Run Deep Research in ChatGPT

Some questions are too big for a normal chat. A market scan, a policy comparison, a competitor review, or a due diligence brief all need many sources and a real read across them. That is where Deep research helps. It plans a small project, searches multiple sources, and returns a report you can review.

Running Deep research in ChatGPT is powerful, but it is not magic. If your plan is available, use it well. Set a clear scope, review the plan before it runs, and always check the citations before you act on the findings. This guide walks through how to brief it, how to steer it, and how to sanity check the report.

What Deep research is good for

Deep research shines when a question needs many sources and a shared frame across them. Comparing three vendors, reviewing a market, summarising a legal area, or building a short primer on a niche topic. It is less useful for one liner questions where a normal search does the job faster.

• Comparing vendors, tools, or plans
• Market or industry scans
• Policy or legal overviews across regions
• Company and competitor primers
• Long form background reading for a decision

When Deep research is the wrong tool

Skip it for quick facts, current prices, or news. A normal Search reply is faster. Skip it when you have already read the sources yourself and only need a synthesis. In that case a summary prompt with your notes will be cheaper.

• Quick facts with one right answer
• Live prices or availability
• Questions you can answer from your own uploaded files

What to give ChatGPT to brief Deep research

A good brief has five parts. The question, the scope, the audience, the shape of the report, and the sources you trust. A weak brief brings back a wide report that answers everything a little and your question not enough.

• Question: the exact decision the research must support
• Scope: regions, industries, dates, or products
• Audience: who will read the report and how much they know
• Shape: sections, length, tables, and any must include claims
• Sources: what to prefer, what to avoid, and how many is enough

A prompt frame you can reuse

Please run Deep research on this topic.
Question: Which of these three project tracking tools fits a 25 person remote agency best in 2026: Linear, Asana, ClickUp?
Audience: our head of operations, who has used none of these tools
Scope: 2025 to 2026 features, pricing for teams under 50, and integrations with Slack and Google Workspace
Shape:
- 1 page summary at the top
- One section per tool with strengths, weaknesses, and pricing
- A comparison table
- A recommendation with a reason
Sources: prefer official pages and reviews from named outlets. Avoid affiliate blogs.
Before you run, please show your plan and ask me any clarifying questions.

Why this prompt works

It sets a real decision, a real reader, a real time frame, a real output shape, and a source policy. Deep research now knows what a good report looks like. The last line invites a plan review, which is the single best step you can take.

Review the plan before it runs

Deep research usually starts by showing a plan. Read it. If a section is missing, name it. If a section is bloated, cut it. If the sources look weak, tell it to prefer specific pages or outlets. Five minutes on the plan saves an hour of report cleanup.

• Check the sections cover the decision you actually need
• Check the time frame matches the question
• Check the sources include first party pages, not only aggregators
• Add or cut any section, then approve the plan

Four realistic examples

Example 1: a vendor selection

You are choosing a payments provider for a small store. Ask Deep research to compare three named providers with columns for fees, dispute rates, and integrations. Restrict sources to official docs and named finance outlets. Ask for one recommendation with reasons.

Example 2: a policy scan

You need a plain overview of the new UK data protection rules for a non lawyer. Set the audience as a general reader. Ask for a short glossary, a timeline, and a table of who is affected. Restrict sources to official government pages and named legal outlets.

Example 3: a market primer

You are new to a market and need a 1000 word primer. Ask for the size of the market, the main players, the main customer types, and 3 open questions to explore. Ask for citations after every number and every named player.

Example 4: a technical background read

You want to understand vector databases before a meeting. Ask for a plain intro, a comparison of 3 named tools, and a short table with columns for use case, cost, and community size. Restrict sources to official docs and engineering blogs from named companies.

What makes a good Deep research report

• Every claim has a citation
• The report answers the decision, not just the topic
• Numbers include units and dates
• Trade offs are named plainly, not hidden in soft language
• There is a short summary at the top for scanning

How to check the report before you use it

Pick the 3 to 5 claims that would matter if they were wrong. Open the citation for each. Read the paragraph that supports the claim. If the page does not support the wording, ask Deep research to redo that section with a better source. Never publish or send a report based only on the summary at the top.

• Open at least 3 citations, from different sources
• Check the date on each cited page
• Read the paragraph the citation points to, not the page title
• Ask for a redo when the source and the claim do not match

How to refine after the first draft

If the report is close but a section is thin, name that section and ask Deep research to dig deeper only there. If the report is too broad, cut sections by name. If the sources are weak, restrict them by outlet or type.

• Please expand only the pricing section for each tool
• Please remove the section on history and keep the current features
• Please redo the market share numbers using only reports from named analysts

Common mistakes

• Skipping the plan review
• Setting the scope too wide, which spreads the report thin
• Trusting the summary without opening the sources
• Using Deep research for a question a normal Search would answer

More Deep research prompt examples

Different research jobs need different briefs. Here are three ready to paste and adapt.

Compare tools for a buying decision

Run Deep research on this question.
Question: Which project management tools are best for a 8 person remote team that needs simple task boards and time tracking?
Constraints: under $20 per user per month, works on web and mobile, no heavy enterprise setup
Deliver:
1) shortlist of 5 options
2) comparison table of must-have features
3) pricing notes with dates
4) risks or lock-in concerns
5) sources for every major claim

Policy or regulation overview

Run Deep research.
Question: What are the main GDPR requirements a small SaaS company must handle for EU customers?
Audience: non-lawyer founder
Deliver:
- plain language summary
- checklist of practical steps
- common mistakes
- what still needs a lawyer
Cite sources and note the date of each source.

Market scan for a side project

Run Deep research.
Question: Is there demand for a simple meal planning app aimed at busy parents?
Scope: English-speaking markets, last 3 years of evidence
Deliver:
- demand signals
- competitor examples
- pricing ranges
- open gaps
- weak spots in the evidence
Separate facts from guesses.

Takeaway rule

Brief it like a junior analyst. Set the question, the scope, and the shape of the report. Review the plan before it runs. Open the citations before you act. Deep research earns its name only when you steer it.