You have 90 minutes to brief a partner on an unfamiliar industry. Two years ago this meant skimming 30 tabs and writing fast. Today, Deep Research (ChatGPT, Gemini, Claude) produces a 10-page brief with citations in 20 minutes — and the remaining 70 minutes go to spot-checking, tightening, and forming your own opinion. This tutorial is for strategists, consultants, PMs, and founders who need a credible brief fast, and shows where Deep Research wins versus where it still misleads.
What this covers
The full loop: framing the question, choosing the right Deep Research engine, structuring the output, spot-checking the citations, and re-researching the weak sections. We treat Deep Research as a research analyst who needs your editing — not as an oracle. The output is a brief you can defend in a room, with sources you have actually opened.
Who this is for
Strategy and consulting teams scoping a new vertical, PMs writing competitive landscape decks, founders pre-checking a thesis before a board meeting, and any analyst who normally bills the first three hours of a project to “skimming.” If your bar is “summary I can re-read tomorrow,” a regular chat is enough. If the bar is “I have to defend this to someone smarter than me,” you need Deep Research.
When to reach for it
When you need a 5-15 page briefing on a market, technology, or competitive landscape. The same Deep Research muscles also work for historical context — see the AI history-timeline workflow for turning a broad topic into dates, actors, and causal links. Not the right tool for: very recent breaking events (real-time gaps remain), narrow internal company questions (sources are too sparse), or topics where most knowledge is behind paywalls.
Before you start
- Frame the actual decision the brief supports. “Should we enter X market?” is a question; “tell me about X” is not. The decision shapes which sections matter.
- Decide the audience and depth. An exec-only brief is 3 pages; a strategy-team working doc is 12. Tell the agent before it starts; refactoring after is expensive.
- Know your “must-include” angles ahead of time — TAM, top 5 players, regulatory risks, key tech bottlenecks. The agent does not know what your reader cares about.
- Have one credible domain source bookmarked. You will use it to sanity-check the brief’s numbers in 90 seconds.
Step by step
- Pick the Deep Research mode that matches your need: ChatGPT for broad coverage and clean formatting, Gemini for the freshest web data, Claude for analytic depth and clearer reasoning. If you go with ChatGPT, the ChatGPT Deep Research tutorial walks through Project setup, question challenge, and the spot-check loop step by step.
- Start with one clear research question. Strong: “How is the AI coding tool market segmenting in 2026, and which segments will consolidate by 2027?” Weak: “Tell me about AI coding tools.” The strong version forces the agent to make a defensible argument.
- Tell the agent the format expectation: number of sections, target length, executive vs detailed, audience seniority. A typical brief: Executive Summary, Market Size, Segments, Top 5 Players, Tech Trends, Risks, Outlook (12 months), Sources.
- Let it run (5-30 minutes depending on the engine). Read with a pen: mark every claim where a number, name, or causal link looks too clean.
- For each weak claim, ask the agent for sources or have it re-research that section. “Re-research Section 3 with at least two sources per claim and prefer 2025-2026 primary sources” is a strong follow-up prompt.
- Spot-check 3-5 cited sources manually. Deep Research engines still cite weak sources (Medium posts, AI-generated articles, outdated reports). One bad source in a brief kills the whole document’s credibility.
First-run exercise
- Pick an industry you actually know something about — e.g., your own domain. You will catch errors instantly and learn the engine’s failure modes cheaply.
- Run a 5-page brief on it. Time yourself: how long until first-draft? How long until you trust it?
- Mark which sections are 90%+ usable, which need re-research, and which are wrong. Most first runs have one fully wrong section — that is normal.
- For the second run, change only the engine (try Gemini if you used ChatGPT). Compare strengths. Most teams settle on one engine after three real runs.
Quality check
- For every numeric claim, click through to the source and check the original number. Deep Research occasionally rounds aggressively or pulls from secondary sources.
- For every player named, verify they still exist and operate in this segment. The web has many ghost companies cited by stale articles.
- Read the executive summary aloud. If it sounds confident but says nothing falsifiable, ask the agent to take a position.
- Diff against your one bookmarked credible source. If the brief contradicts a source you trust, the brief loses.
How to reuse this workflow
- Save the framing question, format spec, and any prompts that re-researched weak sections as a template. New industry, new variables, same skeleton.
- Build a personal “source whitelist” of domains the engine should prefer (or avoid). Pin it in the system prompt of each project.
- Re-run the brief every quarter on the same template. Diffing two quarterly briefs is the cheapest market-tracking system you can build.
Recommended workflow
Research question → engine choice → format spec → agent draft → spot-check 5 sources → re-research weak sections → final synthesis → cite-ready brief. Once the brief lands on a few rival players you actually need to watch closely, follow up with an AI competitor content teardown to read what their content choices reveal about strategy.
Common mistakes
- Trusting the brief without source spot-check — one fabricated stat ends the meeting.
- Asking too broad a question (“tell me about AI”) — the agent gives a Wikipedia article in 10 pages.
- Skipping the format spec — output ends up the wrong shape and you reformat for an hour.
- Running Deep Research on private information not on the web — it cannot find it; you get hallucinations.
- Not committing to a single engine for the project — switching mid-project means inconsistent voice and incomparable drafts.
- Treating the brief as final — it is a draft for editing, not a document for shipping.
FAQ
- How long does Deep Research take?: Usually 5-30 minutes for a substantive brief. The engine reports progress in real time.
- Which engine is best?: ChatGPT for clean formatting and broad coverage. Gemini for newest data. Claude for analytic depth. Pick based on the brief’s main weakness.
- Can I use it for paywalled industries (legal, medical)?: Limited. Engines skip most paywalled content. Pair with a domain database.
- How many sources are enough?: At least 15 distinct sources in the bibliography, and at least 3 you would cite to a partner. Less and you are vulnerable to one bad input.
- Should I name the brief’s audience in the prompt?: Yes. “For a private equity partner who wants to test a thesis” produces different output than “for a junior analyst.”
- Can it produce charts?: Some engines yes, but the numbers are often wrong. Generate the data and chart yourself.
Related
- NotebookLM getting started
- AI multi-source synthesis
- How to Build an AI Research Workflow That Actually Holds Up
- ChatGPT Deep Research tutorial
Tags: #Tutorial #Research #Deep Research