Gemini Deep Research

Gemini Deep Research is the agentic research feature inside Google's Gemini app. Given a research question it generates a visible, editable multi-step plan, then autonomously browses dozens to hundreds of sources, synthesizes the findings and produces a multi-page report citing each claim back to its source. The plan can be edited before execution, so a researcher can correct course before the agent starts. Google states the only requirements are a Gemini account and being 18 or over, so the full agentic workflow can be run before any payment. Best fit: analysts, consultants and researchers who need multi-source synthesis on complex topics in one automated pass. Google Search is the default source, and Deep Research can also draw on a user's own Gmail and Drive, on uploaded files, and on Gemini Notebook notebooks, so it is not confined to the open web. Finished reports export to Google Docs for collaborative editing. Google publishes no MCP server for the Gemini app, re-verified August 2026 against the vendor's MCP path, its llms.txt and the mcp subdomain. Google's own MCP announcements cover Google Cloud, Gemini Enterprise and the Gemini API rather than the consumer app. Named gaps remain: no access to paywalled academic databases such as PubMed full text, Scopus or Web of Science, no Zotero or Mendeley citation management, and no scheduled or recurring research runs. All figures here are USD, read from Google's own United States storefront after switching region, because one.google.com geo-serves local currency. Deep Research runs on the free tier at standard limits. Google AI Plus at $4.99 per month adds 2x usage limits. Google AI Pro at $19.99 per month adds 4x limits, a one million token context window, and the Pro model for Deep Research specifically. Google AI Ultra is sold as a 5x tier at $99.99 and a 20x tier at $199.99 per month, adding Deep Think and report visuals such as charts and diagrams. The tier below is the one to understand: Plus runs the same Deep Research model as the free tier and carries no million token context, so it buys volume rather than capability. Limits are daily and concurrent request caps rather than a monthly report count. Google publishes no annual rate anywhere public, though its help center states Pro can be billed monthly or annually. Deep Research is not suited to systematic reviews needing peer-reviewed coverage, because it cannot reach paywalled journals or institutional databases. For academic-database-grounded and PRISMA-compliant workflows, Elicit ($49/mo billed annually) and Rayyan (free basic tier, paid from $4.99/mo) provide academic-native pipelines. ChatGPT Deep Research ($20/mo) is the closest direct competitor, and Perplexity AI ($17/mo billed annually) suits source-display-first workflows. Teams handling confidential material should note three things. Gemini activity auto-deletes after 18 months by default, adjustable to 3 or 36 months or never. Chats already sampled for human review are disconnected from the account and kept up to 3 years. Turning off Keep Activity to stop model training also disables Workspace Connected Apps. Google's SOC 2 Type II, ISO/IEC 27001 and HIPAA coverage names the Gemini app explicitly, but does so within Google Workspace and Cloud Identity scope, and HIPAA additionally requires a Business Associate Amendment. A consumer Google One AI plan carries none of those agreements, so those certifications describe the app as bought through Workspace rather than at $19.99. Current state Q3 2026: the paid model line is Gemini 3.1 Pro, Deep Research itself runs on 3 Pro, NotebookLM has been renamed Gemini Notebook, and Google AI Ultra now sells as separate 5x and 20x tiers. Independent review evidence for this listing is the parent Gemini profile on G2 rather than a Deep Research specific corpus.