AI tools

How to Use Deep Research in ChatGPT and Gemini: Brief It, Then Audit It

By Sauravjit Singh · Updated 2026-11-07 · 6 min read

Short answer: Deep research is an AI mode that spends several minutes searching and reading many sources, then writes a cited report. Start it from the tools menu in ChatGPT or Gemini, give a detailed brief (decision, scope, sources, format), answer its clarifying questions, then audit the sources before using the findings. Use it for multi-source questions, not quick facts.

Normal chat answers in seconds from what the model already knows, plus a quick search if you ask. Deep research works more like a junior researcher: it plans, runs dozens of searches, reads pages, and returns a structured report, usually within 5 to 30 minutes.

The quality of that report depends far more on your brief and your checking than on which tool you pick. This guide gives you a briefing template, a worked Indian market question, and the audit we run on every report before anything from it goes into a client deck.

When is deep research worth using?

Use it when the answer needs many sources combined and you would otherwise spend an hour or more searching. For anything shorter, it is slower and no better.

TaskBest tool for itWhy
A single fact or current ruleNormal chat with web search, or a quick Perplexity searchSeconds instead of minutes
Market or competitor overview for a new ideaDeep researchNeeds many sources combined into one picture
Background before a client or investor meetingDeep researchStructured report you can skim and verify
Studying from documents you already haveNotebookLMAnswers only from your sources, not the open web
Literature scan for a college projectDeep research, then NotebookLM on the papers it findsDiscovery first, then grounded reading

If you are deciding between tools rather than learning the mode itself, our comparison of Perplexity and ChatGPT for research covers that choice.

What does a deep research report actually contain?

Expect a long, structured document: an executive summary, sections that follow the questions it planned, often a table or two, and a list of sources with links. Length ranges from a few pages to well over ten, depending on the tool and the brief.

What it does not contain matters as much. It cannot see paywalled reports, most private databases, or data that exists only offline, such as a local market's actual rents or a trade association's member surveys. For small, local Indian questions, the online evidence is often thin, and a good report will say so; a weak one will fill the gap with national figures dressed up as local ones. Reading for that substitution is a habit worth building early.

How do you start deep research in ChatGPT, Gemini and Perplexity?

In all three it is a mode you switch on before sending your message. Locations below are as of November 2026; interfaces change often.

ToolWhere to find itWhat happens next
ChatGPTThe plus (+) button or tools menu in the message box, then Deep researchOften asks a few clarifying questions, then works in the background and notifies you
GeminiTools or mode selector in the Gemini app, then Deep ResearchShows a research plan you can edit before it starts; reports can be exported to Google Docs
PerplexityMode selector by the search box, then ResearchUsually faster and shorter; lists sources as it goes

All three limit how many reports you can run, with tighter caps on free plans and higher ones on paid plans. Check the current limits in each tool before planning a big project around it.

A briefing template for deep research

A one-line request produces a generic report. A brief with these seven parts produces something you can use. Copy it and fill in the brackets with your own details.

  1. Decision: what you will decide using this report.
  2. Context: who you are and what you already know, so it does not waste effort on basics.
  3. Scope: geography, time period, segment, and what is out of scope.
  4. Questions: three to six specific questions the report must answer.
  5. Source rules: which sources to prefer, which to avoid, and how recent they must be.
  6. Output: structure, length, tables wanted, and a sources list with dates.
  7. Honesty rules: label estimates as estimates, say when data is missing, flag conflicting figures.

Worked example: an Indian market question, briefed properly

Say you are planning a Punjabi home-style tiffin service from a cloud kitchen in West Delhi. Here is the filled brief:

'Decision: whether to launch a lunch tiffin subscription for office workers in West Delhi in the next six months. Context: I have run a dhaba for eight years and know food costs; I do not know online ordering or subscriptions. Scope: Delhi NCR, focus on West Delhi; 2023 onwards; exclude fine dining and grocery delivery. Questions: (1) How large and fast-growing is India's cloud kitchen and meal-subscription segment, with sources? (2) What commission and fee structures do food delivery platforms typically charge restaurants? (3) What FSSAI licence or registration does a cloud kitchen need? (4) What pricing per meal do existing tiffin services in Delhi advertise? (5) What are the main reasons cloud kitchens shut down? Source rules: prefer government sites, FSSAI, company filings and annual reports, established business newspapers; avoid market-research press releases unless the methodology is stated; nothing older than 2023 for numbers. Output: one-page summary first, then a section per question, a table of pricing examples, and a dated source list. Honesty: mark any figure from a single source, and say clearly if you could not find data.'

