Who checks the fact-checkers? How it works, who pays, its record

Series: How minds are moved · Fake news, fact-checks and how to check for yourself · 2 of 7

In 91% of the stories where a fact-checker shows up in LSN News, my news engine, it stands alone: in 342 of 377 stories since 7 November 2026, my engine grouped no other outlet’s article with the fact-check. Part of that is format. A fact-check is written as a verdict on a claim, with its own kind of headline, so it rarely lands in the same story as the news reports about the event. The eight fact-checking outlets among my sources, from Spain’s Maldita.es and Germany’s CORRECTIV to Snopes, Lead Stories and Ukraine’s StopFake, each count as one voice in my system, like any other outlet. People argue about fact-checkers as if they were judges. In my data they behave like what they are: small newsrooms with a method, a budget and a funder.

Contents · 9 sections

TL;DR #

• A fact-checker is an outlet. Its verdicts carry no special standing, and a US court has treated its labels as protected opinion.
• About 180 organisations are verified signatories of the International Fact-Checking Network’s code. An annual outside assessment checks them for transparency, not for being right.
• The money: in 2024, Meta’s programme paid on average 45.5% of fact-checkers’ revenue. In 2025 that fell to 34.3% of revenue, after Meta ended its US programme and switched to Community Notes.
• The record has documented misses: a lab-leak rating withdrawn, a dispute with a medical journal, plagiarism at a well-known site. There are also documented strengths: experiments in four countries found that fact-checks reduce false beliefs.
• Judge a fact-check the way you would judge any article: by its sources, its date, and the exact claim it rated.

What a fact-checker is, and what it isn’t #

Fact-checking became a genre of its own in the 2000s: outlets that pick a public claim, research it and publish a verdict, often on a scale from “True” to “Pants on Fire”.

The International Fact-Checking Network (IFCN) was set up at the Poynter Institute in 2015. In September 2016 it launched a Code of Principles, first signed by 35 organisations from 27 countries. Signatories commit to five things: non-partisanship and fairness; transparency of sources; transparency of funding and organisation; transparency of methodology; and an open and honest corrections policy. Every year, outside assessors check each signatory against the code. The IFCN’s 2026 survey reached 141 organisations, which it said were 77.5% of its signatories. That puts the network at roughly 180 members.

Note what the assessment checks: whether an outlet discloses its funders, explains its method and corrects its errors. Not whether its verdicts are right. That is left to readers, other outlets and, occasionally, courts.

A court did weigh in once. The journalist John Stossel sued Meta and Science Feedback over labels on his climate videos, and in 2022 a federal judge in California dismissed the case, holding that the labels were opinion, not statements of objective fact. That is a precise description of a fact-check’s legal status: a reasoned opinion by an outlet.

How the platform deal worked #

Most people met fact-checkers through Facebook. Its third-party programme started in December 2016: paid partner organisations rated posts, and posts rated false got a warning label and were shown to fewer people. In early 2019 Snopes and the Associated Press stepped back from the programme. Snopes, which had received $100,000 for 2017, said it was weighing the cost of the work.

On 7 January 2025, Mark Zuckerberg announced that Meta would end the programme in the United States. He said fact-checkers had “been too politically biased and have destroyed more trust than they’ve created, especially in the US”. Meta began testing Community Notes on 18 March. On 7 April 2025, Meta’s global affairs chief Joel Kaplan said the US programme was over: no new fact checks, and notes with no penalties attached.

Outside the US the programme continues. Meta says it works with more than 90 certified organisations in more than 60 languages. In March 2026 Meta’s Oversight Board noted that Meta published about 900 Community Notes in the first six months of the US rollout. In the same period in the EU, professional fact-checkers helped Meta label about 35 million Facebook posts. The Board said notes were not an adequate substitute outside the US. On 9 September 2026 Meta began testing notes in 16 Spanish-speaking Latin American countries, with Brazil excluded. Fact-checking there continues during the test, according to AFP.

Google made a quieter change: in June 2025 it stopped showing fact-check snippets in search results.

Who pays #

The money explains a lot about how fact-checkers work, as it does for any outlet. The IFCN’s State of the Fact-Checkers report for 2025 surveyed 141 organisations in 71 countries:

• Grants were the largest source of revenue, at an average of 46.2%.
• Meta’s programme fell from 45.5% in 2024 to 34.3% in 2025.
• TikTok’s programme fell from 11.9% to 5.1%.
• For 34.3% of organisations, one funder provided 75% to 100% of their income.
• 76% described themselves as financially vulnerable or in crisis.

The report adds that the end of USAID cut media funding in Eastern Europe, Asia and Africa. In Europe, the EU funds the European Digital Media Observatory, where fact-checkers take part.

Duke University’s Reporters’ Lab counts the result. As of 1 June 2026 there were 437 active fact-checking projects, down from a peak of 464 in 2024. In 2025, 10 projects opened and more than 30 stopped publishing. Logically Facts was one; it stopped after losing its Meta and TikTok revenue.

None of this proves bias. It describes dependence, which is why we ask any newspaper about its owner and advertisers. The code requires fact-checkers to disclose their funders, so you can check.

