AI Guides › Playbooks
By Nigel Guy · 6 min read
Most people switch Claude's Research mode on, type a one-line question such as "research our competitors", and wait. A few minutes later they get a long, well-cited report that looks authoritative and answers a question nobody quite asked. Research is built to go wide, so a vague brief produces a wide, vague report.
The rule: write the decision first, then the brief. Research is for gathering sourced material toward a decision you have named; the checking and the judgement stay with you.
Anthropic's help centre describes Research as work that takes five or more tool calls over one to three minutes, building a report from several sources across the web and any integrations you have connected. It runs multiple searches that build on each other, and the report carries citations so you can check claims. Three details matter for planning:
At time of writing, Pro is listed at $20 a month ($17 a month billed annually), about £15 at time of writing; Max starts from $100 a month, about £75 at time of writing. Check the £ price at checkout, as it moves with the exchange rate and any VAT.
To turn it on, click the "+" button in the chat box and choose "Research"; a blue indicator appears. If Claude answers normally instead, say "please use the research tool to..." at the start of your message. Menu labels change, so if yours differs, check the help centre page in the sources.
Fill in six lines before you press send. It takes two minutes and is the difference between a report and a useful report.
| Line | What goes in it | Why it matters |
|---|---|---|
| Decision | What you will do differently depending on the answer | Stops the report becoming a general tour |
| Scope | Companies, regions, product tiers, what is out | Research otherwise widens the net on its own |
| Date window | "Only sources from the last 12 months" | Old pages rank well and look current |
| Source rules | Prefer official sites, filings, pricing pages; flag forum posts | Gives you a fast way to weight claims |
| Output | Table, ranked list, or memo of a set length | Saves a reformatting round |
| Doubt flag | "Mark anything you could not confirm" | Makes gaps visible rather than smoothed over |
Here is the card as a reusable prompt.
You are a careful business research analyst working for [YOUR_ROLE] at [COMPANY_AND_SECTOR].
Decision this research supports: [DECISION_I_AM_MAKING].
Topic: [TOPIC].
Scope: include [IN_SCOPE]; exclude [OUT_OF_SCOPE]. Region: [REGION].
Date window: only use sources from [DATE_WINDOW]; if a key fact has no source in that window, say so.
Source rules: prefer official company pages, regulators, filings and published pricing. Treat forums, anonymous reviews and press-release rewrites as weak evidence and label them.
Steps:
1. Restate the decision and scope in two lines and list any assumption you are making.
2. Research the topic, following leads that bear on the decision.
3. Present findings as [OUTPUT_FORMAT], with a citation beside every factual claim.
4. Add a section called "Not confirmed" listing claims with only one source or none.
5. End with what each plausible finding would mean for the decision, in no more than [WORD_LIMIT] words.
If any bracketed input is missing or unclear, ask me before you start rather than guessing.
Before you answer, check that every number has a source and a date, and that nothing in the report goes beyond what the sources say.
Fill in the square-bracket fields; leave nothing blank, as blanks invite guessing.
Each job reuses the card. Only the Decision and Output lines change.
For job 4, use this variant.
You are a fact-checker for [PUBLICATION_OR_TEAM]. Below is a draft intended for [AUDIENCE].
Draft:
[PASTE_DRAFT]
Task: extract every checkable factual claim (names, figures, dates, product features, prices). For each one, return: the claim; a verdict of supported, contradicted, or unverifiable; the best source with its date; and a suggested correction if needed.
Use sources from [DATE_WINDOW]. Prefer primary sources. Do not rewrite my prose or judge opinions.
If a claim is ambiguous, ask me which meaning I intend instead of choosing one.
Self-check: confirm that no verdict rests on a source you have not actually opened.
Fill in the team, audience, draft and date window.
Keep the Research Brief Card as a saved note, or in a Claude Project's instructions, so you only change the top lines. Run the report, then open three sources by hand: the one behind the claim you care most about, the one that surprises you, and one at random. If the three hold up, trust the rest a little more; if one does not, treat the report as a draft.