Critique: the decision line keeps the report focused; question 3 has a clear official answer you can check; question 1 is where reports most often go wrong, so the source rules target it directly.

Running the same question in three tools: the audit sheet

If you can, run the same brief in two or three tools and compare. The point is not to crown a winner but to see which claims hold up across them. Fill this audit for each report:

Audit checkWhat to recordRed flag
Total sources citedCount from the source listVery few sources for a broad question
Primary sourcesGovernment, regulator, company filings, original studiesNone at all for regulatory or financial claims
Market-size figuresTrace each figure to its original publisherFigure appears only on a press-release or report-reseller site
DatesPublication date of each key sourceNumbers from before your scope period presented as current
Spot-check 5 claimsOpen the page and find the exact sentenceThe page does not say what the report claims
ContradictionsSame metric, different numbers across toolsReport presents one figure as settled

The most common weak point in reports on Indian markets is market-size figures: many trace back to report sellers whose methodology is not public. Treat those as rough signals, never as facts for a pitch deck. The full verification method is in our guide to fact-checking what AI tells you.

After the report: verify, then work with it

  1. Read the summary, then jump to the sources list before reading the body.
  2. Run the audit sheet above, especially the five-claim spot check.
  3. Delete or mark any claim you could not verify.
  4. Download the strongest primary sources and load them into NotebookLM, so follow-up questions are answered only from verified material. Our NotebookLM walkthrough shows how.
  5. Ask the original tool follow-ups in the same chat: 'Which three claims in this report rest on the weakest evidence?'
  6. Write your own one-paragraph conclusion. If you cannot, the report has not done its job.

Common mistakes with deep research

  • Using it for questions a quick search answers in seconds.
  • Skipping the clarifying questions or the plan review, which is where you fix the scope cheaply.
  • Pasting report figures straight into presentations without tracing them.
  • Asking for 'everything about' a topic. Broad briefs produce long, shallow reports.
  • Forgetting that the report reflects what is findable online; small local markets are often thinly covered.

A good brief is just a structured prompt, so the principles in our guide to writing clear AI prompts apply directly. Research workflows with deep research and NotebookLM are part of our Applied AI Professional course.

Frequently asked questions

Is deep research available on free plans?

As of November 2026, ChatGPT, Gemini and Perplexity have each offered some deep research on free tiers, with low monthly or daily caps. Paid plans raise the limits; check each tool's current plan page.

How long does a deep research report take?

Typically somewhere between a few minutes and half an hour, depending on the tool and how broad the brief is. You can usually leave the page and come back when it finishes.

Which is better, ChatGPT or Gemini deep research?

Both produce useful reports; quality varies by question. Gemini's editable plan and Google Docs export suit people in Google Workspace, while ChatGPT's clarifying questions help tighten the scope. Running both on an important question and comparing sources is the safest approach.

Can I trust the citations in a deep research report?

Not blindly. Citations can point to weak or outdated pages, or to pages that do not say what the report claims. Spot-check at least five key claims before using the report.

Can deep research read my own files?

Some tools let you attach files or connect apps alongside web research, depending on the plan. For questions that should be answered only from your documents, NotebookLM is the better fit.

Want to learn this with a trainer?

Leave your details and our counsellor will call you within one working day with batch dates, fees and the right course for you.

  • Free counselling call, no obligation to join
  • Weekday and weekend batches, online or offline
  • Small batches of 15 to 20 students

Or chat on WhatsApp

Enter a 10-digit mobile number.

We use your details only to contact you about Spettro School courses.

Chat on WhatsApp