Mr AI Decoder

Home / How we check

How we check a claim

Each episode takes one claim about AI and work that people are sharing, and checks it against the source it came from.

Five steps, every episode

  1. Pick a claim people are sharing. A headline, a stat or a forecast that is going around.
  2. Read the primary source. The report, study or dataset itself, and what it actually says.
  3. Do the maths. Every number traced to a source. Where we model something, the model is ours, kept simple and labelled.
  4. Give a verdict. One word, such as true, misleading or false, and who it holds for.
  5. Give you one thing to do, and a free file to run your own numbers.

Every source gets a grade

  • A

    Official statistics. Government data such as the US Bureau of Labor Statistics.

  • B

    Published study or disclosed data. A paper or dataset you can read in full.

  • C

    News report, announcement, survey or forecast. Useful, but someone’s estimate or summary.

  • D

    Our own assumption. Labelled as ours on screen and in the file.

Rules we keep

  • No number on screen without a source.
  • Our assumptions are labelled as ours.
  • Models are simplified, never forecasts.
  • Each episode says what it doesn’t show.
  • Mistakes get fixed: DM @mr.ai.decoder and we correct the page and note the change.

Chad and Claude

Chad brings the claim and Claude checks it. They are our animated characters. Mr AI Decoder is an independent page, not affiliated with Anthropic or with any organisation whose claims we check.