Before you start
AI apps look things up in two ways. Web search answers a question from a few cited pages. A research mode plans a search, reads many pages and writes a report. You pick the mode, and the price of picking wrong is time or usage limits.
Plans change quickly, so these are as of 5 November 2026. Research in Claude needs a paid plan and web search turned on. Deep Research in Gemini asks only that you are 18 or older and signed in, with daily limits. Researcher in Microsoft Copilot is for Microsoft 365 Premium and Pro subscribers, and for business users with a Copilot add-on license.
This lesson follows the help pages of Anthropic, Google and Microsoft. For ChatGPT, check OpenAI's help center for the search and research options your plan has. The habits below carry over.
Run a research job in five steps
The prompt below follows the five steps. It is our own sample. Replace the topic and keep the shape.
I manage a ten-person office and need to choose a shared password manager.
Research this. Use sources from the last two years, and prefer security reviews and official documentation over blog posts.
Give me:
1. A one-paragraph answer.
2. The three questions I should ask before choosing.
3. A list of sources, each with its link and one sentence on what it says.
Mark any claim you could not confirm in a source as "unconfirmed".- Write the question and what it is for. "Which password manager should a ten-person office use?" beats "password managers", because it says who decides and what they need.
- Choose the mode. For one fact, use web search. For a decision, use the research mode. In Claude, click the + button at the bottom left of the chat and choose Research. In Gemini, choose Deep Research from the Add Files menu in the text box. In Microsoft Copilot, open Researcher under Agents.
- Set the scope. Give a date range and the kinds of source to prefer. Microsoft says more specific prompts and a defined search scope give more relevant results, and Gemini lets you choose sources and edit its research plan before it starts.
- Ask for a source list. Request the link and one sentence on what each source says, and tell the app to mark what it could not confirm.
- Open the sources before you use the answer. Find the sentence behind every number you plan to quote.
Which tool does what
The same job has a different button in each app. The table lists where to find each mode and the one fact about it that surprises people.
None of these modes is two years old yet, and the model inside them keeps changing. Claude's Research help page now names a model released on 22 September 2026, and Research uses it whatever the chat is set to. A report you ran in the spring came from a different model than one you run today.
Gemini Deep Research
Rolled out to Gemini Advanced subscribers, in English on desktop and mobile web.
Source: Google: Try Deep Research and our new experimental model in Gemini (read November 5, 2026)
Claude Research
Launched with a Google Workspace integration, so one job could read the web and your work files.
Source: Anthropic (@AnthropicAI) on X: Research launch (read November 5, 2026)
Claude Opus 5.5
Research runs on it now, with Claude Opus 4.8 as the fallback, whichever model you picked for the chat.
Source: Anthropic: Claude release notes (read November 5, 2026)
| App | Quick lookup | Longer report | Good to know |
|---|---|---|---|
| Claude | Click +, then Web search. Claude searches when a question needs current information and cites its sources | Click +, then Research. It needs web search on and a paid plan | Research can use connected tools such as Google Workspace. Started on mobile, it searches only the web. It uses your limits faster than a normal chat |
| Gemini | An answer can carry a Sources button. No button means Gemini gave no links for that answer | Deep Research from the Add Files menu. Edit the plan, then select Start research | Reports usually take about 5 to 10 minutes. You can leave the chat and get a notification. Past reports stay only if Keep Activity is on |
| Microsoft Copilot | Copilot Chat suits quick answers and short summaries | Researcher, under Agents in the Microsoft Copilot app | It may ask clarifying questions first. Your administrator may need to enable it. Reports include cited sources |
Check what it found
A research report reads as if an expert wrote it, and that is the risk. Google's page on sources describes the Sources button as a way to dig deeper, and it adds that not every response includes sources. Treat a link as a lead, not as proof. Anthropic's web search page says to cross-check cited sources and to rely on authoritative ones for critical decisions.
Checking one claim takes about a minute. Do it for every number and every name that your decision rests on.
Find the link
Open the source the app cites for the claim. No link means the claim is unconfirmed.
Find the sentence
Search the page for the number or the name. The claim should be there in plain words.
Check the date and the author
An old page may be out of date, and a vendor's own page sells its own product.
Look for a second source
For anything that decides money or safety, find one independent page that says the same.
When the app can read your email and files
Research tools can read more than the web. Claude's page says Research searches connected tools such as Gmail, Google Calendar and Google Docs, and that it uses them automatically, without asking for approval each time. Gemini lets you add Gmail and Drive as sources once Workspace is connected, and Microsoft's Researcher reads files, emails, meetings and chats you can access.
That reach is useful, and it has a cost: a research job can open material you did not think about. Connect only what the question needs, and run a question about the open web in a chat with no connectors.
