Everything I check,
and what turned out to be true.

AI in plain language, for your actual life. Every claim a chatbot made on this channel, checked against the actual source — dated, linked, and searchable. When someone sends you a screenshot at 11pm, start here.

The archive opens with episode 01.

The archive

Nothing checked yet.

Episode 01 is being made. When it airs, every claim in it lands here — dated, sourced, and linkable. One claim, one page, so you can send someone a single link instead of a whole video.


It isn't lying, and it isn't broken.

It's answering from what it read, and what it read has a date on it. Every episode, I ask a chatbot about something in the news at normal human speed — false starts included — and then I check what it told me against the real source. This is where the checking lives.

How to talk
to a computer
AI in plain language, for your actual life

One prompt a week, and the claim it caught.

The bad version, the fixed version, and the thing the fix actually caught. The prompt is never invented for the newsletter — it's always the one that turned up a real answer worth checking.

The prompt that caught it.

One prompt: the bad version, the fixed version, and the thing the fix actually caught. Once a week at most.

The first issue goes out with episode 01. Every prompt in it will be one that caught a real claim — nothing invented for the newsletter.

Sign-up opens with episode 01.