You have ninety minutes to produce a briefing on how three competitors have priced a product over the past year, drawing on their sites, recent coverage and analyst commentary. You need something you can defend in a meeting. Which approach fits?

The principle — matching a capability to a task means asking two questions: is the information external and recent, and does the answer need one fact or many stitched together? The pairing decides the tool, and getting it wrong wastes either time or credibility.

Why the key is correct — Anthropic describes research as optimal for comprehensive information gathering requiring five or more tool calls over one to three minutes, creating in-depth reports by synthesising information from multiple sources across the web and your integrations. This task is that description almost line for line: three companies, three source types, a year of movement, and an output that must hold together as a document. It is also the only choice that produces citations across the whole briefing, which is what "defend in a meeting" actually requires — every claim traceable to something a colleague can open.

Why the others are wrong — a plain chat rests on knowledge that stops at a cutoff, so a year of pricing changes is precisely the material it lacks. Extended thinking is documented for complex reasoning that does not require recent information; here the reasoning is easy and the facts are the hard part, so it solves the wrong half. And a single web search suits a straightforward query answerable in one or two calls, which this is not — extrapolating from one result to three competitors over a year manufactures the claims you will be asked to support.

Remember this — one fact, search. No external facts, thinking. Many facts that must hold together as an argument, research.

Sources — Anthropic's guidance on choosing between web search, extended thinking and research.