ChatGPT Deep Research: How to Use It, Steps and Prompts

In brief

ChatGPT deep research is a tool that browses the web for 5 to 30 minutes and returns a cited report. You start it from the + menu in the message box. As of November 2026 OpenAI lists it as Limited on Free and Go, Expanded on Plus and Maximum on Pro, with no task counts.

How do you use ChatGPT deep research?

You start ChatGPT deep research from the message box and let it run for 5 to 30 minutes. The steps, from OpenAI's help page read on 15 November 2026:

A brushed chrome magnifying glass hovering over an open cream report booklet embossed with the OpenAI logo, a thin thread of rainbow glass running across its pages, on a charcoal plinth under a warm rim light.
  1. Open a new chat and choose Deep research. Three paths work: type /Deepresearch in the prompt, pick Deep research from the + menu in the prompt window, or pick it from the sidebar menu.
  2. Write the outcome you want, not a topic. Say who the report is for, what decision it feeds and what a good source looks like. Attach files or spreadsheets if you have them. OpenAI's launch post says you can.
  3. Choose where it may look. The public web and your uploaded files are the defaults. You can restrict it to specific sites, and connected apps count as sources where your plan, region and workspace allow them.
  4. Read the research plan ChatGPT proposes, then start the run. Edit the plan first if a step is wrong. It costs nothing to fix a plan and a lot to fix a finished report.
  5. Watch or leave. A sidebar lists the steps taken and the sources used. You can interrupt to narrow the focus or add a source, or close the tab and wait for the notification.
  6. Open the report. It fills a fullscreen view with a table of contents, a sources-used section and an activity history. Download it as Markdown, Word or PDF.
Quoted run times: ChatGPT 5 to 30 minutes, Gemini 5 to 10, Perplexity 2 to 4
0102030minutes per reportChatGPT deepresearch5 min30 minOpenAI launch postGemini DeepResearch5 min10 minGoogle help page, longer if complexPerplexity DeepResearch2 min4 minPerplexity launch post

All three ranges come from the vendors. OpenAI's launch post says the tool "accomplishes in tens of minutes what would take a human many hours". Google's Gemini help page says reports usually take about 5 to 10 minutes and complex ones may take longer. Perplexity's launch post says 2 to 4 minutes. None of the ranges is a promise. Plan on the long end for a broad brief.

What can ChatGPT deep research do that normal chat search can't?

Deep research plans a multi-step search, reads many pages and writes one documented report, where normal chat search answers in seconds from a few pages. OpenAI's help page says standard chat or search may be faster for quick lookups. Its launch post describes deep research as a tool that "conducts multi-step research on the internet for complex tasks".

The differences you can feel are four. A run can use uploaded files next to the web. It shows its plan before it starts. It keeps an activity history you can check. And every output carries citations or source links. Connected apps are read only: deep research reads from a document store such as Google Drive or SharePoint, and it doesn't write to one. App connections don't override your provider's permissions or your workspace's rules.

Which prompts get a good ChatGPT deep research report?

A good deep research prompt names the decision, the scope and the source standard in the first two lines. These five are original to this page. Swap the bracketed parts:

  1. Market scan: "I'm deciding whether to enter [market] in [country]. Compare the five biggest vendors on pricing, target customer and weaknesses users complain about. Use vendor sites and published filings only, and mark any number you couldn't source."
  2. Literature review: "Summarize what peer reviewed studies since [year] say about [topic]. Group them by finding, flag where they disagree, and list sample sizes. Skip blogs and press releases."
  3. Buying decision: "I need a [product] under [budget] for [use]. Find the three best options, what reviewers who tested them agree on, and one reason to skip each."
  4. Policy check: "What are the current rules on [topic] in [state or country]? Cite the statute or agency page for each rule and say when each page was last updated."
  5. Briefing before a meeting: "Brief me on [company] before a call with its [role]. Cover products, recent funding or leadership news, and three questions worth asking. Cite every claim."

Two habits lift all five. Say what you already know, so the report skips it. And ask for uncertainty by name. OpenAI's launch post admits the tool shows weakness in confidence calibration, often failing to convey uncertainty accurately, so asking it to mark weak claims does real work.

Why is the ChatGPT deep research button missing?

The button is missing when your plan, country or workspace doesn't allow it, or when you're looking in the wrong place. OpenAI's help page says availability depends on your plan and your country or territory. It also says Enterprise and Edu admins control access through role-based access control, and that not every app or plugin supports deep research.

