sexual exploitation law, or any other federal criminal statute cannot invoke section 230 as a defense. The exclusion is one-directional: it preserves existing criminal enforcement without creating new offenses or expanding prosecutorial authority. Most courts have also held that the exclusion does not preserve civil lawsuits merely because the underlying conduct also violates a criminal law, since reading it that way would let any civil plaintiff evade the statute by framing the claim in criminal terms. The carve-out is for criminal enforcement itself, not for private claims dressed in criminal-law language.

Q: What does the intellectual property exclusion in section 230 leave to other laws?

Subsection (e)(2) provides that nothing in the section shall be construed to limit or expand any law pertaining to intellectual property, which courts have read to cover copyright and trademark claims. Copyright disputes over user-uploaded material therefore proceed under the Digital Millennium Copyright Act at 17 U.S.C. 512, with its own safe harbors conditioned on requirements section 230 never imposes, such as designating an agent for infringement notices and acting expeditiously on proper takedown requests. Trademark claims over counterfeit listings proceed under trademark principles, including any contributory liability doctrines that law provides. The exclusion is neutral: it neither creates intellectual property liability nor reduces it. Section 230 simply steps aside, leaving each claim to stand or fall under the intellectual property laws as they exist.

Q: Does section 230 preempt state laws that conflict with it?

Yes, through subsection (e)(3), which bars any cause of action and any liability under state or local law inconsistent with the section, while preserving state laws that are consistent with it. Courts apply the provision claim by claim from the pleadings, asking whether the state claim as pleaded treats the defendant as the publisher or speaker of third-party content or targets the defendant’s own conduct. A state defamation or negligence claim seeking to hold a service liable for hosting or failing to remove a user’s post is inconsistent and barred. A state claim grounded in the service’s own promises or misrepresentations, such as promissory estoppel based on a specific assurance, has sometimes survived because it does not ask the court to treat the service as anyone’s publisher. The paragraph thus preserves state regulatory authority while preventing states from achieving through local law what the federal statute forbids.

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

Yes, under the Zeran construction that courts have followed since 1997. At common law, distributors such as bookstores faced liability only with knowledge of unlawful content, a lighter standard than the strict liability of publishers. The Fourth Circuit held that distributor liability is a subset of publisher liability rather than a separate category the statute left intact, because publication in the legal sense includes communicating content to third parties. Since section 230(c)(1) forbids treating providers and users as publishers or speakers of another’s information, the distributor theory is foreclosed by the same words. The practical consequence is the notice rule: being told about unlawful third-party content does not create a duty to remove it, because the underlying claim still seeks to punish the defendant for its role in publishing someone else’s words.