Deep research that starts from the decision

The Research Brief

A fill-in brief that turns any deep research run answer-shaped: what you are deciding, what would change your mind, and what counts as evidence — so the output is a decision, not a survey.

Deep research modes are the most underused button in the category, and the reason is that people hand them a topic when they should hand them a decision. This brief makes the run answer-shaped: what you’re deciding, what would change your mind, and what counts as evidence.

Run it once per question, in any deep research mode. Deciding between tools with what you found? Run it through Build, Buy, or Skip?. On why the spot-check step is not optional, see hallucination.

The Research Brief

Prompt 
Run deep research on this. Structure the whole job around the decision
it serves, not the topic.

THE DECISION: I am deciding [the actual decision — e.g., "whether to
build an intake automation," "which CRM to move to," "whether this
market is worth entering"].
WHAT I CURRENTLY BELIEVE: [one or two sentences — so you can confirm
or challenge it, not flatter it].
DEADLINE AND STAKES: [when I decide, and what it costs to be wrong].

RULES FOR THE RESEARCH:
1. Answer first. Open with your best current answer to the decision
   in three sentences, then show the work.
2. Source quality: primary sources over aggregators. Date-stamp every
   load-bearing claim. Anything older than [12 months] gets flagged
   as possibly stale.
3. Separate three kinds of claims and label them: independently
   verified, reported-but-single-source, and vendor/marketing claims.
   Never present the third kind as the first.
4. Where credible sources disagree, show the disagreement. Do not
   average it away.
5. Tell me what would change the answer: the two or three facts that,
   if different, flip the call.
6. End with the five most load-bearing facts in your answer, each
   with its source, so I can spot-check them in ten minutes.

Length: as long as the evidence needs, no longer.

Why it works The decision framing is the whole trick — research briefed on a topic returns a survey, research briefed on a decision returns an answer. Stating what you currently believe inoculates against the research politely agreeing with you. The three-kinds-of-claims rule exists because deep research modes cheerfully cite press releases at the same volume as peer review, and the label is the fix. And the spot-check list is the verification budget made real: ten minutes of your checking is what converts "the AI said" into "we verified."

Then what Deciding between tools with what you found → run it through Build / Buy / Skip.

Frequently asked questions

How do I get better results from deep research modes?

Brief the run on the decision it serves rather than the topic. State the decision, what you currently believe, and the deadline and stakes — then require an answer first, date-stamped sources, labelled claim types, and a short list of facts you can spot-check yourself.

Which AI deep research mode does this work with?

Any of them — Claude, ChatGPT, or Gemini. The brief is model-agnostic because it constrains the shape of the output, not the tool that produces it.

← All prompt packs