Operations · Engineered

Run a Research Brief Without Getting Confident Nonsense

Separate verified facts from inference before you make a decision on them.

Act as a research analyst. Accuracy matters more than completeness here.

RESEARCH QUESTION
[YOUR QUESTION — as specific as you can make it]
Why I need it / the decision it feeds: [DESIRED OUTCOME]
Scope: [GEOGRAPHY, INDUSTRY, TIME PERIOD]
How current it must be: [RECENCY REQUIREMENT]
Tool: [CHATGPT / CLAUDE / GEMINI / PERPLEXITY — note whether browsing is enabled]

METHOD
1. Restate the question and state explicitly what would count as a good answer.
2. Answer in three clearly separated tiers:
   TIER A — Verified: things you can cite to a source. Include the source and date. If browsing is not available to you, say so and put nothing in this tier that you cannot attribute confidently.
   TIER B — Reasoned inference: your conclusions and the reasoning chain behind each.
   TIER C — Unknown: what would change the answer but you cannot determine.
3. Where sources disagree, present both positions and say what the disagreement hinges on.
4. Rate your confidence per tier-A claim as high, medium or low, and say why.

HARD RULES
- Never fabricate a source, statistic, study, URL, quote or date. If you cannot verify it, it belongs in Tier B or C.
- Do not present a recalled figure as a current one. Flag anything whose value changes over time.
- Do not smooth over uncertainty to sound helpful.

CLOSE WITH
- The three things I should verify myself before acting, and where to look.
- The single assumption in this brief that, if wrong, would most change the decision.

How to use this prompt

  1. Copy the whole template — the structure matters as much as the words.
  2. Replace every bracketed placeholder with your real details. Vague inputs produce vague output.
  3. Paste it into ChatGPT, Claude or Gemini and let it finish before you judge it.
  4. Give one round of feedback: name what to keep, what to cut, and what to change, then ask for the revised version.

Why it works

This prompt gives the model a role, the background it cannot guess, the outcome you actually want, the limits it must respect and the exact format for the answer. Those five pieces are the difference between a generic paragraph and something you can use. The words behind them are explained in the prompting glossary.

A template does not know your business. The Prompt Engineer does the asking for you.