What Anthropic posted when it launched Claude Research
Anthropic launched Research on X on 15 April 2025, and the same post announced a Google Workspace integration. The pitch fit in one line: Claude now brings together information from your work and the web.
Those two arrived together on purpose, and that is why you decide what to connect before a run starts. The help page today adds what the launch post left out. Connector tools run automatically during research, with no approval asked each time. A run started in the mobile app searches only the web and skips your connectors, and no error tells you so.
Pick the place to start by the question. Ask about the open web from a chat with no connectors. Ask about your own work from the web or the desktop app, where the connectors come along.
Claude now brings together information from your work and the web.
A screenshot of X's own embed of the post, taken on 5 November 2026, with the time in UTC. Select it to open the post on X.
Common mistakes
- Taking a link as proof. A source can fail to support the sentence next to it, so open it.
- Asking a vague question. A report on "password managers" covers everything and decides nothing.
- Using a research mode for a one-line fact. Claude's page says research sessions use your limits faster, and Microsoft says Copilot Chat is better for quick answers.
- Trusting old pages. Ask for sources from a date range and look at the date on each one.
- Pasting the report into your document as the final text. Use it as notes, then write the conclusion yourself.
Try it yourself
- Pick a decision you face this month and write the question with its purpose.
- Run it with web search first, then with a research mode if your plan has one, and compare the two answers.
- Open the sources behind three claims and note which ones hold up.
- Ask the app: "What is the strongest argument against this answer?" and read what comes back.
Next: summarize documents
Research hands you pages to read. The next lesson, how to summarize a PDF with AI, covers the other half: you already have the document, and you need its main points without reading every page.
If you want your AI app to reach more of your tools, connect MCP servers shows how an app links to them, and are MCP servers safe covers the risks.
Check yourself
When is a research mode a better choice than plain web search?
Show answer
B. For a decision that needs many sources and a written report A research mode plans a search, reads many pages and writes a report, which costs more time and usage than a one-fact lookup needs.
You get a research report with ten links. What should you check first?
Show answer
D. Whether each claim is on the page it links to A link is a lead, not proof. Google says not every response includes sources, so you open each link and find the claim on the page.
Claude Research started in the mobile app skips your Google Workspace. Where should you start it instead?
Show answer
A. On the web or in Claude Desktop Claude's page says research started on mobile searches only the web and skips connectors without warning.
Gemini makes a research plan after you send a Deep Research prompt. What can you do with it?
Show answer
C. Edit the plan, then select Start research Google's page says you can edit the research plan before you continue with Start research.
Frequently asked questions
What is the difference between AI search and deep research?
Mostly depth and time. A search answers from the pages it found. A research mode keeps going: Claude's page says it runs several connected searches and chooses what to look into next from what it finds, then hands back a report.
Can AI make up sources?
It can cite a page that does not say what the answer claims. The fix is the four-move check above: open the link and find the sentence. Gemini's page adds that a missing Sources button means it gave no links for that answer, so check the claim another way.
Do I need a paid plan for AI research?
Not always. Claude reserves Research for paid plans, Gemini's Deep Research page lists no paid plan as a requirement, and Microsoft ties Researcher to a Microsoft 365 or Copilot license. Plans move, so check the page for your app.
Can AI research use my own email and files?
Yes, in some apps, if you allow it. Claude can use connected tools such as Google Workspace, Gemini can add Gmail and Drive once Workspace is connected, and Microsoft says Researcher respects the permissions you already have at work.
Should I cite the AI in my report?
Cite the pages you opened, not the chat. If your school or employer has a rule about disclosing AI use, follow it.
Is deep research always better than a normal search?
No. For a single fact it costs more time and usage than it is worth. Use web search, or a plain chat, and keep the research mode for questions that need many sources.
Sources
- Anthropic: Using Research on Claude https://support.claude.com/en/articles/11088861-using-research-on-claude
- Anthropic: Web search on Claude https://support.claude.com/en/articles/10684626-enabling-and-using-web-search
- Google: Use Deep Research in Gemini Apps https://support.google.com/gemini/answer/15719111?hl=en
- Google: Check responses and sources in Gemini Apps https://support.google.com/gemini/answer/14143489?hl=en
- Microsoft: Get started with Researcher in Microsoft 365 Copilot https://support.microsoft.com/en-us/topic/get-started-with-researcher-in-microsoft-365-copilot-e63ab760-f3de-4c47-ae87-dad601b0e9c4
- Google: Try Deep Research and our new experimental model in Gemini https://blog.google/products/gemini/google-gemini-deep-research/
- Anthropic (@AnthropicAI) on X: Research launch https://x.com/AnthropicAI/status/1912192384588271771
- Anthropic: Claude release notes https://support.claude.com/en/articles/12138966-release-notes

