Skip to main content

Deep Research

Deep Research is for when you need a lot of sources, not a quicker answer. Claude goes off, reads widely, and comes back with a structured brief. It takes minutes, and it uses your plan limits. Most questions don't need it.

We use Claude's research mode for this. A normal question with web search is enough when you need one fact or a quick check.

How to access it​

On paid plans that include Research (Pro and above). Check the current price at claude.com/pricing.

In the chat, switch on Research, or run /deep-research. This is not the effort control, and it isn't an "extended thinking" toggle. Effort is how hard one reply thinks. Research is a separate mode that goes out and reads.

The switch people miss: using Research for a question you could have answered from a file you already have. If the answer is in the folder or the transcript, don't research it.

When to use it​

Use it when you genuinely need a wide read. A rough line: 20 or more sources, or a question you couldn't answer by opening three tabs.

  • A market or a competitor, before a meeting that matters
  • A topic you need to get up to speed on: a regulation, a channel, a sector you're entering
  • Background before a negotiation, so you can check it, not so you can bluff

Don't use it for a definition, a rewrite, or anything your own files already cover.

The prompt​

GOAL
[The decision this brief has to support.]

CONTEXT
What I already know: [facts].
What I don't need repeated: [skip].
Who will read it: [role].

CONSTRAINTS
[Format: one page, a table, a briefing.]
Name the source for anything factual.
If a source doesn't say it, leave it out.
Only do this as research if you need many sources. Otherwise answer directly.

If anything is missing, ask me first.

Worked example.

GOAL
A one-page briefing I can read the night before I meet their CEO.

CONTEXT
I'm meeting the CEO of a mid-sized professional services firm next Thursday. About £30m turnover, Bristol. I know they've been growing through acquisition. I don't need a company history.

CONSTRAINTS
What they do, recent news, likely priorities, and 3 questions that show I've done the work.
Name sources. One page.
If you only need a couple of pages to answer this, don't run a long research job.

If anything is missing, ask me first.

Pro tips​

  • Say what you already know, or you'll pay to be told it.
  • Open the links before you repeat a claim in the meeting. A citation can be the wrong sentence from the right page.
  • Plan the questions in normal chat first. Research is the expensive step.

Limits​

It still misses nuance, and it still states a weak source with confidence. For anything you'll act on, check the few claims that would embarrass you if they were wrong.

It draws on the same usage limits as everything else. A research run on Fable, in a long chat, is an easy way to spend the week.

It doesn't replace a transcript of your own meeting, or a number in your own pack.

Where to start in 30 minutes​

  1. Pick one meeting in the next fortnight where being under-prepared would show.
  2. Write what you already know in five lines, so it doesn't research that.
  3. Run the briefing prompt with Research on, Sonnet or Opus, not Fable.
  4. Open every link you might quote.
  5. Save the prompt if this is a type of meeting you have often. Otherwise leave it as a one-off.