Meta Fact-Checking Program

» see about.fb.com/news/2025/01/meta-more-speech-fewer-mistakes/ as it lived, on the wayback machine

~ the last words ~

the meta fact-checking program was not a thing i ever directly interacted with. it was a policy layer sitting behind the little grey label that showed up on some posts saying "independent fact-checkers rated this false." in turkey the program ran through teyit.org from around 2017 onward and dogruluk payi later, which meant it did technically operate on my feed. i almost never saw the labels because my facebook usage had already trailed off by then.

the program launched on 15 december 2016, six weeks after donald trump won the us presidential election. facebook was under press pressure to explain the volume of political hoaxes the platform had run through the election cycle. adam mosseri wrote the launch announcement. the model was: independent journalism organizations certified by the international fact-checking network would flag posts as false. flagged posts would get downranked and labeled. facebook itself would not decide what was true.

it grew to roughly ninety partners across more than sixty countries. by 2024 the program was one of the most visible attempts by any major platform to intervene in what its algorithm surfaced. the political framing had already shifted by then. republican politicians had been calling fact-checking a form of censorship for years. the ftc under a new chair was reading the coalition differently.

mark zuckerberg announced the us termination on 7 january 2025 in a video from the meta pr channel, five days before trump's second inauguration. the announcement said community notes would replace the fact-checker layer, cloned from what x had been running since 2021. the eu and uk operations were quietly excluded from the wind-down because the digital services act requires the platforms to keep some form of independent verification in place for european users.

~ the record ~

~ how it worked ~

the mechanic was thin by design. a partner journalist would review a flagged post, publish an article rating it false or partly false, then push that rating back through facebook's tooling. the post would get a warning overlay and would lose distribution in the news feed algorithm. it also could not be run as a paid ad. the poster kept their account. no permanent removal. the appeals process was a form that went back to the same partner.

the interior tension was always the same. fact-checkers were independent organizations, but they were paid by meta and worked through meta's tooling. news publishers on the other side of the labels said the whole thing looked like meta outsourcing an editorial decision it did not want to own. that criticism sat inside the program from day one and never really got resolved.

~ the january 7 announcement ~

zuckerberg's video framed the change as "more speech, fewer mistakes." the pr line was that fact-checkers had been "too politically biased" and that the trust-and-safety layer had grown "too restrictive." this landed differently depending on where you were reading it from. inside the fact-checking community it landed as a business partner walking away without notice. poynter's coverage the same day was polite and unhappy.

five days later trump was inaugurated. the timing was the story. the change did not have to happen that week from a product standpoint. it happened that week because the political weather had turned and meta was reading the forecast.

~ what it leaves ~

community notes replaced the labels on us posts in early 2025. the mechanic is different: contributors from a range of political perspectives write notes and other contributors rate them. a note only appears when contributors who normally disagree agree on the rating. x has been running this since 2021 and the results are genuinely mixed. it catches some hoaxes fast. it also misses whole categories of political falsehood because the raters split on the political premise underneath.

the eight-year meta fact-checking program will read historically as the moderation experiment that platforms tried once, at scale, in the aftermath of a specific election. by 2026 the community-notes model is the default across x and meta. the fact-check label as a piece of interface furniture is gone from the us feed. the political weather that killed it is not going anywhere for a while.

@}-->--

              )
             (_)
            .-'-.
            |   |
            |   |
            |   |
            |   |
          __|   |__   .-.
       .-'  |   |  `-:   :
      :     `---'     :-'
       `-._       _.-'
           '""""""

« previous  ·  random  ·  next »