The most common prompt mistake isn't vagueness. It's loading.
A vague brief produces vague answers, and you notice immediately. A loaded brief produces confident, specific, useful-sounding answers that all agree with you — and you notice nothing at all. That is the failure worth engineering against.
Three components
A well-framed brief has three parts. You won't need all three every time, but you should know which one you're leaving out and why.
- Context — who you are, what situation you're in, what has already been established or decided. Two to four sentences. Be specific; this is what separates a useful answer from a generic one.
- Question — stated precisely. One question. Specific enough that a clear answer is possible, and narrow enough that three platforms will understand it the same way.
- Output format — what kind of response you need: a recommendation with reasoning, a list of options, a structured analysis, a yes/no with explanation. Unspecified format is the main reason three answers come back incomparable.
Add one more line that most people omit: tell each platform it is being asked independently and that you will compare responses. It costs nothing and it measurably changes the register of what comes back.
Neutral phrasing, concretely
Neutral phrasing means structuring the question so that any answer is genuinely possible. The test is not whether the question sounds fair. It is whether a platform could disagree with you without seeming to have misread the prompt.
LOADED
- "I'm thinking of moving forward with Option A — can you confirm this is the right choice?"The answer is in the question. You will get three confirmations and learn nothing.
- "Why is this approach better than the alternative?"Presupposes it is better. The platform's job is now to justify, not evaluate.
- "Is there anything wrong with my plan?"Invites a token objection followed by reassurance. Reads as a request for permission.
NEUTRAL
- "Evaluate Option A and Option B. Identify the strengths and risks of each and recommend one, with reasoning."Both outcomes are available. The recommendation now carries information.
- "Compare these two approaches against [stated criteria]. Which better satisfies them, and where does the weaker one still win?"Names the criteria, so three answers become comparable.
- "Identify the weakest assumptions in this plan and what would have to be true for it to fail."Asks for the failure case directly rather than hoping it volunteers.
If you find yourself rewriting a brief because it sounds too negative, stop. That instinct is the sycophancy trap operating from the inside — you softening the question so the answer will be pleasant. The discomfort you feel writing the harsh version is the signal that the harsh version is the one worth asking.
CONTEXT: [Who you are, what situation you are in, and what has already been established or decided. 2–4 sentences. Be specific.] QUESTION: [State your question precisely. One question. Specific enough that a clear answer is possible.] OUTPUT FORMAT: [What kind of response you need: a recommendation with reasoning, a list of options, a structured analysis, a yes/no with explanation.] NOTE: Please approach this independently. I am asking the same question of several platforms and will compare responses before synthesizing. Where you are uncertain, say so and say why rather than resolving the uncertainty for me.
Documents
The mechanics don't change: identical inputs across all three. Upload the same file, with the same question, to each platform independently.
Upload — do not summarize on their behalf. The moment you paste your own summary instead of the source, every platform is now reasoning about your reading of the document rather than the document. Whatever you missed, all three will miss identically, and their agreement will look like corroboration. This is one of the easiest ways to manufacture false consensus without noticing.
Ask specific questions rather than "summarize this." A general summary request produces three overlapping summaries that are tedious to compare. Three answers to "identify the clauses that create risk" are directly comparable, and the differences are immediately meaningful.
DOCUMENT CONTEXT: [What this document is and why you are analyzing it.] TASK: Please analyze the attached document and: · [Specific question 1 — e.g. identify clauses that could create risk] · [Specific question 2 — e.g. flag anything that contradicts X] · [Specific question 3 — e.g. summarize the key obligations in plain language] Do not summarize the document generally. Focus on the questions above. Note anything that requires verification against current sources, and quote the passage you are relying on for each finding.
What to keep constant, what may vary
The brief must be identical. The platform settings around it should also be as close as you can make them — a reasoning mode enabled in one and not another is a real difference in the input, even though the text matched. If you turn on deep research or extended thinking for one, turn it on for all three or note that you didn't.
What may legitimately vary is the follow-up. Once all three initial responses are in and you've read them, targeted follow-ups to individual platforms are fine and often necessary. The independence requirement applies to the first pass, which is the pass that produces the fix.
"Identical inputs. Independent outputs. That is the briefing."
If you rephrase the question for each member, you are not triangulating. You are conducting separate interviews.