An unscientific peek inside Meta’s Community Notes shows the pilot is 'not ready for prime time'

When Meta killed its third-party fact-checking program in the U.S. earlier this year, the social media giant behind Facebook and Instagram said it would replace its professional fact-checkers with a crowdsourced Community Notes program similar to one launched by Twitter.

Meta has said it will continue to use automated systems and some human moderators, though the humans will be reserved for what Meta considers “the highest-severity harms, such as terrorism,” according to reporting in Platformer. “It will be up to individual users to do the rest.”

Three months after phasing out the third-party fact-checkers, the crowdsourced program remains in pilot phase. But Indicator co-founder Alexios Mantzarlis found a way to peek behind the curtain. Mantzarlis asked three Facebook and Instagram users in the Community Notes pilot to screenshot what they saw over four days. (“This is obviously not a representative analysis of the program, but absent any kind of data from the company, it’s also all we can do,” Mantzarlis writes.)

Mantzarlis knows a little something about fact-checking — he ran the International Fact-Checking Network after all — and has written about both the pros and cons of crowdsourced programs. For his newsletter Indicator, he evaluated 61 posts collected by the Community Notes testers.

Based on the screenshots he collected, Mantzarlis concluded “it’s clear that Community Notes on Facebook and Instagram is not yet ready for prime time.”

There were some bright spots. Mantzarlis found helpful Community Notes appended to posts with out-of-context or AI-generated content, including two videos from 2020 that created the false impression they were from recent protests.

Some Community Notes, while not obviously wrong, weren’t “terribly helpful” in Mantzarlis’s estimation. Several were appended to what Mantzarlis deemed un-fact-checkable content, including a community note on three very blond dudes flexing over the caption “Summer Aesthetics 2025.”

But Mantzarlis also found some Community Notes that were “flat out inaccurate.”

A post by a Los Angeles TV station about a family detained by ICE during a routine immigration appointment was annotated with an unrelated story about a different young woman. Based on the url suffix, it looks like the note writer may have been sent off track by ChatGPT.

A note on an anti-Trump post about crimes committed by undocumented immigrants linked to a page of statistics by ICE that does not contradict the statement. It also appears to misunderstand the concept of “on average.”

You can find the full (paywalled) post and a spreadsheet with annotated Community Notes posts here.