Meta’s election Community Notes caught politicians — but weren’t always fact checks

The 2026 US midterms were the first nationwide vote to be held since Meta abolished the third-party fact-checking program in the country and replaced it with Community Notes.

So, how did the crowdsourced fact-checking program do? To answer that question, I reviewed all notes that appeared in the month before the election.

Fully a quarter were about politics, including some that debunked claims by politicians. However, many notes weren’t fact checks and instead often provided unnecessary context or opinions.

In March, the Oversight Board said Meta promised it would publish data "on par" with X about its crowdsourced fact-checking program. The Board took this to mean "the full text of notes and their associated geographies, ratings for each note, status history, user enrollment and the requests for notes on content."

For now, all Meta publishes is a dashboard linking out to 40-odd helpful notes published in the previous 24 hours. (Meta declined to comment on its data sharing plans.)

But I’ve been able to collect a more complete dataset because the dashboard exposes a data feed that appears to contain all of the helpful notes from the previous 24 hours. Over the past month, the number of crowdsourced context cards deemed helpful ranged between 119 and 369 a day. (This excludes notes that were proposed but not upvoted by other users and therefore not shown on Meta's platforms.)

A Meta spokesperson told Indicator that an average of 240 notes a day are published through the program. That's roughly in line with my count, which gives me confidence that I collected most of the helpful notes. The company also confirmed that the portal doesn't contain notes from the recent expansion of Community Notes to Spanish-language Latin American countries.

First, I wanted to understand how much of the corpus was actually about politics.

To do that, I used Claude Haiku via an API to read the text of 6,519 notes posted across Facebook, Instagram, and Threads between Oct. 4 and Nov. 3 and determine whether they were about US politics. I manually reviewed 5% of classified notes and agreed with the ratings 90% of the time. (The prompt and more details are in the Info Box at the end of this article.) 

Overall, about a quarter of Community Notes were about US politics, with a notably higher share on Threads and Facebook than on Instagram.

I used a similar method to code the topic of each political note and grouped similar notes to find recurring narratives.

I also extracted and categorized the sources cited by the Community Notes, along with any engagement data I could gather from the three platforms. 

Finally, I reviewed who got fact-checked the most, with a particular focus on accounts belonging to elected officials and political parties. Meta used to forbid fact-checkers from flagging false posts by politicians, so the (relatively limited) number of notes on their content is a change from the past.

Indicator members can search and filter the 1,669 political notes in the interactive below and also download the full dataset of 6,519 helpful notes collected in the past month.