AI Guides › Prompting
Asking For The Method, Not Just The Answer
By Nigel Guy · 2 min read
Most people ask a question, get an answer, and check whether the answer looks
right. That's a weak check, because a confidently wrong answer and a
confidently right one can read identically on the page — same tone, same
tidy formatting, same air of certainty. You're not actually verifying
anything; you're checking that it sounds plausible, which any fluent wrong
answer will also do.
The rule: ask for the method alongside the answer — the reasoning is easier
to fact-check than the conclusion, and bad reasoning is where wrong answers
come from.
The mechanism
- Ask for the steps, not just the result. "What's the answer, and walk
me through how you got there" rather than "what's the answer." For
anything involving a calculation, a comparison, or a chain of if-this-
then-that logic, the method is where errors hide.
- Read the method before the conclusion. Skim the reasoning first. If a
step doesn't follow from the one before it, the conclusion is already
suspect regardless of how tidy it looks.
- Check the weakest link, not every link. You don't need to verify every
step equally — find the one step the whole conclusion actually depends on
and check that one properly.
- Ask "what would change this answer?" if the method still looks sound
but you're not confident. A method that can't name what would overturn its
own conclusion is often reciting a pattern rather than reasoning through
the specific case.
Where this matters most
This is worth the extra step on anything with a calculation, a ranking, a
recommendation between options, or a claim about cause and effect. It's not
worth it for a request with no reasoning to check — a rewrite, a rephrase, a
list of ideas. There's no method to inspect in "make this sound friendlier";
the output either reads right or it doesn't.
What to skip
Skip asking for the method on subjective or generative tasks where there's no
underlying logic to audit. And skip accepting a method that's just the
conclusion restated in more words with "because" in front of it — that's not
reasoning, it's the same claim wearing a longer sentence.
Guardrails
- A plausible-sounding method isn't a verified one. Fluent step-by-step
reasoning can still contain a wrong step; read it, don't just note its
presence.
- This catches logic errors, not factual ones. A method can be internally
consistent and still rest on a fact that's simply wrong — verify the
premises too, especially anything time-sensitive.
- For anything with real stakes — money, health, legal standing — checking
the method is a floor, not a substitute for a second, qualified opinion.
All 751 AI guides · JulieMango plans from £17/mo