An 18-month study found that fact-checking on Facebook reduced engagement with false info

It has been exactly one year since Mark Zuckerberg sacrificed the US arm of Meta’s fact-checking program to endear himself to Donald Trump. The rollback has, among other things, resulted in a resurgence of viral hoaxes that the initiative was originally created to combat.

60 percent of Americans believe platforms should take steps to restrict false information online. Across eight countries around the world, large majorities think social media should be held responsible for showing potentially false information to users. Perhaps for this reason, Meta has yet to pull the plug on its fact-checking program globally.

Stephan Mündges, coordinator of the European Fact-Checking Standards Network, said that “Meta hasn't announced an end to their fact-checking program outside the US, neither publicly nor privately, which is why we expect it to continue.”

A spokesperson for Meta told Indicator that the fact-checking program, now in its tenth year, is still operational outside the United States. In fact, Indicator has separately confirmed that Meta renewed many of its contracts with fact-checking partners outside the United States through to the end of 2026, although it is often paying significantly less than in the past. However, the company is still planning to roll out Community Notes, the program that replaced fact-checking in the United States, to the whole world.

When Zuckerberg ended the US fact-checking program, he said “fact-checkers have just been too politically biased and have destroyed more trust than they have created.” He provided no evidence for that assertion.

Now, a new working paper led by researchers at Sciences Po in Paris shows that the fact-checking program can reduce the reach of false and misleading on Facebook. Moreover, part of this reduction may be driven by user behavior rather than algorithmic penalties. (Disclosure: as director of the International Fact-Checking Network at the time, I played an important role getting the program off the ground in 2016.)

The study has a unique methodology. One the authors embedded as a fact-checker with AFP Factuel, the fact-checking unit of the French newswire, for 18 months starting in December 2021. Along with publishing fact checks, he collected detailed data on the origin, topic, and virality of 944 hoax narratives reviewed by Factuel.

Emeric Henry, the head of the Economics Department at Sciences Po and one of the paper’s co-authors, told me the setup was a “win-win situation” because the research group got deep access to the subject they were studying, while AFP got an additional fact-checker at no cost.

Grégoire Lemarchand, deputy news director for digital strategy at AFP, said the newswire engaged in the partnership because it was “genuinely interested in better understanding the real-world impact of fact-checking.” He added that “the project didn’t feel intrusive because it was conducted over a long period, which allowed everything to proceed naturally.”

During the study period, Meta applied prominent visual labels and a related link to posts that were rated false by fact-checkers. Meta uses the “false” rating as a signal to reduce future distribution, though it has not disclosed the exact mechanism and has also softened language about penalties in public documentation.

The French study compared fact-checked narratives with similar claims that the AFP team decided not to verify for a range of reasons, including that definitive evidence could not be found or that there wasn’t enough time. Such unchecked narratives were used as a control to quantify the effect of a fact check on future reach.