Check in order. Type /Deepresearch in a new chat. Open the + menu in the prompt window. Open the sidebar menu. If all three show nothing, check your plan: as of November 2026 the ChatGPT pricing page lists deep research on Free and Go as Limited, on Plus as Expanded and on Pro as Maximum. On a work account, ask your admin. Agent mode is a separate door. We cover where it went in our guide to agent mode in ChatGPT.

How does ChatGPT deep research compare with Gemini, Claude and Perplexity?

All four assistants now sell a deep research mode. They differ most in whether you can steer a run and whether the vendor tells you how long it takes. We read each vendor's own pages on 15 November 2026:

ChatGPTGeminiClaudePerplexity
Where you start it+ menu, /Deepresearch or sidebarAdd Files, then Deep Research+ button, then Research (web search must be on)Deep Research in the mode selector of the search box
Time the vendor states5 to 30 minutesUsually about 5 to 10 minutesNone on its help page2 to 4 minutes (launch post)
Who gets itFree and Go (Limited), Plus, ProAll users, higher limits on AI Pro and UltraPro, Max, Ultra, Team, EnterpriseFree at 1 research query a month, Pro and Max on monthly limits
Change course mid runYes, interrupt to refineEdit plan before Start researchNot statedNot stated
ExportMarkdown, Word, PDFExport to DocsNot statedPDF, document or Perplexity Page
Count published for the main paid planNoneNone, daily request limitNone, shares chat limitsNone, monthly limits

Pick ChatGPT when you want to steer a long run and download a Word or PDF file. Pick Gemini when you want a free start and a Google Docs export: its plan editor works before the run, and Google's page says all users can run Deep Research with Thinking. Pick Claude Research if you already pay for Claude. Its help page says research counts against the same limits as standard conversations and can use them up faster. Pick Perplexity for a source-first report. Its plan comparison is the only one that still prints a free count: 1 research query a month. Caps are covered in our deep research limit guide and the Gemini Deep Research page.

What can't ChatGPT deep research do?

ChatGPT deep research can still state false facts, and OpenAI says so. Its launch post says the tool "can sometimes hallucinate facts in responses or make incorrect inferences", and that it may struggle with "distinguishing authoritative information from rumors". Read the sources-used section before you trust a number. Open the two or three links behind your most important claim.

It also can't act. Connected apps are read only, so it won't send, edit or file anything for you. It isn't instant, so a quick lookup belongs in ordinary chat. And the allowance has no published number. OpenAI's help page says your in-product counter shows remaining tasks and, on plans with a fixed monthly allowance, resets every 30 days from your first use. If you hit zero, the limit guide lists the next moves.

If you want deep research and Claude, GPT and Gemini chat in one place, Whizi's web app has its own Deep Research. A run costs 200 credits from your plan, and it needs a paid plan. Pro includes Claude, GPT and Gemini, and its first week is $1.99. It is Whizi's tool, not ChatGPT's, so it doesn't share ChatGPT's plan or counter. The free plan gives 10 messages a day on Whizi AI, Auto and Claude Haiku 5.5, with no card, but it doesn't include Deep Research.

Workflow checklist
  • Start from the + menu or type /Deepresearch, then read the proposed plan before you press start
  • Write the decision the report feeds and the sources you trust, not just a topic
  • Interrupt the run early if the sources look wrong; ChatGPT lets you add one mid run
  • Open the links behind your two most important claims before you reuse them
  • Send quick lookups to normal chat so your deep research allowance goes on real synthesis
Common questions

Frequently asked questions

How long does ChatGPT deep research take?

OpenAI's launch post says a run may take anywhere from 5 to 30 minutes. The help page gives no estimate. You get a notification when it finishes, so you can leave the tab. Broad briefs sit near the long end.

Is ChatGPT deep research free?

Partly. As of November 2026 the ChatGPT pricing page lists deep research on Free and Go as Limited, with no count. Plus is Expanded and Pro is Maximum. OpenAI's current pages give no count, and the limit guide has the last numbers it published.

Can ChatGPT deep research use my own files?

Yes. Uploaded files are a default source next to the public web, and OpenAI's launch post says you can attach files or spreadsheets. Connected apps such as Google Drive or SharePoint also work where your plan, region and workspace allow, as read only sources.

Is ChatGPT deep research better than Gemini Deep Research?

It depends on what you value. ChatGPT lets you interrupt a run and downloads Markdown, Word or PDF, while Gemini asks you to edit the plan before the run and exports to Google Docs. Google says Gemini runs usually take 5 to 10 minutes, OpenAI says 5 to 30.

Why is my ChatGPT deep research taking so long or stuck?

A run may take 5 to 30 minutes, so a long wait is normal. OpenAI's help page says you can follow progress and interrupt to refine the focus, and that search or standard chat is faster when time is critical. The sidebar lists the steps taken and sources used.