AI Guides › Prompting

The Difference Between A Leading Question And A Fair One

By Nigel Guy · 2 min read

It's easy to write a prompt that already contains the answer you're hoping for — "isn't a subscription model obviously the better fit here?" — and then read the model's agreement as if it were independent confirmation. It usually isn't. A prompt that states your preference before asking for judgement tends to get that preference reflected back, dressed up as analysis.

The mechanism: the leading question audit

Before sending a prompt you actually want a genuine judgement on, check it against these:

1. Does it contain a value word before the judgement is made — "clearly," "obviously," "the better option"? If so, the conclusion is already in the question.
2. Does it name only one option and ask for a defence of it, rather than naming all the real options neutrally and asking how they compare?
3. Does it reveal a decision you've already made — "we've gone with X, help me justify it" — while nominally asking for an assessment?
4. Does it ask "why is X good" instead of "is X good, and how does it compare to the alternatives"? The first presumes the answer; the second actually leaves room for "it isn't."

If any of these are true, rewrite before sending:

• Leading: "Isn't a subscription model obviously better here?"
• Fair: "Compare a subscription model against a one-off purchase model for this product — what favours each one?"

What to skip

Skip rewriting every casual question through this filter — if you already know you want a defence written for a decision that's been made, and you're asking for the case in support of it rather than a test of it, leading phrasing is the right tool, not a mistake. Skip assuming neutral wording alone removes your influence entirely — the model can often infer your leaning from other context in the conversation even when the question itself is phrased fairly.

Guardrails

• Fairer phrasing improves the odds of a genuine comparison. It doesn't guarantee an unbiased one — the model has its own defaults and tendencies independent of how you ask.
• This is a tool for judgement calls and open comparisons, not for factual questions that have a settled answer — there's no "leading" version of asking what a number is.
• For a decision that actually matters, one fairly phrased prompt is a useful check but not a complete one — a second, independently worded prompt, or a second opinion from an actual person, catches more than rephrasing alone.

All 751 AI guides · JulieMango plans from £17/mo