How Community Notes Succeeds and What It Can Do Better

A deep dive into X's Community Notes system, analyzing its strengths and exploring opportunities for improvement.

If you've spent any time on X (formerly Twitter) in the past few years, you've probably noticed those gray boxes that sometimes appear beneath posts. "Readers added context they thought people might want to know," they say, followed by a correction or clarification. This is Community Notes in action, and it represents one of the more interesting experiments in fighting misinformation at scale.

But here's the question we've been asking ourselves: Does it actually work? And if so, what can we learn from both its successes and its failures?

What Community Notes Actually Is

At its core, Community Notes is a crowdsourced fact-checking system. Instead of relying on a team of professional fact-checkers (which doesn't scale) or algorithmic content moderation (which lacks nuance), X decided to let users themselves annotate potentially misleading content.

The approach is refreshingly transparent. The algorithm that decides which notes get displayed is completely public. Researchers at the University of Rochester and elsewhere have been able to dig into exactly how it works. This matters because one of the persistent criticisms of content moderation has always been its opacity. With Community Notes, you can actually see the math.

Here's how it works: users who opt into the program can write notes on any post they think might be misleading. Other contributors then rate those notes as helpful or not helpful. But here's the clever part: a note doesn't just need a lot of "helpful" ratings to get displayed. It needs helpful ratings from people who typically disagree with each other.

The Surprising Success Story

When we started digging into the research, we expected the results to be mixed at best. Social media interventions often look good in theory and fall apart in practice. But the data on Community Notes is genuinely impressive.

A study published in PNAS found that posts with Community Notes attached saw a 46% reduction in reposts and a 44% reduction in likes. That's not a marginal effect. That's nearly cutting viral spread in half. An INFORMS study found something equally striking: authors of flagged posts were 32% more likely to delete them entirely. The notes aren't just warning readers; they're changing the behavior of the people posting misleading content in the first place.

Perhaps most compelling is the accuracy data. Researchers at UC San Diego analyzed vaccine-related Community Notes and found a 97.5% accuracy rate. When the crowd says a claim is misleading, the crowd is almost always right.

Internal Twitter research (conducted before the platform became X) found that notes led to a 25-34% reduction in retweets of misleading content. That's consistent with the external studies and suggests these effects are robust across different measurement approaches.

The Bridging Algorithm: Why Cross-Partisan Agreement Matters

The secret sauce of Community Notes is what's called the "bridging algorithm." This is worth understanding because it addresses one of the thorniest problems in content moderation: accusations of political bias.

Traditional fact-checking by journalists, algorithms, or platform employees inevitably gets accused of favoring one side or the other. If a conservative claim gets flagged, conservatives cry bias. If a liberal claim gets flagged, liberals cry bias. This dynamic has made effective content moderation politically toxic.

Community Notes tries to sidestep this problem with a simple but powerful requirement: notes only get displayed if they receive support from people across the political spectrum. The algorithm doesn't just count votes. It weights them by the rater's position in a multi-dimensional "viewpoint space" derived from their rating history.

In practice, this means a note that only conservatives find helpful won't get displayed, even if thousands of conservatives vote for it. Same for notes that only appeal to liberals. To show up, a note needs to convince people who usually disagree.

This design choice has real consequences. It means Community Notes surfaces claims that reasonable people across the political spectrum can agree are genuinely misleading, rather than claims that are merely controversial or politically inconvenient for one side.

Where It Falls Short

So Community Notes works. But it doesn't work perfectly, and the limitations matter.

The biggest problem is speed. Research published on arXiv shows that misinformation spreads fastest in its first few hours. This is the viral window when a misleading post goes from a few hundred views to millions. Community Notes, unfortunately, tends to show up after this window has closed.

Why? Because the bridging algorithm requires time. Contributors need to find the post, write a note, and then other contributors need to rate that note. And not just any raters, but raters from diverse viewpoints. All of this takes hours or sometimes days. By the time a note appears, the damage is often done.

There's also a coverage problem. Community Notes works best for English-language content on topics that have engaged contributor communities. Misinformation in other languages, or on niche topics, often goes unnoted entirely.

And then there's the sustainability question. As X has grown and the information environment has become more chaotic, the demands on Community Notes contributors have increased. Research suggests that contributor burnout is a real risk. The system depends on volunteer labor, and volunteers eventually get tired.

Opportunities for Improvement

If we were advising the Community Notes team (we're not, but bear with us), here's what we'd suggest:

Faster deployment during viral windows. This is the big one. The current system optimizes for accuracy at the cost of speed. But if a note shows up after 90% of a post's engagement has already happened, how much does accuracy matter? A provisional note system with a lower confidence threshold during the early viral stage, upgraded to a full note once the bridging threshold is met, could help.

Expanded language coverage. Misinformation doesn't respect language boundaries, and neither should fact-checking. This probably requires both recruitment of contributors in underserved language communities and improvements to the tooling to make cross-language contribution easier.

Better tooling for contributors. The people who write Community Notes are doing difficult, often thankless work. Better dashboards, clearer feedback on what makes notes effective, and some form of recognition for high-quality contributors could help reduce burnout and improve note quality.

What Other Platforms Can Learn

Community Notes isn't a complete solution to misinformation. Nothing is. But it demonstrates something important: crowdsourced fact-checking can work, at scale, with high accuracy, in a way that's resistant to accusations of political bias.

The scalability point matters most. Professional fact-checkers are expensive and slow. Algorithmic moderation is fast but brittle and opaque. Community Notes occupies a middle ground: fast enough to be useful (most of the time), accurate enough to be trusted, and transparent enough to be defensible.

The transparency itself is worth emphasizing. By making the algorithm public, X has enabled independent verification of how the system works. Researchers can (and do) audit the results. This builds trust in a way that black-box moderation never can.

And the cross-partisan consensus model addresses the political legitimacy problem that has plagued content moderation for years. It's hard to claim bias when the very definition of a good note is one that people across the political spectrum agree is helpful.

We don't know what the future of Community Notes looks like. X as a platform has been through substantial changes, and the information environment continues to evolve in ways that challenge any moderation system. But the underlying ideas of crowdsourced fact-checking, transparent algorithms, and cross-partisan consensus feel like they'll outlast any particular implementation.

The question isn't whether Community Notes is perfect. It isn't. The question is whether it points toward something better than what we had before. On that front, the evidence is encouraging.

Sources & References7