Skip to content
eWiki The practical AI reference

AI Tool Questions

How to Use ChatGPT Deep Research: Define the Question, Choose Sources, and Review the Report

Open sketchbook with geometric drawings resting on two gray books beside a pen and neutral color swatches on a sunlit white table.

Learn how to start ChatGPT deep research in Work, frame a focused request, provide relevant source context, and review the resulting report and its sources.

What ChatGPT Deep Research Is For

OpenAI describes deep research in Work as a multi-source investigation producing a reviewable report.

As an editorial suggestion, use deep research when you need to compare evidence, organize findings from several materials, or prepare a report that will inform later work. For a short explanation, brainstorm, or simple draft, regular Chat may be the more proportionate option: OpenAI distinguishes those uses from Work tasks with clear outcomes and reviewable deliverables.

Before starting, write down the decision or task the report should support. This is a suggested preparation step, not a guarantee of report quality. It gives you a concrete standard against which to assess the finished work.

Stack of gray and white notebooks with blue ribbon markers beside an open sketchbook, black pen, ruler, and loose diagram sheets.

Start Deep Research in ChatGPT Work

OpenAI documents Work → + → Deep research for ChatGPT web and Work → Plugins → Deep research → Try now for the desktop app, enabling the plugin if necessary. Availability varies by account and workspace. State the question, scope, requested result, and relevant files or connected-app context.

Editorial suggestion: Add only materials relevant to this assignment that you are authorized to use.

Person writing in an open document at a white table with stacked booklets and a mug, in front of shelves filled with binders and books.

Define the Question and Deliverable

OpenAI recommends telling ChatGPT Work the outcome you need, the sources or plugins to use, applicable constraints, what a good result looks like, and when it should stop for review or approval. The same Work guidance suggests setting boundaries such as using only specified sources or comparing a fixed number of options.

A useful prompt can include:

  • The audience and the decision the report should inform.
  • The exact question and subjects that are in scope.
  • Relevant time period, geography, or comparison criteria.
  • Materials or tools to use.
  • Constraints and exclusions.
  • The requested output, such as a brief with findings, uncertainties, and recommendations.
  • A direction to stop for your review before any later action.

These details are editorial suggestions for making the assignment easier to evaluate. They do not establish that every requested constraint will be followed perfectly. Consider defining ambiguous terms in the prompt and asking the report to distinguish sourced findings from recommendations.

Choose Relevant Source Context

ChatGPT Work can use files, plugins, and approved tools to retrieve information and create outputs, subject to what is available in the workspace. OpenAI’s Work documentation also says plugins can connect Work to services such as Google Drive, SharePoint, email, calendars, and project trackers.

As a practical suggestion, choose sources according to the question rather than adding everything available. For a policy comparison, you might identify the governing documents and a date cutoff. For an internal analysis, you might name the approved files and connected tools that contain the relevant evidence. If discovery is part of the assignment, say what kinds of sources should receive priority.

Also consider asking the report to flag missing information and separate evidence from interpretation. That request can make gaps easier to notice during review, but it does not prove that every gap will be found.

Review the Report and Its Sources

OpenAI’s instructions call for reviewing the report and its sources and asking follow-up questions.

A suggested review sequence is to identify the claims most important to your decision, open their cited sources, and check whether those sources support the report’s wording. Look for omitted qualifications, inconsistent dates, unclear definitions, and conclusions that extend beyond the cited material. Then ask focused follow-up questions about discrepancies or missing evidence.

For consequential work, consider having an appropriate person review the relevant claims and underlying materials. This is a risk-management recommendation, not an assurance that review will eliminate errors.

Conclusion

Editorial suggestion: Treat a research assignment as a question with boundaries, an output to inspect, and a decision about what to do next.

The editorial recommendation is to begin with a bounded request that names the audience, decision, scope, source context, constraints, and output. Afterward, verify the claims that matter most and use follow-up questions to resolve gaps before relying on the report for a consequential decision.

Sources

Disclosure

This eWiki article was prepared with AI assistance on October 23, 2026.

Improve this guide

Report a problem or suggest the next question

Submissions enter a private editorial queue. They are reviewed and deduplicated before any public update or new page is created.