# What is Gemini Deep Research? Concept TL;DR Gemini Deep Research is a mode in Google's Gemini app that researches a question for you. It writes a plan, searches and reads up to hundreds of websites, reasons over what it found and hands back a multi-page report with its sources linked, usually in 5 to 10 minutes. For a website, that means a report can name your page as a source, and it can reach your pages two ways: through Google Search, and by reading them directly. ## What it is Ask an ordinary chatbot a question and it answers in one breath. Deep Research is closer to handing the question to a research assistant who goes away, reads for a while and comes back with a written brief. Google calls it an agent, meaning it decides its own next step instead of waiting for yours. A request goes through four stages: - **Plan.** It turns your question into a research plan, a list of what it intends to find out, which you can change or approve before it starts. - **Search and read.** It runs searches, opens the pages that look useful and follows leads, across up to hundreds of websites. If you allow it, it can also read your own Gmail, Google Drive and Google Chat. - **Reason.** It works through what it has read, notices gaps and contradictions, and searches again where it needs to. You can watch its thinking as it goes. - **Report.** It writes a multi-page report with links to the sources it used. Google says this usually takes 5 to 10 minutes, and you can leave and be told when it is ready. It lives in the Gemini app on the web, Android and iOS, and it is included in the free plan with limits. The report opens in Canvas, where you can export it to Google Docs or turn it into an Audio Overview, a spoken summary of the findings. Developers can run the same agent through the Gemini API. Three of Google's own names blur into it. **Ordinary Gemini chat** answers in seconds, with no plan and no report. **AI Overviews and AI Mode** are Google Search features that answer on the results page, and AI Mode's Deep Search is the closest thing to Deep Research inside Search. **Deep Think** is a reasoning mode for hard maths and science problems, not a web researcher. Outside Google, ChatGPT and Perplexity each have their own version of the same idea. ## Why it matters - **A report names its sources.** Deep Research ends in a long report with its sources linked, and people reading a report like that check the claims that matter to them. Being one of the named sources is a citation somebody may actually follow, which is the whole idea behind [being cited by AI](https://metapilot.io/learn/what-is-geo). - **It reaches your site two ways.** By default it uses Google Search, so a page Google has not indexed is a page Deep Research is unlikely to find. It also reads pages directly while it works, and those visits show up in your server logs with `Gemini-Deep-Research` in the user agent. - **It reads the page you serve, not the one you meant.** What it takes away is what your HTML states. A price, an opening time or an author that a person works out from the layout is a guess for a machine unless the page says it plainly, which is what [structured data](https://metapilot.io/learn/what-is-structured-data) is for. - **Google publishes no rules for which sources it picks.** What it does say is that Deep Research weighs a diverse range of sources, favours authoritative ones and ties each claim to where it came from. So the part in your hands is the ordinary part: be indexed, be reachable, and state your facts plainly. ## How MetaPilot handles it The honest part first: **MetaPilot cannot make Deep Research cite you**, and nobody can promise that, because Google does not say how it chooses. What MetaPilot does is show you whether it is reading your site, and put the controls and the machine-readable facts in your hands. - **See when it reads you.** On a site connected through Cloudflare, MetaPilot's crawler analytics recognises `Gemini-Deep-Research` by name, alongside GPTBot, ClaudeBot and the rest, and shows how often it visits and which of your pages it reads most. - **Decide who gets in.** Allow or block it, and the Google-Extended token, from the dashboard, and MetaPilot writes the rule into its own section of your [robots.txt](https://metapilot.io/learn/what-is-robots-txt). If your connection does not offer the per-bot switch, the same lines go in MetaPilot's robots.txt editor. - **State your facts in a form machines take literally.** MetaPilot writes structured data for the pages your platform left out, so what a page is, who publishes it and what it offers is stated rather than inferred. Ready to see who reads your site? [See and control which AI bots crawl you →](https://metapilot.io/guides/monitor-ai-crawlers) ## Can you keep your site out of Deep Research? Partly. It reaches you two ways, and each has its own control with a different amount of certainty behind it. - **Through Google Search: the Google-Extended token.** Google says a `Disallow` for Google-Extended in robots.txt controls whether content it crawls may be used to train Gemini models and to ground answers in Gemini Apps. Deep Research is part of the Gemini app and uses Search by default, so this is the most likely lever. Two gaps: Google's wording never names Deep Research, and it says nothing about pages read directly. It does not affect your place in Google Search. - **Through direct reads: a rule for `Gemini-Deep-Research`.** This is the only rule aimed at the agent itself, and it comes with no promise. Google has not documented the agent, and its stated rule for fetches a person asked for is that they "generally ignore robots.txt rules". The exact lines are on our [Gemini-Deep-Research crawler page](https://metapilot.io/bots/gemini-deep-research). - **Not Googlebot.** Blocking Google's main crawler takes your site out of Google Search altogether, which costs far more than Deep Research ever could. Our take: leave it open. Deep Research only reads your page because someone asked a question your page might answer, and the report it writes names its sources. For most sites, being one of them is the goal. Related - [What is an LLM (and an answer engine)?](https://metapilot.io/learn/what-is-an-llm) - [What is GEO (Generative Engine Optimization)?](https://metapilot.io/learn/what-is-geo) - [What is robots.txt?](https://metapilot.io/learn/what-is-robots-txt) - [What is structured data?](https://metapilot.io/learn/what-is-structured-data) ## FAQ ### Is Gemini Deep Research free? Yes, with limits. It is included in the free Gemini plan, and Google's paid plans raise the limits. Google does not publish a fixed number of reports per plan, and a Deep Research report uses more of your allowance than an ordinary chat. At busy times free users can lose access to it for a while. ### How long does Gemini Deep Research take? Google says usually 5 to 10 minutes, and longer for complex questions. You do not have to watch it work: you can leave the chat and Gemini lets you know when the report is ready. ### Is Deep Research the same as AI Mode or AI Overviews? No. AI Overviews and AI Mode are part of Google Search and answer on the results page, built from Google's search index. AI Mode's Deep Search is the closest relative: it runs many searches and returns a cited report inside Search. Deep Research lives in the Gemini app, works from a plan you can change, and can also read your own Gmail and Drive if you allow it. ### Can I stop Gemini Deep Research from reading my website? Partly. A `Disallow` for the Google-Extended token is Google's documented control over Gemini using content from its crawl, though Google does not name Deep Research in it. A `Disallow` for `Gemini-Deep-Research` targets the agent's direct visits, but Google has not documented that agent, and fetches a person asked for generally ignore robots.txt. Neither affects your Google Search rankings.