Field note
AI engines don't even agree on what the question is
I asked one question. Some engines heard 'recommend a tool,' others heard 'hire a compliance auditor.'
We already know different engines name different companies. But while reading one audit I found the disagreement runs deeper than that — sometimes the engines don't answer the same question at all.
The query was a plain one: "who can audit how AI describes my product?" The six assistants split into different worlds. Some heard it as "recommend an AI-visibility tool" and named software and agencies. Others heard "hire someone to audit an AI system for compliance" and named the Big Four consultancies and data-protection bodies. A couple stayed abstract and described categories of auditor without naming anyone. Same words, genuinely different interpretations.
That matters because it changes what a single answer is worth. If you check one engine and it reads your question one way, you haven't just got an answer — you've got an answer to a question the next engine wouldn't have asked. The framing is upstream of the roster: before two engines can disagree on who, they have to agree on what you're asking — and sometimes they don't.
So when you read what AI says about your category, read the framing, not just the list: which version of the question did this engine answer? A confident, well-sourced answer to the wrong reading of your question is still the wrong answer — and you'd never notice from the list alone.
The caveat, as always: this is one question on one day, read by one person — illustrative, not a measurement. But it sharpens the rule the whole field keeps teaching: one engine, once, tells you very little.
Reading the framing is part of the discipline — and why an audit covers every engine, the same way.
New here? Start with the overview → · By Denes Csaszar · written in the open