How Meta’s take on Community Notes misses the mark

Crowdsourced moderation is better than nothing — but how much better remains an open question

Casey Newton

Mar 13, 2025 — 10 min read

A Community Note added to an Instagram post. (Meta)

Eager to curry favor with the incoming Trump administration, Meta kicked off 2025 by announcing a large-scale retreat from its content moderation and fact-checking efforts in the United States. The company had invested billions of dollars in building sophisticated automated systems that detected and removed vast quantities of misinformation, hate speech, and other things that most people would rather not see. But scared of retribution from Trump, and frustrated with high error rates in its systems, CEO Mark Zuckerberg said the company would stop scanning most new posts for violations and instead begin relying more on the user base to report instances of harm.

As part of these changes, Meta also said it would stop funding third-party fact-checking organizations, which it said had broadly failed to improve trust in its platforms. In their place, Zuckerberg said the company would borrow a feature from X: Community Notes, a way for users to add context to posts by providing additional information about what they are seeing. If a viral photo of a celebrity was actually generated by artificial intelligence, or an image presented as breaking news was actually taken years earlier, Community Notes can help audiences understand the truth of what they’re seeing.

In a sign of how abruptly Meta had decided to abandon years of investment in content moderation, work on its Community Notes feature had barely begun when Zuckerberg announced it was coming. But today the company announced that they are almost ready to launch: on Tuesday, the company will begin accepting notes across Facebook, Instagram and Threads. (It said it would take an unspecified amount of time to evaluate them before posting them publicly.)

“We expect Community Notes to be less biased than the third party fact checking program it replaces because it allows more people with more perspectives to add context to posts,” the company said in an unsigned blog post.

Let’s start with the one true innovation that Community Notes brought to content moderation after Twitter began experimenting with them in 2021. Community Notes are the most prominent example to date of a product that uses what’s known as bridging-based ranking — algorithms that reward behavior that bridges people with different views. When Twitter expanded the program then known as Birdwatch, it decided to display notes on tweets only if they had been marked as helpful by people who normally disagree. If a post was upvoted only by people with left- or right-leaning views, it would not appear. But if a post could get left-leaning and right-leaning users to agree that a post was helpful, Twitter would display it right on the tweet.

This incentivized people to add notes that were based in fact rather than opinion. In its early tests, Twitter found that surveys showed people were 20 to 40 percent less likely to agree with tweets rated as misleading after they read Birdwatch notes.

This is good and useful work, even if it stops short of what author Aviv Ovadya called for in his original proposal for bridging-based ranking. “Imagine if Facebook rewarded content that led to positive interactions across diverse audiences, including around divisive topics,” they wrote in 2022. “How might that change what people, posts, pages, and groups are successful?” I’d still love to find out.

That core innovation aside, though, Community Notes have several important limitations, according to former Twitter executives I’ve interviewed.

One, despite being available globally, they appeared relatively rarely outside the United States. Some Twitter executives speculate that this is due to lack of trust in the company, particularly in countries with more authoritarian leaders. If you correct a comment made by your fascist president, will X share that comment with the government? The fear alone has been enough to limit notes’ usefulness abroad.

Two, Community Notes take longer to appear on viral posts than Meta’s fact checks typically do. This is a downside of bridging-based ranking: it typically takes many hours for a post to accumulate enough votes from people with different perspectives for the system to deem it helpful. In situations where every minute matters — the aftermath of a tragedy, or in the midst of an election — Community Notes may not arrive until they’re too late to do much good.

Three, Community Notes on X often rely on information from the independent fact-checking programs that Meta just pulled the funding for. Meta has long said it doesn’t want to be an “arbiter of truth,” but it has funded those arbiters for the past several years, and it’s not clear whether anyone will step up to replace it. If no one does, Community Notes will suffer both on X and on Meta’s platforms.

Most importantly, though, Meta said it will not reduce the distribution of posts that its users have determined to be false. That’s the opposite of the approach it took with fact checks. Until last year, when third-party fact checkers rated a post as false, Meta reduced its distribution significantly. While fact checks themselves seem to have done little to change people’s beliefs, showing them fewer false posts overall seems like an obvious benefit to the quality of Meta’s services.

Now, no matter how false or misleading a post is, Meta has committed to showing it to as many people as its ranking systems predict will find it engaging.

When I asked about its rationale, the company told me that having a note attached doesn’t mean that a post is false — only that people thought it needed more context. And even in cases where someone is lying, Meta said it wants to move away from a more “punitive” system, instead showing people additional information and letting them make up their own minds.

“Notes … won’t have penalties associated with them the way fact checks did,” the company said in its blog post today. “Fact checked posts often had their distribution across our platforms reduced. That won’t be the case with posts that have notes applied to them. Notes will provide extra context, but they won’t impact who can see the content or how widely it can be shared.”

And to be fair, Twitter took this approach as well — it never downranked content that has Community Notes attached. And notes can result in content being downranked in an indirect way: seeing them generally makes people less likely to interact with a post, which in turn results in platforms showing them to fewer people.

At the same time, Twitter’s Community Notes — like Meta’s fact-checking program — was never intended to be a load-bearing pillar of either company’s content moderation efforts. Rather, it was intended to complement other, more robust efforts, such as relying on internal teams and automated systems to label obvious falsehoods.

Meta will continue to use some automated systems and human moderators, though it has said it will focus them on what it considers the highest-severity harms, such as terrorism. It will be up to individual users to do the rest. (The company says “around” 200,000 people in the United States have signed up to participate so far.) How this will affect the average experience of using Facebook, Instagram, or Threads is still anyone’s guess.

Still, it seems likely that people will now see more nonsense that they did before. If all goes well, it will come appended with a note informing you that you’re looking at nonsense. Meta has expressed confidence that this approach will build trust in its systems. Even if it does so, though, it will come at the expense of wasting huge amounts of its users’ time and attention.

An earlier generation of leaders at Meta argued, among other things, that filling up feeds with obvious falsehoods made for a bad user experience. But those leaders are gone, and in the place of third-party fact checks you can now find Shrimp Jesus and other AI slop.

There was a moment when Meta seemed to take seriously the idea that its users might want the truth — might want journalism, even. But that belief has been thoroughly bullied out of them, and there’s no telling what that might mean for the rest of us.

Corrections, 6:29 p.m.: This post was updated to reflect the proper authorship of the bridging-based ranking paper and to note that while it will begin accepting notes next week, Meta will not publish them until later.

On the podcast this week: Kevin and I sort through what's going on with Apple Intelligence. Then, the Times' Adam Satariano joins us to explain how Starlink became the most powerful telecom in the world. And finally, a new study asks: is AI making us dumber?