The Research Prompt Stack

Asking an AI “tell me about X” gets you a confident summary of the obvious. Research is different: you need the landscape, the primary sources, the disagreements and the parts nobody knows yet. This framework breaks research into six prompts that build on each other in one conversation.

How to use it

Run the prompts in order in a single chat so each one uses the context of the last. Replace the bracketed parts. Check each step off as you go.

1. 1

   Map the territory. Get the shape of the topic before the details.

   I am researching: [topic]. My goal: [decision or output you need]. Give me a map of this topic: the main sub-areas, the key terms (with one-line definitions), the major players (companies, projects, people, institutions), and the 5 questions an expert would ask first. Flag which areas are fast-moving, where information from the last 6 months matters most.

2. 2

   Find primary sources. Push past blog summaries to the original material.

   For the sub-areas most relevant to my goal, list primary sources: official docs, papers, datasets, repositories, filings, changelogs, original announcements. For each: what it is, why it is authoritative, and the link. Only include sources you can actually cite; if you are unsure a source exists, say so instead of guessing.

3. 3

   Argue the other side. Surface the strongest disagreements before you commit.

   What are the strongest arguments against the mainstream view on [topic/claim]? Who makes them and on what evidence? Then: what would have to be true for each side to be right, and what evidence would change your mind?

4. 4

   Extract and verify the claims. Turn the conversation into a checklist of facts, then check them.

   List every factual claim made so far in this conversation as a table: claim | source cited | confidence (high/medium/low) | how I can verify it. Mark any claim without a checkable source as UNVERIFIED.

   Now open the sources and check the high-impact claims yourself. Correct the model in the chat when something does not hold up.

5. 5

   Synthesize a brief. Compress everything into something you can act on or share.

   Using only the verified claims, write a one-page brief: summary (3 sentences), what is known, what is contested, what is unknown, and the 3 most important sources. Keep uncertainty explicit.

6. 6

   Decide what to do. Research is only useful if it changes an action.

   Given this brief and my goal ([goal]), what are my 3 best options? For each: first concrete step this week, cost/effort, main risk, and the signal that would tell me it is working. Recommend one and explain why.

Make the prompts your own

The best research prompts include your context: your goal, your constraints and what you already know. The Prompt Builder helps you turn any of the six prompts above into a reusable, well-structured version for your use case.

Where this pays off

• Evaluating a new tool or model before adopting it.

• Researching a market before building a product.

• Preparing content that is accurate and more specific than what is already out there.