What you’ll know by the end of this check
- What Research mode actually does under the hood — and why it’s different from just enabling web search
- The decision matrix: Research vs. web search vs. extended thinking vs. Enterprise Search
- How to write a Research prompt that gets a useful report, not a rambling summary
What Research mode is
Research is Claude operating as an autonomous investigator. You give it a question; it decides what to search, reads what it finds, decides what to search next based on that, and keeps going until it has enough to synthesize a comprehensive answer. You see the progress as it works. At the end, you get a structured report with citations.
The key word is agentic. This isn’t a web search that returns a list of links. Claude is reasoning about what it needs to know, pursuing leads, filling gaps, cross-referencing sources — the same thing a skilled analyst does over a few hours, compressed into 5-45 minutes. Extended thinking activates automatically when Research is on, so the planning step is baked in.
Citations are included and linked. That’s not a nicety — it’s what makes Research output trustworthy enough to act on.
The decision matrix
This is worth having in your head before you reach for Research:
| Situation | Use |
|---|---|
| Multi-source investigation, need a comprehensive report | Research mode |
| Quick specific fact (stock price, a company’s address, a date) | Web search |
| Complex reasoning on a problem that doesn’t need external info | Extended thinking |
| Question about internal docs, Slack threads, company knowledge | Enterprise Search |
Research is overkill for quick lookups and the wrong tool for internal-only questions. The mistake most people make is using regular web search when they actually want Research — then being disappointed by shallow answers.
One powerful move: turn web search off inside Research mode to do internal-only synthesis. “What did our team discuss about the Q3 launch across Slack and Docs?” — no public web results, just your connected tools.
How to write a Research prompt that actually works
Research can take up to 45 minutes for complex questions. Writing a better prompt upfront is leverage — you’re not going to interrupt and redirect halfway through.
Four things that improve Research outputs:
- Specify the goal, not just the topic. “Tell me about the EV market” → bad. “Analyze the EV battery market: key players, technology trends, supply chain risks that affect investment decisions” → good.
- Specify the structure you want. Claude organizes its report around the sections you provide. Tell it the sections.
- Include constraints. Budget range, geography, timeline, industry vertical — anything that narrows the search space.
- Ask Claude to help you write the prompt first. Not a joke. If you’re not sure how to frame a complex research question, ask Claude to write you a better Research prompt before you enable the feature. Takes 2 minutes and saves you from a 30-minute run in the wrong direction.
Things to try right now (5 minutes)
Find a question you’ve been putting off because it would take too much research time. Enable Research mode (blue button, bottom left of chat — web search must also be on). Write a prompt with a goal, structure, and one constraint. Start it. Watch the first few steps of the plan it produces. You don’t have to wait for the full report — seeing the plan alone is useful.
The canonical version
Full official lesson at anthropic.skilljar.com/claude-101/383399 — includes the “Researching with Claude” video and the full multi-step process breakdown.
Ready to verify this check?
Finish the official lesson, then come back and mark this check verified on your flight log.