---

Community Notes Resources

If you're interested in learning more about the machine learning and scoring algorithm behind Community Notes, these links should be helpful:

• Short overview talk: How did X use bridging-based ranking for community notes?
• This is a similar short talk I gave at d/acc day: Community Notes: Scaling Public Epistemics
• The main original paper on Community Notes (previously called Birdwatch) describes the algorithm/product, surveys of note quality, and an A/B test measuring reduced engagement rates on posts with notes.
• For some of the history and behind-the-scenes of Community Notes' origin story, see this interview: The Making of Community Notes: Asterisk Magazine
• The code and data are 100% public and free! Also see the current algorithm description.
• How should LLM-written notes be incorporated into Community Notes? One promising way is proposed in Supernotes, where generated notes synthesize the best elements of multiple proposed notes, and are scored by a simulated jury of raters with diverse viewpoints.

---

External Community Notes Analyses and Research

Because our data and algorithm code are fully public, many external researchers have been able to study Community Notes! Here is a very incomplete sampling of research done externally:

• Community notes increase trust in fact-checking on social media finds survey participants rate posts as less accurate when displayed with a note vs. without.
• In 2024, researchers analyzed notes about vaccines and found (according to them) that ~97.5% of notes were entirely accurate and ~99.5% of notes were partially accurate: Characteristics of X Community Notes Addressing COVID-19 Vaccine Misinformation.
• Vitalik Buterin messed around with the algorithm to verify that it does what it's supposed to, at a high level: What do I think about Community Notes?
• 3 separate independent research groups found reductions in spread of misleading posts due to Community Notes, after notes are applied. This is notable because the ranking algorithm ignores whether posts have notes, so these behaviors are driven by organic user behavior.
  • Community Notes Moderate Engagement With and Diffusion of False Information Online finds -46% reposts, -44% likes, -23% replies, and -14.0% views after posts get notes attached.
  • In Collaboratively adding context to social media posts reduces the sharing of false news, Renault et al. found that after a note is applied, reposts drop nearly in half, and authors delete their posts ~80% more often.
  • Community-based fact-checking reduces the spread of misleading posts on social media finds a 62% drop in spread, and 103% increase in voluntary deletions.
  • Unfortunately, these three papers all use data from before we dramatically sped up the speed at which notes are applied in mid-late 2024. I'm very curious to see the updated numbers on more recent data which I'd expect would show much larger total drops in reposts/views/etc.

---

Find me on X

True

— Elon Musk (@elonmusk) July 2, 2024

As always, critically important that Community Notes apply to all accounts, and are written and rated by users with no process for Elon or any employee to make exceptions https://t.co/5svIsuudFi

— Jay Baxter (@_jaybaxter_) October 25, 2024

Community Notes don't affect post ranking (and shouldn't, in general -- they often add useful context to correct posts)

But we measured a ~25-35% drop in organic likes & RTs (& smaller drop in QTs, since QTs are sometimes dunks...) in an A/B vs. the same post w/o a note

We may… pic.twitter.com/OPwN3LxLQW

— Jay Baxter (@_jaybaxter_) January 4, 2024

Imitation is the sincerest form of flattery <3

This will be very positive for the quality of information on the internet. Fingers crossed they also go open source & open data https://t.co/dKC18spUJx

— Jay Baxter (@_jaybaxter_) January 7, 2025

Jack called it https://t.co/VNc69pS4p4

— Jay Baxter (@_jaybaxter_) January 7, 2025

To be honest I thought they'd take the core idea and then rebuild it more from scratch, so it's a big vote of confidence to see them copy many of the lower-level details too.

I do believe Community Notes should be everywhere. We open source our data and code after all, and I…

— Jay Baxter (@_jaybaxter_) June 17, 2024