The final substantive excerpt from the First Amendment section of my Social Media as Common Carriers? article (see also this thread); recall that the key First Amendment arguments are in this post, which relies on the PruneYard, Turner, and Rumsfeld precedents, and in this one, which explains why Miami Herald, Hurley, and the various other “common theme” precedents don’t apply.
[* * *]
For all these reasons, I think Congress could categorically treat platforms as common carriers, at least as to their hosting function. But Congress could also constitutionally give platforms two options: (1) Be common carriers like phone companies, immune from liability but also required to host all viewpoints, or (2) be distributors like bookstores, free to pick and choose what to host but subject to liability (at least on a notice-and-takedown basis).
Indeed, this would just return the law to something close to the pre-§ 230 common-law rules, as modified by the First Amendment protections that have developed starting with New York Times v. Sullivan. Historically, American law has divided operators of communications systems into three categories—publishers, distributors, and conduits—and has set up different standards of liability for each.
Newspapers, magazines, and broadcasters, which themselves print or broadcast material submitted by others (or by their own employees). 
Free to choose what to include.
Fully liable for material they include, as for their own speech.
Bookstores, newsstands, and libraries, which distribute copies printed by others; likely also property owners on whose property people might post or write things.
Free to choose what to distribute.
Liable on what we might today call a notice-and-takedown basis.
Telephone companies, cities on whose sidewalks people might demonstrate, or broadcasters running candidate ads that they are required to carry.
Generally legally forbidden from choosing what goes on their property.
Not liable at all.
The two pre-§ 230 Internet libel decisions, Cubby v. Compuserve, Inc. and Stratton Oakmont, Inc. v. Prodigy Services Co., seemed to roughly distinguish services that edited, which were treated as publishers and were thus potentially legally liable for others’ material that they didn’t edit out, from services that didn’t edit, which were treated as distributors and were thus at least largely immune. The Cubby / Stratton Oakmont results encouraged providers not to restrict speech in their chat rooms and other public-facing portions of their service: If they were to try to block or remove vulgarity, pornography, or even material that they were persuaded was libelous or threatening, they would lose their protection as distributors, and would become potentially strictly liable for material their users posted. At the time, that looked like it would be ruinous for many service providers (perhaps for all but the unimaginably wealthy, will-surely-dominate-forever America Online).
Congress, then, chose to reject the Cubby / Stratton Oakmont approach, and instead to deliberately provided conduit immunity to all entities—including those that, unlike traditional conduits, could and did select what user content to keep up. It did so precisely to encourage (though without requiring) conduits to block or remove certain speech, by removing a disincentive (loss of immunity) that would have otherwise come with such selectivity. It gave them this flexibility regardless of how the entities exercised this function. And Congress chose conduit liability (categorical immunity) rather than distributor liability (notice-and-takedown immunity). Online sites thus had the best of both worlds: the selection power of distributors, but the liability of conduits.
But that was Congress’s decision in 1996. It’s not set in stone, and not constitutionally mandated. Publisher and distributor liability is consistent with the First Amendment, despite the chilling effect it might sometimes create, so long as it complies with the New York Times v. Sullivan / Gertz v. Robert Welch rules immunizing honest mistakes (or sometimes just reasonable mistakes).
If Congress wants to return to a world where social media immunity for libel (and other torts) turns on whether social media platforms act as common carriers, it can do so. I’m not at all sure that’s the right approach as a policy matter, especially since immunity from tort liability has helped many small and midsized online platforms thrive, and those platforms’ editorial power has often been valuable. But it does reflect an important practical reality: Immunity from tort liability is what also helped the major platforms become so big, powerful, and capable of influencing public debate—thus helping create the problems to which common carrier status might be a solution.
This sort of conditional immunity might also apply to platforms’ recommendation function and conversation functions. (Again, I’m not certain that such conditional immunity is a good idea, but here I’m speaking only of the constitutional question.) As I mentioned, platforms have a First Amendment right to choose what to recommend, just as newspapers have such a right. But it doesn’t follow that they have complete First Amendment immunity from (say) defamation liability when they recommend something that proves to be libelous. And Congress can offer this extra immunity but only for recommendations or conversation management actions that operate in viewpoint-neutral ways.
Some such conditional benefits may violate the unconstitutional conditions doctrine. But I tentatively think that a narrowly crafted and viewpoint-neutral condition such as this one would be constitutional.
When the government offers speakers a benefit not mandated by the First Amendment, it can generally attach conditions to that benefit, so long as the speakers remain free to say what they want when they aren’t using the benefit. Thus, for instance, the government may provide that charitable contributions to advocacy groups are tax-deductible, but only if those groups don’t use such contributions for electioneering, so long as the groups remain free to engage in electioneering using non-tax-deductible funds. The government may set forth “conditions that define the limits of the government spending program—those that specify the activities [the government] wants to subsidize.”  The conditions become unconstitutional only if they “seek to leverage funding to regulate speech outside the contours of the program itself.”
I think the same would apply to subsidies that come in the form of financially valuable immunity from tort law claims, and not just to subsidies in the form of financial grants or tax deductions. Congress can’t say to a platform, “so long as any of your actions are immune under § 230(c)(1), you must accept restrictions on all your First-Amendment-protected activities, even if those don’t take advantage of the immunity”: That would be unconstitutionally “leverag[ing] funding to regulate speech outside the contours of the program.” But it can say, for instance, “if you want § 230(c)(1) immunity for your decisions about which material to recommend—so that if you recommend something defamatory, for which you might be held liable under standard defamation principles, you would be immunized from such liability—that will only be available for recommendations that you make in a viewpoint-neutral way.”
 Cf. Tushnet, supra note 36.
