AI Guides › Judgement & Guardrails
The Honest Case For Distrusting Your Own First Reaction To A Good Answer
By Nigel Guy · 2 min read
There's a specific feeling that shows up when an AI tool hands you an
answer that resolves something you were stuck on, or tells you something
you were already hoping was true. It feels like recognition — "yes, that's
it." That feeling is real, but it isn't measuring what it seems to be
measuring. It's mostly measuring relief, and relief has no opinion on
whether the answer is actually correct.
The rule: your immediate positive reaction to an AI answer is a signal
about how much you wanted that answer to be right, not a measurement of
whether it is — treat the two as separate and check the second one
separately.
The mechanism: the second read
- Notice the feeling itself, specifically, when it shows up. A small
flush of relief, the urge to stop looking, the sense of "finally" — name
it as what it is rather than folding it silently into your assessment of
the answer.
- Reread the answer a second time looking specifically for its weakest
claim, not for confirmation of the parts you already liked. The first
read finds what you were hoping for; the second read has to go looking
for what you weren't.
- Ask whether you'd feel the same confidence in this answer if a stranger
had said it, with the same words, and no context about how badly you
wanted it to be true. If the answer only feels solid because of how
relieved you were to hear it, that's the tell.
- Treat "this conveniently confirms what I wanted to hear" as a reason
for more scrutiny, not less. The answers that most deserve a second
look are exactly the ones that feel like they need it least.
What to skip
Skip applying this to every trivial exchange — most answers don't carry
enough emotional charge to distort your judgement, and second-guessing all
of them is exhausting for no benefit. This is specifically for the moments
where the answer landed with a noticeable feeling of relief, resolution, or
validation attached — that's the signal that the check is worth running.
Guardrails
- This isn't a claim that your instincts are generally unreliable — it's
specifically about the gap between an answer feeling right and an answer
being right, which is a narrower and more common failure than general
distrust of your own judgement.
- The check works better paired with an actual verification step — tracing
the claim to a source, redoing the arithmetic — than as a feeling-based
exercise on its own. Noticing the relief is the trigger, not the
substitute, for checking properly.
- This applies just as much to your own writing and conclusions as to an
AI's — the same relief-as-verification trap catches people reviewing
their own first draft of an idea, with or without a model involved.
All 751 AI guides · JulieMango plans from £17/mo