Some questions deserve more than a quick answer, such as whether to enter a new market or how a competitor is changing course.
ChatGPT deep research plans a multi-step investigation, reads across many sources and returns a report with citations you can check. Think of it as delegating a research assignment, not asking a question. It can still miss evidence or misread a source, so the judgment stays yours.
Where to find deep research
In a new chat, type @deep research, or select + next to the message box and choose Deep research.
What your plan includes
- Every plan: deep research is available on all ChatGPT plans. OpenAI describes the allowance as limited on Free and Go, expanded on Plus and the highest on Pro.
- Business and Enterprise: deep research is included, and your workspace uses credits for anything beyond the included amount.
- Your remaining tasks: ChatGPT shows how many deep research tasks you have left. A fixed allowance resets 30 days after your first use, not on the first of the month.
- Admin settings: if you do not see deep research, your admin may not have turned it on for your role or group.
When deep research earns its place
There are four situations where deep research pays off quickly:
- A strategic decision, such as comparing markets, competitors or business models.
- A board or leadership briefing that needs sources behind every claim.
- A body of evidence to synthesize, such as studies on productivity or AI adoption.
- An unfamiliar topic, such as a new regulation, before you make a decision.
For a quick fact or a short summary, regular chat with web search is faster. Save deep research for questions where the extra time pays for itself.
How to run a deep research report, step by step
Step 1: Start deep research. Open a new chat and choose Deep research from the + menu.
Step 2: Write your brief. Say what decision the research supports, how it should search and what the report should look like. The decision you name shapes everything the research looks for. A template is below.
Step 3: Choose your sources. By default, deep research uses the web and any files you upload. You can also turn on connected apps, such as Google Drive or SharePoint, where your admin allows them. To keep the research on trusted sources, limit web searches to specific sites.
Step 4: Review the plan. ChatGPT proposes a research plan and starts a short countdown. Select Edit and describe your changes in plain language before it begins.
Step 5: Steer while it works. A report usually takes between 5 and 30 minutes. If the research heads in the wrong direction, interrupt it with new instructions to narrow the scope or change the sources, without starting over.
Step 6: Check the sources. Open Activity to see the searches and the path the research took, and Sources to see everything it read. Open the sources behind your most important claims and confirm they say what the report says.
Step 7: Export and keep working. Open the report full screen to navigate by its table of contents, then export it to Markdown, Word or PDF. Keep asking follow-up questions in the same chat, which now holds all the research as context.
A CODO prompt to start with
Paste this into deep research, using our CODO framework, and replace the bracketed parts. The Character is optional here, but the Objective, Do and Don't lines and the Output still matter.
Character: You are a strategy analyst who prepares research briefings for a board of directors. Objective: Research [2026 expansion opportunities in the US market] so my board can decide [which market to enter first]. Do: - Prioritize primary and reputable sources, such as [the World Economic Forum, BCG and government statistics]. - Use sources from the last [18] months, and flag anything older. - Compare the options in a table with size, growth, risks and cost to enter. - Separate what the sources establish from your own recommendations. - Cite a source for every material claim. Don't: - Don't rely on vendor marketing pages or unsourced blog posts. - Don't present estimates as facts. Label them as estimates. Output: A meeting-ready report with an executive summary, key findings, risks, recommendations and a comparison table.
Want a ready-made starting point? These how-to articles show practical ways to research with ChatGPT:
- Create a competitive brief and slide deck using ChatGPT Agent: research a market, then turn the findings into a deck.
- Write a custom report from your data using ChatGPT connectors: bring your connected apps into a report.
- ChatGPT Slides: turn a finished report into a deck your team can edit together.
What will your IT and legal teams ask?
- What can it reach? The web, the files you upload and the connected apps your admin allows. It only reads from those apps, and it follows the permissions you already have.
- Can we restrict the sources? Yes. You can limit web searches to specific, trusted sites.
- Is our data used for training? On Business and Enterprise plans, OpenAI does not train its models on your workspace data by default.
- Who controls access? Admins decide which roles and groups can use deep research and which apps it can connect to.
ChatGPT deep research or another research tool?
ChatGPT deep research is the natural choice if ChatGPT is already your main AI assistant, especially when you want to steer the research while it runs.
If your company runs on Google Workspace, Gemini Deep Research draws on Gmail and Drive. On Microsoft 365, Copilot Researcher works with your emails, meetings and files. For what the scientific literature says, Consensus puts academic papers first. Our overview of AI deep research tools compares all six options.
Key takeaways
- Delegate a research assignment, not a quick question.
- Name the decision the research should support.
- Edit the plan before the countdown ends.
- Steer while it runs instead of starting over.
- Check the key sources, because a citation does not make a claim correct.
