An inside look at X’s Community Notes

Keith Coleman (VP of Product) and Jay Baxter (ML lead) are the product and machine-learning leaders behind X's Community Notes, a crowdsourced system for adding context to potentially misleading posts. The conversation covers how the bridging-based ranking algorithm works, the team's operating model and principles, how the product survived multiple leadership changes, and where it is headed.

01Key learnings

• Surface notes only when raters who normally disagree both find them helpful; seeking cross-polarized agreement is what keeps notes neutral, accurate, and hard to manipulate.
• Open participation to everyone with basic verification and random selection, rather than curating contributors, so the system feels fair and trustworthy.
• Set the quality bar conservatively: showing a bad note damages trust far more than withholding a good one, so the team accepts showing only about 8% of proposed notes.
• Make the product prove itself at each stage with data before expanding, which built internal confidence and eased skepticism from trust and safety teams.
• Run a small, fully focused team with one clear owner and one senior decision-maker, using light tools like a shared doc instead of heavy task management and OKRs.
• Keep core principles like transparency, open-sourced code and data, and no manual override button, since breaking them would make the product harder to trust.
• Allowing pseudonymous contributors surfaced more honest cross-partisan agreement than real-name identities, a counterintuitive finding worth testing in your own community features.
• Deleting unnecessary code and features is as valuable as shipping new ones, because maintenance burden from small incremental wins quietly accumulates.

02Frameworks mentioned

Summary and takeaways written by PM Atlas; quotes are short excerpts. © the original author.