The short answer
Deep research is an AI mode that turns one question into a research plan, runs many web searches, reads the sources it finds and writes a long report with citations. Google says Gemini Deep Research builds a multi-step plan you can edit and delivers a multi-page report in a few minutes. In November 2026 it is on ChatGPT's free plan with limits, in Gemini for free, in Claude as Research from the Pro plan and in Perplexity on paid plans.
| Also called | Research mode, Research (Claude), Deep Research (Gemini, Perplexity) |
|---|---|
| What it produces | A long report with links to its sources |
| Free options | ChatGPT (limited), Gemini Deep Research |
| Paid only | Claude Research (Pro and up), Perplexity Deep Research |
| Takes | Minutes rather than seconds |
| Main risk | Citations that do not support the claim |
What deep research is, in plain English
A normal chat answer comes from the model's training plus, at most, a quick web search. Deep research is closer to giving a junior analyst a brief. The assistant works out what it needs to know, searches many times, opens and reads pages, follows leads, and then writes a structured report with links so you can check where each point came from.
Google's description of Gemini Deep Research is a useful template for how all of them work. It turns your prompt into a multi-step research plan and shows it to you so you can adjust it. It then searches the web and, if you allow it, your Gmail, Google Drive and Google Chat content. You can follow its reasoning as it goes, and after a few minutes you get a detailed, multi-page report, which can also be turned into an audio summary.
Where to find deep research in November 2026
| Tool | Name | Plans |
|---|---|---|
| ChatGPT | Deep research | Limited on Free and Go, included on Plus, maximum on Pro |
| Gemini | Deep Research | Included on the free plan |
| Claude | Research | Pro, Max, Ultra, Team and Enterprise; not Free |
| Perplexity | Deep Research | Paid plans only |
| Vibe (formerly Le Chat) | Deep Research skill | A skill in Work mode; Mistral does not say which plans include it |
| Kimi and Qwen | Deep Research modes | In Kimi and Qwen Chat |
Limits and plan details change often. Check ChatGPT's pricing page, Google's Gemini plans and Claude's pricing page for the current allowances.
Why it matters at work
Deep research is at its best for questions that need many sources pulled together: a market scan before a pitch, how competitors price a product, what regulators have said about a topic, a briefing on a company before a meeting, or a first pass at a literature review. Work that would take an afternoon of tabs can come back as a draft in minutes.
It is not the right tool for a quick fact (a normal search is faster), for questions about your own documents only (a source-grounded tool such as Gemini Notebook, formerly NotebookLM, fits better) or for anything where you cannot check the sources. Three habits make the difference:
- Write a real brief. Say who the report is for, what decision it supports, the time period, the regions, which sources to prefer and which to avoid. A vague question produces a vague report.
- Edit the plan. Gemini shows its plan before it starts and lets you change it. Use that moment to narrow the scope.
- Check the citations that matter. Open the source behind every number or claim you will repeat. A link next to a sentence does not prove the page says that.
One safety point for connected tools: Anthropic's help centre warns that in research mode Claude can call connector tools without asking for approval each time, and recommends disabling any connector tools that write or change data before you run Research. Our guide on using AI at work safely covers connected tools in more depth.
An example you can try today
- Open Gemini (free) or ChatGPT and switch on Deep Research for a new chat.
- Paste the first prompt below with your own industry and competitors filled in.
- When you see the plan, or any questions it asks first, cut anything you do not need and add a date range such as "the last 12 months".
- While it works, write down three facts you already know are true about this market.
- When the report arrives, check whether your three facts appear and are correct, then open the sources for the two most important numbers.
That ten-minute check tells you how far you can trust this tool on this kind of question. Our guide on how to use ChatGPT deep research walks through the ChatGPT version step by step.
Common misconceptions about deep research
- "It has citations, so it is correct." Citations show where the AI looked, not that it read them correctly. Reports can still contain hallucinations attached to real links.
- "More sources means a better report." Fifty weak sources are worse than five strong ones. Tell it which sources to prioritise, such as official sites, regulators or peer-reviewed journals.
- "Deep research is just a longer web search." It plans, reads and reasons across results, which is why it takes minutes rather than seconds.
- "It works the same everywhere." The tools differ in which sources they can reach. Gemini can include your Gmail and Drive if you allow it, and Perplexity can search and cite Wiley journals and books.
- "Custom assistants can run it for me." Not always. Google says Deep Research does not yet work inside Gemini skills, the replacement for Gems.
Related terms
- Reasoning model: the kind of model that plans the research.
- AI agent: deep research is an agent that only reads and writes.
- RAG: retrieving sources before answering, the core idea behind it.
- Hallucination: the error you still have to check for.
- Context window: how much of the material the model can hold at once.
For the right tool by task, see the best AI for research, ChatGPT vs Perplexity, Gemini vs Perplexity and our research prompts.
Prompts to copy
Run a competitor pricing scan
Research how [competitor A], [competitor B] and [competitor C] price their [product category] for [customer type] in [region], using only information from the last 12 months. Prefer the companies' own pricing pages and reputable press over blogs. Produce: a comparison table of plans and what each includes, a list of recent pricing changes with dates, and three implications for a company that charges [our model]. Cite a source for every number and say plainly when you could not confirm something.
You get: A sourced pricing comparison you can check line by line
Brief yourself before a meeting
I am meeting [person and role] at [company] on [date] about [topic]. Research the company's last 12 months: strategy statements, results, leadership changes, products and any press coverage relevant to [topic]. Write a one-page briefing with five likely priorities for them and three smart questions I could ask. Cite every claim.
You get: A one-page meeting brief with sources
Check a deep research report
Here is a research report: [paste report]. List every factual claim that includes a number, date or quote. For each, say which cited source supports it and whether the source actually says that, quoting the relevant sentence. Mark anything unsupported.
You get: A claim-by-claim audit of the report