The prompt
Score 10 of 10, Excellent
Use it when You need a scoped research request with the questions, preferred sources and output format before you run it.
- {{research_topic}}
- {{decision}}
- {{what_i_know}}
- {{scope_limits}}
- {{preferred_sources}}
- {{sources_to_avoid}}
- {{output_wanted}}
Full prompt and why it works
Prompt
What I want to research: {{research_topic}} The decision or piece of work this is for: {{decision}} What I already know or have read: {{what_i_know}} Limits on the research (countries, time period, industries, company size): {{scope_limits}} Sources I trust or want included: {{preferred_sources}} Sources to leave out or treat with caution: {{sources_to_avoid}} What I want back: {{output_wanted}} Task: Write a research request I can paste into a deep research tool, such as the research modes in ChatGPT, Gemini, Claude or Perplexity. These tools search the web for several minutes and write a long report, so a vague request produces a long, vague report. The request should keep the tool on the questions that matter for my decision and make its sources easy to check. Format the request with these labeled parts: Objective: one or two sentences on what I need to decide or produce, and who will read the result. Key questions: 3 to 7 specific questions, each answerable with evidence, most important first. Turn broad wishes into questions with a clear answer, for example "understand the market" into "Which regulations most affect this product in this country, and when does each take effect?" Scope: what is in and out, the geography, and the time period, including how recent sources must be. Sources: what to prefer (primary sources such as official statistics, regulators, company filings, peer-reviewed studies and original data), what to treat with caution (vendor marketing, sponsored content, undated pages) and what to exclude. Evidence rules: give a link for every factual claim and the publication date of each source, separate reported facts from estimates and forecasts, show conflicting figures side by side instead of averaging them, and write "not found" rather than filling a gap. Output: the format, length and sections I asked for, ending with a section on what the tool could not find and how confident it is in each main finding. Requirements: - Keep the request under 400 words, written as direct instructions to the research tool. - Use only what I have told you about my situation. Do not add facts, names or numbers of your own. - Do not answer the research questions yourself. Your job here is the request, not the research. A good request is one where a colleague reading it would know exactly what the report must answer and how to judge whether it did. After the request, list up to three things I should decide or check before running it, such as a question that is still too broad or a time period I have not set. If something essential is missing, do not guess silently: put your best assumption in square brackets inside the request, for example [assumed: sources from the last 3 years], so I can change it.
Variables
- {{research_topic}}
- The subject, in a sentence or two.
- {{decision}}
- What you will use the result for, for example "whether to enter a new regional market next year".
- {{what_i_know}}
- What you already know, believe or have read. Write "nothing yet" if you are starting fresh.
- {{scope_limits}}
- For example "EU only, sources from 2023 onward, companies under 500 staff". Write "none" if it is open.
- {{preferred_sources}}
- For example "government statistics and trade association reports". Write "none" if you have no preference.
- {{sources_to_avoid}}
- For example "vendor blogs". Write "none" if there are none.
- {{output_wanted}}
- The format and length you want back, for example "a two-page memo with a table of options".
Why it works
- What good looks like. Evidence rules (a link and date for every claim, facts kept apart from forecasts, conflicts shown side by side) make the long report the tool returns something you can check.
- Length. Broad wishes become 3 to 7 answerable questions with a set time period and geography, which keeps a deep research run from producing a long, generic survey.
- If unsure. Missing details become bracketed assumptions you can change before you run it, rather than silent guesses buried in the request.