How We Rate Claims

Most AI news comes with a claim attached: a model is the best, a product saves hours, a deal changes everything. Each story we publish carries a verdict on its main claim, so you can see at a glance how well it holds up.

The verdicts

Supported The main sources back up what was claimed. Supported means the evidence matches the claim as stated, not that the product or decision is a good idea.

Promising but unverified There is evidence, but it comes only from the people behind the claim, such as a company’s own benchmark or a lab’s own paper with no outside replication yet.

Needs follow-up Important details lack confirmation or depend on things that have not happened yet: a deal that still needs approval, a feature that has not shipped, a number with no source.

Did not hold up The evidence goes against the claim, or the claim is much bigger than what the sources show.

News A plain report with no claim to test, such as a launch date, a leadership change or a policy being published.

How we decide

  • We judge the main claim in the story, usually the one in the headline.
  • We weigh primary evidence first (see How We Source). A company’s announcement shows what it says. Independent tests, filings and replications show whether it holds.
  • Verdicts are about evidence, not opinion. A story can be critical of a product and still rate its claim Supported.

Verdicts can change

New evidence can move a verdict either way. When that happens we update the story, add a dated note, and update the matching Hype Ledger entry, which records when each claim was last checked.

Disagree with a verdict?

Send us the evidence through the Contact page. If it changes the picture, we will update the verdict and log it on our Corrections page.