How Claim Brief checks claims

Claim Brief checks where claims about AI come from and whether they hold up. This page is the whole method. Nothing is held back, and you're welcome to use it yourself.

Which claims get checked

Claims come from what people send me, comments and messages, and what's circulating in search and on social media. A claim gets checked only if it passes all five of these questions:

  1. Is it actually landing on people? Shared, forwarded, a worried comment section, or someone mentioned it.
  2. Is it checkable? A specific number, study, quote, or event.
  3. Can it be checked with sources that can be shown on screen? The post isn't the source. The check starts with who said it first.
  4. Would the verdict help someone decide something? About a job, money, family, a business, or a vote. A verdict may inform a vote, but the check never tells anyone how to vote.
  5. Is the claim the target? Never anyone who repeated or believed it.

Two kinds of claims

Claims about now — a study, a number, a quote, a layoff, a product — can be verified. The check finds the source and shows how close the claim comes to it.

Claims about the future — "by 2028," "half of all jobs" — can't be verified, and Claim Brief doesn't pretend they can. Instead, the check looks at who is saying it, whether they benefit from you believing it, their track record, what would have to be true for it to happen, and what the evidence so far shows. A forecast verdict describes what a prediction is built on; it is not a finding that the prediction will or won't come true.

Sometimes the honest answer is that nobody knows yet. When that's the case, the check says so rather than forcing a verdict the evidence doesn't support.

How a check works

Each claim check names the source, gives its date, shows the gap between the claim and the evidence, and ends in one of five verdicts. In practice:

  1. Find who said it first. The post, headline, or video where you saw it is usually a copy. The check traces it back to the original study, report, quote, or event.
  2. Name the source, its date, and who benefits. Who published it, when, how the information was gathered, and whether the publisher gains from the claim being believed.
  3. Compare the claim to the source. The claim as it circulates, side by side with what the source says, and the gap between them, if there is one.
  4. Give the verdict. One of the five below.
  5. Say what it means for you. What the verdict changes, or doesn't, for someone deciding what to do next.

The evidence is shown on screen, and every video and page links to its original sources so you can check them yourself. A link to a source is not an endorsement of it.

Every figure and quote in a check is read from the original source. When AI tools help with research and production, they're never the source for a number or a quote.

The five verdicts

Every check ends in one of these, and only these:

  • Supported. The evidence backs up the claim as it's stated.
  • Overstated. The claim starts from real evidence and stretches it: a bigger number, a wider group, a stronger word, or more certainty than the source has. The check names the stretch.
  • Unsupported. The evidence doesn't back the claim, or the claim can't be traced to any evidence.
  • True but not what you think. The claim is accurate as stated, but the evidence measures something different from what most people take it to mean.
  • Forecast — here's what it's built on. A prediction. It can't be verified yet, so the check lays out what it rests on.

A few things hold for every verdict:

  • A verdict is Claim Brief's reasoned opinion, based on the sources shown. It is not a statement of objective fact about any person, company, or product.
  • A verdict is about the claim, not about anyone who repeated or believed it.
  • Evidence changes. A verdict reflects the sources available when the video or page was published.

Political and election claims

Claim Brief takes no political position and never tells you how to vote. A factual claim gets a verdict no matter who said it. When two sides are making competing predictions, both get the same treatment: what each is predicting, what each has to gain, and what you control either way. A verdict may inform your vote, but the check never tells you how to vote.

Disclosures and independence

Claim Brief uses Claude, made by Anthropic, in its research and production. When a claim involves Anthropic, Claude, another tool used on the channel, or a company Claim Brief has a financial stake in, the video says so. No company can pay to influence a verdict. The full list is on the Disclosures page.

Corrections

If you think a check got something wrong, email hello@claimbrief.com with the source. If there's an error, a correction is posted at the top of the video's description and in its pinned comment, with the date and what changed. If the error changes the verdict, the correction says so. Claim Brief doesn't delete published pieces to hide a mistake.

Claims and stories you send

If you send me a claim, I may check it in a video. I won't name you, or tell your story, without your permission, and only at the level of detail you confirm.

What this isn't

Everything here is for general information and education. Nothing is professional advice of any kind, including legal, financial, investment, tax, medical, career, or employment advice. For a decision that matters to you, talk to a qualified professional who knows your situation. The full terms are in the Terms of Use.

Use it yourself

The method is public, and you're welcome to use it yourself. Find who said it first, check the date, look for the gap, and ask who benefits.

Claim Brief · @claimbrief — where AI claims come from, and whether they hold up. Comment or DM me a claim. It might be the next one I check. Want it set up live, with a safety check for your situation? See the live sessions.