 Restatement (Second) of Torts § 578 (1976).
 A newspaper, for instance, can be sued for libel in a letter to the editor. See id. In practice, there is some difference between liability for third parties’ speech and for the company’s own—a newspaper would be more likely to have the culpable mental state for the words of its own employees. But, still, publishers are pretty broadly liable, and have to be careful in choosing what to publish.
 Id. § 581.
 Hellar v. Bianco, 111 Cal. App. 2d 424, 425 (1952) (dealing with “libelous matter [written on a tavern restroom wall] indicating that appellant was an unchaste woman who indulged in illicit amatory ventures”; “[t]he writer recommended that anyone interested should call a stated telephone number, which was the number of the telephone in appellant’s home, and ‘ask for Isabelle,’ that being appellant’s given name”).
 A bookstore, for instance, isn’t expected to have vetted every book on its shelves, the way that a newspaper is expected to vet the letters it published. But once it learns that a specific book included some specific likely libelous material, it can be liable if it kept selling the book. See Restatement (Second) of Torts § 581; Janklow v. Viking Press, 378 N.W.2d 875 (S.D. 1985).
 Phone companies are common carriers. Cities are generally barred by the First Amendment from controlling what demonstrators say. Federal law requires broadcasters to carry candidate ads unedited. 47 U.S.C. § 315(a). New York’s high court adopted conduit immunity in 1999 for e-mail systems, even apart from § 230; though e-mail services are not legally forbidden from excluding certain messages based on viewpoint, the court stressed that their “role in transmitting e-mail is akin to that of a telephone company, which one neither wants nor expects to superintend the content of its subscribers’ conversations.” Lunney v. Prodigy Servs., 94 N.Y.2d 242, 249 (1999).
 For instance, even if a phone company learns that an answering machine contains a libelous outgoing message, and does nothing to cancel the owner’s phone service, it can’t be sued for libel. See Anderson v. N.Y. Telephone Co. 320 N.Y.2d 746, (1974); Restatement (Second) of Torts § 612. Likewise, a city can’t be liable for defamatory material on signs that someone carried on city sidewalks (even though a bar could be liable once it learned of libelous material on its walls), and a broadcaster can’t be liable for defamatory material in a candidate ad. Farmers Educ. & Coop. Union v. WDAY, Inc., 360 U.S. 525, 528–29, 531 (1958).
 Cubby held that ISPs (such as Compuserve) were entitled to be treated as distributors, not publishers. Stratton Oakmont held that only ISPs that exercised no editorial control (such as Compuserve) over publicly posted materials would get distributor treatment, and service providers that exercised some editorial control (such as Prodigy)—for instance, by removing vulgarities—would be treated as publishers. Cubby v. Compuserve, 776 F. Supp. 135, 140–41 (S.D.N.Y. 1991); Stratton Oakmont, Inc. v. Prodigy Servs. Co., No. 31063/94, 1995 WL 323710, at (N.Y. Sup. Ct. May 24, 1996).
Neither considered the possibility that an ISP could actually be neither a publisher nor a distributor but a categorically immune conduit, perhaps because at the time only entities that had a legal obligation not to edit were treated as conduits. And Stratton Oakmont‘s conclusion that Prodigy was a publisher because it “actively utilize[ed] technology and manpower to delete notes from its computer bulletin boards on the basis of offensiveness and ‘bad taste,'” is inconsistent with the fact that distributors (such as bookstores and libraries) have always had the power to select what to distribute (and what to stop distributing), without losing the limited protection that distributor liability offered. Id. at *4.
 See Tushnet, supra note 45, at 1010 n.100 (distinguishing “equal” “treatment of privilege and liability” for traditional conduits like telephone companies from “disjoined” treatment for Internet intermediaries).
 Nor is there an unconstitutional conditions problem here, just as there wasn’t with the Solomon Amendment in Rumsfeld v. FAIR: As I’ve argued above, Congress has the power to make social media platforms common carriers generally, so it has the power to make them common carriers if they want the benefits of immunity.
 Cf. Balkin, supra note 91, at __ (suggesting that intermediary immunity for social media platforms could be conditioned “on accepting obligations of due process and transparency,” though concluding that it would be a bad idea to require viewpoint-neutral moderation). But see Eric Goldman, Want to Kill Facebook and Google? Preserving Section 230 Is Your Best Hope, Balkinization, June 3, 2019 (arguing that § 230(c)(1) immunity, by “reduc[ing] barriers to enter the online republishing marketplace,” “keeps the marketplace open for the next generation of startups that hope to usurp the current Internet giants”).
 Of course, they would have the immunity secured by New York Times Co. v. Sullivan, 376 U.S.254, 282-83 (1964), Curtis Publishing Co. v. Butts, 388 U.S. 130 (1967), and Gertz v. Robert Welch, Inc., 418 U.S. 323 (1974), for speech about public officials, public figures, and private figures (on matters of public concern), respectively. But even the strongest such immunity allows liability for speech conveyed with knowledge that it’s false or likely false; it doesn’t amount to categorical immunity regardless of mental state.
 Regan v. Taxation with Representation of Wash., 461 U.S. 1983 (1997). This is generally done by allowing groups to set up two separate affiliates, a 501(c)(3) that takes tax-deductible contributions but doesn’t electioneer, and a 501(c)(4) that electioneers (so long as supporting or opposing candidates isn’t its primary activity) but doesn’t take tax-deductible contributions. See IRS, Operational Requirements: Endorsing Candidates for Public Office (updated Dec. 8, 2020), https://perma.cc/ZWX4-X8L6.
 Agency for Int’l Dev. v. All. for Open Soc’y Int’l, 570 U.S. 205, 214 (2013).
 Id. at 214–15.