Post-Musk X: The Natural Experiment

In October 2022, Elon Musk paid $44 billion for Twitter, fired most of the workforce, dismantled the Trust and Safety apparatus, eliminated the COVID misinformation policy, and reinstated more than 12,000 previously suspended accounts after a poll where 72.4 percent of respondents approved a general amnesty. Inside three months, a major consumer-internet platform had rolled back its content-moderation regime as completely as any platform has ever been rolled back.

This was, accidentally, the largest natural experiment in platform content moderation ever conducted. The question it answered was not the one the experimenter had in mind.

The Cuts

Twitter’s pre-acquisition headcount was approximately 7,500. By the end of Musk’s first quarter, the company employed roughly 2,000. The Trust and Safety function — which had run at roughly 2,000 full-time employees with thousands of contractors — was reduced to fewer than 20 full-time staff. The COVID misinformation policy was rescinded outright. The general amnesty restored accounts that had been suspended for everything from harassment to coordinated inauthentic behavior to election misinformation.

The cuts were not the experiment’s confound. They were the experiment’s intervention. The platform stopped doing the thing the moderation apparatus had been doing. The question was what would happen.

The Results

Advertising revenue dropped approximately 60 percent within twelve months. Musk himself stated the figure in September 2023. Fidelity, which held a minority position in the company through one of its funds, marked the position down approximately 72 percent from the acquisition price. The Center for Countering Digital Hate documented that X failed to act on 99 percent of flagged hate speech from paid Blue subscribers in its sampling.

Musk amplified an antisemitic conspiracy theory in November 2023, then traveled to Israel and did an apology tour with the families of October 7 hostages. The European Union opened a formal Digital Services Act investigation in December 2023 and issued a €120 million enforcement action by December 2025. Proceedings against Meta and TikTok remained pending.

The financial bleed was the test result. The advertiser revolt was the immediate punishment. The legal exposure was the slower one.

Go Fuck Yourself

At the November 2023 DealBook Summit interview with Andrew Ross Sorkin, Musk told the advertisers who had withdrawn from the platform to “go fuck yourself.” He named Disney’s Bob Iger directly. The clip went viral. The advertisers did not return.

The apology tour ran into the first quarter of 2024. The numbers did not respond.

The Dam Metaphor

The post-acquisition X is the case that demonstrates the difference between a single moderation mechanism and the layered system that produces content moderation outcomes.

The platform-level mechanism — content review, suspensions, policy enforcement, fact-check labels — was dismantled. The advertiser-funded economic incentive that had supported the mechanism remained, and the advertisers responded to the dismantlement by leaving. The legal mechanism — the EU DSA, the UK Online Safety Act, the various national hate-speech regimes — remained, and the European Commission opened the investigation that produced the €120 million enforcement action two years later. The competitive mechanism, alternative platforms, produced Threads, which crossed 100 million users within five days of its July 2023 launch, and Bluesky, which crossed 20 million by late 2024. The internal mechanism, under Linda Yaccarino’s CEO tenure starting June 2023, quietly re-tightened moderation: restoring Community Notes as a labeling mechanism, reintroducing election-integrity policies, settling with advertiser coalitions, hiring back some of the Trust and Safety capability the initial cuts had removed.

You can break the platform-level dam. The water has nowhere to go because there are four dams in series behind it. Knocking down the first one releases water that fills the reservoir behind the second.

Community Notes

Community Notes is the part of post-Musk X that worked. The bridging algorithm, which surfaces notes when raters from historically opposed perspectives both rate them helpful, produces fact-check labels on clear factual claims with measurable accuracy and acceptable latency. The data is public. Researchers can study it. The mechanism is auditable in a way that the previous trust-and-safety regime was not.

Community Notes works for factually testable claims. It does not work for hate speech, harassment, or breaking misinformation. The bridging consensus is too slow to catch viral misinformation in its viral window. The “is this hateful” question is not the kind of factual claim the algorithm is designed to adjudicate. The mechanism is a partial substitute, not a full one. For the parts where it works, it is the most transparent fact-checking system any major platform has deployed.

The Ratchet Question

Did the platform-level reversal at X demonstrate that the moderation ratchet can be reversed?

Mechanically, the moderation regime was reversed. Policies were repealed. Staff were fired. Accounts were reinstated. The visible mechanism that had produced the previous moderation outputs was dismantled.

Economically, the market punished the dismantlement. Advertiser revenue dropped 60 percent. The acquisition financing turned into a problem. The company’s enterprise value dropped substantially. Whatever lesson the next platform CEO takes from the X experiment, “dismantle moderation and the advertisers will love you” is not the lesson.

Legally, the European Union forced compliance through statute. The DSA’s six-percent-of-global-turnover fine architecture was designed precisely for the scenario where a platform refuses to moderate. X was the first major test case. €120 million was the first enforcement action. The architecture is operational.

Competitively, Threads and Bluesky adopted the moderation standards X abandoned. The user base that wanted moderation did not have to remain on X. It went elsewhere. The competing platforms are smaller than X, but they exist. The competitive ratchet on moderation is bidirectional. It can pressure a platform toward less moderation in some periods and toward more in others, but in the post-X period, it pressured toward more.

Internally, the platform’s own management re-tightened. Yaccarino’s tenure was the institutional response to the financial damage. Community Notes was scaled. Election-integrity policies returned in modified form. The walk-back was partial, gradual, and unannounced, but it was a walk-back.

The platform-level ratchet was dismantled. The economic ratchet imposed it from below. The legal ratchet imposed it from above. The competitive ratchet imposed it laterally. The internal ratchet reimposed it as the financial consequences became unsurvivable. You can break one ratchet. The others compensate.

There is not one dam. There are five dams in series.

Corporate Self-Regulation Pattern

The post-X period is consistent with what every adjacent industry has shown about self-regulation. PCI-DSS works because the card brands enforce it. Aviation safety works because the FAA grounds aircraft. UL works because building codes require it. INPO works because the NRC can close a reactor. Pure voluntary self-regulation without external enforcement does not succeed at scale.

The pre-Musk Twitter moderation regime was partially voluntary — the platform’s own policy choices, partially influenced by advertiser preferences and legal exposure. The voluntary layer collapsed when Musk took control. The non-voluntary layers — advertiser economics, EU regulation, competitive substitution — held. The visible part of the system was the voluntary part. The structural part of the system was everything else.

The implication is uncomfortable for both sides of the moderation debate. The advocates of stronger moderation cannot conclude that the X experiment vindicated their model. The moderation outcomes that returned were imposed by economic and legal pressure, not by re-arguing the case for moderation as a policy preference. The advocates of free-speech absolutism cannot conclude that the X experiment vindicated theirs. The cost of the dismantlement was a 60 percent revenue drop, a multi-hundred-million-euro EU fine, and a competitive migration to platforms that retained the moderation regime.

The experiment ran. The water did not disappear. It found new channels.

The platform fired the moderators. The advertisers left. The European Commission opened the investigation. The competitors absorbed the migrating users. The internal management quietly restored the policies. The shareholder mark-downs were absorbed. The walked-back regime arrived at a place not very far from where it started, by a route the experimenter had not planned.

The kill switch is dismantleable. The five dams behind it are not.

Click.

The receipts (free, on this site): the Twitter Files index · deplatforming

This research appears in The Ratchet, Chapter 3.