AI Guides › Prompting
Prompting For A Second Opinion On Your Own Idea, Honestly
By Nigel Guy · 2 min read
The usual way to ask for a second opinion is to describe the idea and also,
in the same breath, describe why it's good — "I've been thinking about
doing X, and I think it makes a lot of sense because…" Framed that way, what
comes back tends to agree, because you've already supplied the case for it
and the most natural continuation of an enthusiastic pitch is more
enthusiasm.
The rule: state the idea in neutral terms, without your case for it,
before you ask anyone — human or model — what they think of it.
The mechanism: strip, then stress-test
- Write the idea as a plain description of what it is, not why it's
good. "We're considering charging a flat monthly fee instead of
per-seat pricing" — not "I think a flat fee is obviously simpler and
fairer, so we're considering it."
- Remove your own enthusiasm markers before sending — "I think this is
great," "this seems like the obvious move," "everyone I've mentioned it
to likes it." Each of those is a leading signal, the same way a leading
question is.
- Ask explicitly for the strongest case against it. Not "what do you
think" — a direct instruction to argue the other side, so you get an
actual test rather than a polite gloss.
- Ask separately for the strongest case for it. Deliberately asking
for both sides gets you real material either way, rather than only the
agreement a neutral-sounding "what do you think" tends to produce by
default.
- Weigh the two yourself. This exercise is meant to hand you material
to reason with, not a verdict — treat the result as input, not an answer.
What to skip
Skip the instruction "be brutally honest" as if the phrase itself does the
work — it's the same kind of flavour text as "as an expert in X," and does
little on its own. The neutral, stripped-down description in steps 1 and 2
is what's actually doing the work. Skip this whole approach for early-stage
brainstorming, where an expansive, generative reaction is what you actually
want — stress-testing is a later step, not the first one.
Guardrails
- The model has no independent stake in your decision and no ability to
tell you it's actually the strongest objection available — it can
generate a plausible one, which isn't the same guarantee.
- This doesn't replace an actual person who disagrees with you, especially
someone with real context on the decision. Use it to surface angles
you hadn't considered, not to arbitrate whether the idea is good.
- For anything with real consequences riding on the answer, treat this as
one input among several, not the deciding one.
All 751 AI guides · JulieMango plans from £17/mo