AI Guides › Judgement & Guardrails
The One Habit That Catches Most Bad AI Advice Before It Costs You
By Nigel Guy · 2 min read
Most guidance on using AI carefully turns into a long list — check this,
verify that, don't trust the other. Long lists don't survive contact with a
busy day; they get skipped wholesale the first time someone's in a hurry.
What actually holds up under time pressure is a single habit, applied
consistently, rather than a checklist people quietly stop running.
The rule: before acting on anything an AI tool told you, ask one
question — "what would tell me this is wrong?" — and if you can't answer
it, that's the signal to slow down, not a formality to skip.
The mechanism: the falsification question
- Ask it before you act, not after something goes wrong. The question
is a gate, not a postmortem tool. Build it into the moment right before
you send, submit, or rely on something.
- Demand a real answer, not a shrug. "I'd notice if it seemed off" is
not an answer — it's the confidence problem restated. A real answer
names something concrete: a source you'd check, a number you'd
recompute, a person who'd know.
- If you can name a real check, do it — right then, not later. The
question is only useful if answering it actually triggers the check,
rather than becoming its own substitute for one.
- If you can't name a real check, treat that as information. It means
you're about to act on something you have no way to catch if it's
wrong. That's not necessarily a reason to stop, but it is a reason to
know you're taking that risk knowingly rather than by accident.
- Let the answer scale with the stakes. For something trivial, "I'd
probably notice" might be a fine answer. For something that costs real
money or trust to get wrong, it isn't, and the habit is partly about
noticing which situation you're actually in.
What to skip
Skip running this as a formal, written exercise every time — for most
day-to-day use it should take seconds, a genuine but quick mental check.
Save the fuller written version — an actual source, an actual independent
check — for the moments where the honest answer to the question is "I
don't currently have one."
Guardrails
- This one habit is deliberately simple, and simple things get skipped too
if they're not actually practised. The value comes entirely from asking
it every time, not from how sophisticated the question sounds.
- It catches the errors you'd have no way to catch otherwise — it doesn't
replace the deeper checks (source-checking, independent verification)
this library covers elsewhere. Think of it as the trigger for those, not
a substitute for them.
- Answering the question honestly requires a moment of real self-scrutiny,
not just going through the motions. If you notice yourself always
answering "yes, I'd catch it" without specifics, that's worth examining
on its own.
All 751 AI guides · JulieMango plans from £17/mo