The record: misses, strengths, and the critics’ own claims #

Misses, from more than one direction:

• Lab leak. In September 2020 PolitiFact rated a lab-origin claim by the virologist Li-Meng Yan “Pants on Fire”. In May 2021 it removed the rating from its database, saying the claim was “now more widely disputed”, and kept the original in an archive. That same month Facebook dropped its own February 2021 rule against posts saying the virus was man-made. Post 3 in this series goes deeper.
• The BMJ dispute. In November 2021 the medical journal The BMJ published an investigation into a contractor that ran three of the 153 sites in Pfizer’s vaccine trial. Facebook put a “missing context” label on it, linked to a Lead Stories fact check. The journal’s editors wrote to Zuckerberg and called the check “inaccurate, incompetent and irresponsible”. Lead Stories defended the label and said it had been designed for content that is true but could mislead. Both documents are public, and you can read them side by side.
• Plagiarism. In August 2021 BuzzFeed News found that Snopes co-founder David Mikkelson had copied material from other outlets into dozens of articles. Snopes retracted about 60 of them and suspended him from writing. The error was not in a verdict but in the integrity of the outlet itself.
• Lie of the Year. PolitiFact’s 2011 pick, the Democrats’ claim that Republicans “voted to end Medicare”, was attacked from the left; its 2013 pick, President Obama’s “If you like your health care plan, you can keep it”, drew pushback too. A one-word verdict on a contested policy claim invites that.

Strengths, also documented:

• Agreement. A 2023 study in the Harvard Kennedy School Misinformation Review matched 749 claims checked by both Snopes and PolitiFact. After it adjusted for their different rating scales, the two disagreed on one.
• Effect. A 2020 meta-analysis by Walter and colleagues pooled 30 studies with 20,963 participants. Fact-checks improved accuracy, but less when the correction ran against a person’s ideology or prior beliefs. A 2021 study in PNAS by Porter and Wood ran experiments in Argentina, Nigeria, South Africa and the UK. Fact-checks cut belief in false claims by at least 0.59 points on a five-point scale, and most of the effect was still there two weeks later.

Agreement on verdicts does not settle the bigger complaint, which is about selection. In 2011 a University of Minnesota analysis found PolitiFact rated 39% of Republican statements “False” or worse, against 12% for Democrats; its author said the numbers could not tell who lies more from which statements were chosen. Nobody samples claims at random.

The critics need checking too. In my library, Ed Dowd’s book “Cause Unknown” tells readers that “the man who started Reuters Fact Check is also on the Board of Pfizer”. Part of that is documented: Pfizer’s own board page lists James C. Smith, a director since 2014, as chairman of the Thomson Reuters Foundation, and he was chief executive of Thomson Reuters until 2020. I found nothing showing that he “started” Reuters Fact Check. According to the journalist Paul Thacker, Reuters has said there is no conflict of interest. Claims about checkers can be checked the same way.

Crowd checking has its own record. A 2024 study in JAMA found that 97.5% of Community Notes on COVID-19 vaccine posts were entirely accurate. A 2025 University of Washington study found that reposts fell by 46% after a note was attached. Other research warns that notes often arrive after most of the sharing has already happened.

What it feels like from the inside #

If your post gets a label, it feels like being judged by someone you never chose, with your reach cut before any hearing. If you share a fact-check, it can feel like having the last word. Both feelings treat the label as a verdict instead of an argument. The article behind it is the evidence, and only the evidence deserves your trust or your doubt.

How to notice it #

• Read the check, not the label. What exact sentence was rated, and is it the claim you actually saw?
• Look at the date and any update notes. Ratings get revised, and the lab-leak case shows they can be withdrawn.
• Find the funders. IFCN signatories must publish them, so look for the funding page.
• Check the scale. “Missing context” and “partly false” are judgements about framing, not findings that something is untrue.
• Ask about selection. Who chose this claim to check, and which claims from the other side went unchecked?
• Look for a second checker. Two unrelated outlets agreeing counts for more than either one.

Frequently asked questions #

Who funds fact-checkers? #

Mostly grants and platforms: in 2025 grants averaged 46.2% of revenue, Meta 34.3% and TikTok 5.1%, according to the IFCN’s survey. Signatories must disclose their funders.

Did Meta end fact-checking? #

In the United States, yes, on 7 April 2025, replaced by Community Notes. Elsewhere Meta still works with more than 90 fact-checking organisations while it tests notes.

Are fact-checkers biased? #

Each is an outlet with editors, funders and choices. Major checkers agree closely on the same claims, but whether their choice of claims is balanced is unsettled. Judge each check by its evidence.

Do fact-checks actually change minds? #

On average, modestly. Experiments in four countries found lasting reductions in false beliefs, weaker when the correction runs against a reader’s politics.

Is Community Notes better than professional fact-checking? #

They do different things. Notes have been accurate on vaccine posts and cut reposts, but they are slower and far fewer. Professional checks reach scale but depend on paid contracts and editorial choices.

The lesson #

A fact-check is a piece of journalism with a verdict on top. The verdict is the least useful part of it; the sources underneath are what you can check yourself. Treat fact-checkers as you would any outlet: notice who pays them, read what they actually rated, see whether they correct themselves, and compare them with someone else. The record holds both documented errors and documented effects. Neither gives fact-checkers, or their critics, the right to do your judging for you.