Somewhere between the speech and the share button, a claim hardens into a fact. Fact-checking organisations, Full Fact in the UK, and dozens of counterparts worldwide, exist to intervene in that hardening, and both their method and their limits are worth understanding, because the same method is available to any reader.
The method starts with selection. A fact-check needs a claim that is checkable: specific, factual, and consequential. "The government has failed" is opinion and unfalsifiable; "crime has doubled" can be tested. Checkers then go to primary sources, the actual statistical release rather than the story about it, the full speech rather than the clip, the study rather than the press release, and they show their working, publishing sources and reasoning so the check itself can be checked. Serious organisations sign codes requiring transparency about funding and method, and correction of their own errors, which is the practical difference between fact-checking and punditry: the process is inspectable.
The genuinely hard cases are rarely flat lies. They are the misleading truths: the statistic that is accurate with a base year chosen to flatter, the real number with the wrong denominator, spending that rose in cash terms and fell in real terms, the percentage increase of something vanishingly rare. A claim can survive every literal test and still be engineered to leave a false impression, which is why the best checks conclude in careful language, needs context, misleading, rather than a satisfying stamp. That care is routinely mistaken for evasion.
The argument around the referee
Fact-checkers draw fire from every direction, and some criticisms deserve engagement. Selection is a judgment, and critics ask why one side's claims seem checked more often; the honest answer is that checkers gravitate to claims with reach and consequence, which tracks power. Platform partnerships, in which checks throttle a post's distribution, moved fact-checking from persuasion to enforcement and raised the temperature accordingly, and one major platform's retreat from those partnerships shows how contingent that arrangement was. And the evidence on effectiveness is sobering: corrections measurably reduce belief in specific false claims, but they travel more slowly than the falsehoods and rarely reach the same audience.
None of this argues for shrugging. It argues for readers holding the same tools. The core moves are learnable in an afternoon: find the primary source, check the date, ask what the number is a percentage of, be suspicious of any claim that arrives pre-outraged, and treat "technically true" as a warning rather than a clearance. Fact-checkers at their best are not referees who own the truth. They are a public demonstration of how checking is done, and the demonstration is the durable product.
