ChatGPTDeep ResearchLevel 3: Workflows

How Do You Use ChatGPT Deep Research?

Type /Deepresearch in the prompt, or choose Deep research from the tools menu (+) or the sidebar, then describe the outcome you need. ChatGPT proposes a research plan, which you can edit, and returns a structured report with citations or source links. OpenAI's pricing page lists it as limited on Free and Go, expanded on Plus and maximum on Pro.

Prompt to copy
Research question: [the single decision or question, in one sentence]
Who will use this, and for what: [audience and the decision they'll make]
Must cover: [3 to 5 specific dimensions]
Leave out: [what's out of scope]
Time range and region: [for example, 2024 onward, EU only]
Report format: [one-paragraph answer first, then a comparison table, then sections on...]
Length: [for example, about 1,500 words]

Where sources disagree, show both and say which is stronger and why. Mark anything you couldn't confirm in a primary source.
  1. 1Describe the outcome
  2. 2Edit the proposed plan
  3. 3Watch and redirect
  4. 4Check the cited report
The plan step is your best chance to fix the scope before the run uses up one of your tasks. Steps from OpenAI's help page.
  1. Free and GoLimited
  2. PlusExpanded
  3. ProMaximum
Deep research allowance by plan. Source: OpenAI's ChatGPT pricing page, checked 6 November 2026. The pages read publish no task counts.

Try it now

  1. 1Open a new chat, type /Deepresearch, and paste the prompt with a real question filled in.
  2. 2Answer any clarifying questions, then read the proposed plan. Remove any part you don't need and add anything missing before it starts.
  3. 3Keep working while it runs, or watch its progress. If it heads the wrong way, interrupt it and redirect it.
  4. 4It worked if the report opens full screen with a table of contents and a sources section, and its first paragraph answers your question.

Use it when

Questions that need many sources combined: comparing vendors, reviewing regulation across countries, summarizing a research field, or checking a market before a decision. OpenAI suggests it for multi-step or in-depth questions, especially when you want control over which sources get used.

Skip it when

A quick lookup or short conversation. OpenAI says standard chat may be faster, and to use search or standard chat if time is critical. It's also not available everywhere, since OpenAI says availability depends on your plan and your country or territory.

Why it works

Deep research plans, reads many sources, then writes, so most of the value depends on the brief. OpenAI's help page says a good prompt describes the question, the desired outcome and any relevant constraints, and that ChatGPT may ask clarifying questions before it starts. By default it can use the public web, files you upload and connected apps. You can follow progress and interrupt at any time to change the focus or the sources. Your in-product counter shows remaining tasks. On plans with a fixed monthly allowance, it resets every 30 days from your first use.

Example

Instead of "research CRM tools", try: "Which CRM should a 5-person B2B sales team choose: HubSpot, Pipedrive or Attio? Cover price at 5 seats, API limits, and migrating from spreadsheets. Use current vendor pricing pages. Output: a recommendation first, then a table." When the plan appears, cut anything about enterprise features, since a team of five won't need them.

Going further

You can check your count before you commit: the in-product counter shows what's left. OpenAI's pages read for this card don't publish exact task numbers per plan, so older figures you may find online shouldn't be trusted. Lockdown Mode (Settings > Security) restricts deep research, which can look like a missing feature. Deep research also follows the same data and privacy settings as regular chats, and deleting the chat deletes its report. You can still choose a legacy model for a run, though OpenAI removed the older "legacy deep research mode" on March 26, 2026.

Common mistakes

  • Skipping the plan because it looks fine. It's the cheapest moment to fix scope, before a run from a limited allowance goes the wrong way.
  • Starting over when you remember a constraint. Interrupt the run and add it; OpenAI's release notes describe updating mid-run without a restart.
  • Not stating the format, then getting long prose when you wanted a comparison table.
  • Deleting the chat to tidy up and losing the report. Download it first, as Markdown, Word or PDF.

Sources

Checked 6 November 2026. Products change often, so confirm details with the vendor.

Related

Beyond the course