AI Guides › Prompting
The One Follow-Up Question That Improves Almost Any Answer
By Nigel Guy · 2 min read
Most people treat the first answer as a finished object: either it's good
enough to use, or it isn't and you rewrite the prompt from scratch. Neither
response makes use of the fact that the model made real choices while
producing that answer — what to include, what to leave out, how deep to go
— and those choices are recoverable just by asking about them.
The rule: ask what was left out and why, before you ask for anything
else — a single targeted follow-up does more for the answer's quality than
a second attempt at the original prompt.
The mechanism: the omission question
- Get the first answer, as normal.
- Ask directly: "what did you choose not to include here, and why?"
Not "is this complete?" or "anything else?" — a specific instruction to
report the choices already made, not to guess at new content.
- Read the list of omissions on its own terms. Some will be genuinely
irrelevant to your case; some will be things the model correctly guessed
you didn't need. That's fine — the point isn't that every omission is a
flaw.
- Decide which flagged omissions actually matter for what you're
using this for. The model doesn't know your full context, so it can't
make this call — it can only tell you what it left out.
- Ask for a revision addressing only the omissions you've decided
matter. Naming them specifically gets a tighter fix than resending the
whole original prompt.
What to skip
Skip the vaguer version of this question — "is there anything I'm missing?"
tends to produce padding rather than the actual list of deliberate choices
behind the first draft. Skip this step entirely for simple factual lookups,
where completeness isn't really ambiguous and there's nothing meaningful to
have left out.
Guardrails
- This surfaces the choices behind the answer you got, not gaps in the
model's knowledge. It won't tell you about a fact it doesn't know, only
about ground it decided not to cover.
- Treat the list of omissions as candidates, not a verdict — deciding which
ones matter is still your judgement call, not something the follow-up
question settles for you.
- This is one useful question, not a substitute for actually reading the
answer carefully yourself — it improves what you're reviewing, it doesn't
review it for you.
All 751 AI guides · JulieMango plans from £17/mo