Research Mode, Skills & Tools Basics

10 examples to get you started with Research Mode, Skills & Tools - 7 basic and 3 intermediate.

Prerequisites

• A Claude.ai account; Research mode, Agent Skills, and connectors are Claude.ai features, not something you install separately.
• Which of these three features you can see depends on your plan tier - Free, Pro, Team, and Enterprise differ in what's available.
• No prior setup is required to try any of the three - each example below starts from a normal conversation.

Basic Examples

1. Recognizing When a Question Needs Research Mode

Not every question needs the full multi-step research process - knowing the signal helps you reach for it at the right moment.

Good fit for Research mode:
"Compare how three specific vendors price their enterprise plans,
using their current public pricing pages."
 
Better as a normal conversation:
"What's a common way to structure a vendor comparison?"

• A question that needs comparison across multiple sources, or current information Claude can't be certain of from training alone, is a Research mode question.
• A question you could answer from general knowledge or a quick explanation doesn't need the extra time Research mode takes.
• If you're unsure, start with a normal conversation - you can always follow up by asking for a deeper, sourced investigation.
• This distinction gets its own full page later in this section.

Related: When to Use Research Mode vs a Regular Conversation - a fuller comparison of the two

2. Starting a Research Mode Query

Starting a research task begins the same way any Claude conversation does - by describing what you need.

Example starting message:
"Research the current landscape of employer-sponsored EV charging
benefits in the US. I want to know what's typical, what a few
named companies offer, and how it's usually structured."

• The clearer and more specific the question, the more useful the sub-questions Research mode ends up investigating.
• Research mode runs autonomously once started - you don't need to guide each step.
• Expect it to take noticeably longer than a normal reply; that time is spent searching and reading, not idling.
• You can keep working in other conversations while a research task runs.

3. Reading the Cited Report

Research mode's output is a structured report, not a short paragraph - reading it well means checking the citations, not just the summary.

A cited report typically includes:
- A synthesized answer to your original question.
- Supporting claims, each traceable back to a specific source.
- Notes on where sources agreed or disagreed, if relevant.

• Treat the citations as the point of the report, not decoration - they're what makes the answer checkable.
• If a claim matters to a decision you're making, open the underlying source before relying on it.
• A well-built report will tell you when sources disagreed, rather than silently picking one.
• This gets a full page of its own later in this section.

Related: Evaluating Research Mode Output: Sources and Confidence - how to judge what you're reading

4. Finding an Agent Skill Claude Already Uses

Agent Skills are folder-based instructions Claude loads on its own when a task matches what a Skill describes - you often don't have to invoke one by name.

Example: if a Skill exists for "formatting quarterly sales reports"
and you ask Claude to format your quarterly numbers, Claude can
recognize the match and load that Skill's instructions automatically.

• A Skill is really a packaged recipe: instructions plus a description that tells Claude when it applies.
• You don't need to know a Skill exists to benefit from it - Claude discovers relevant ones from your request.
• If a task feels like something you (or your team) do repeatedly, there's often value in checking whether a Skill already covers it.
• Skills work the same way across Claude.ai, Claude Code, and Cowork.

5. Asking Claude Directly to Use a Specific Skill

When you know the name of a Skill, you can ask for it by name instead of waiting for automatic discovery.

Example message:
"Use the meeting-notes-cleanup skill on the transcript below."

• Naming a Skill directly is useful when several Skills could plausibly apply and you want a specific one.
• Claude will tell you if no Skill by that name is available, rather than silently ignoring the request.
• This is also a good way to test a newly created Skill before trusting it to trigger automatically.
• Skills are meant to be reusable - once one exists, everyone with access to it can call on the same recipe.

Related: What Agent Skills Are and How Claude Uses Them - a deeper look at what a Skill actually is

6. Checking Which Tools Are Already Connected

Before delegating a task that depends on outside data, it's worth confirming which connectors are actually linked to your account or workspace.

Questions worth answering before starting:
- Is our team's Google Drive or Slack connected?
- Does this connector cover the specific folder or channel I need?
- Am I asking Claude to use a source it doesn't actually have access to?

• A connector has to be explicitly set up before Claude can search that source - it isn't automatic.
• Assuming a connector is live when it isn't is one of the most common reasons a task comes back incomplete.
• Admins typically manage which connectors are available at the team or org level.
• This gets its own pre-flight checklist later in this section.

Related: Connected Tools Checklist Before You Start a Task - the full pre-flight list

7. Asking Claude to Search a Connected Source

Once a connector is live, you can ask Claude to search that specific source as part of a normal request.

Example message:
"Search our Notion workspace for the current onboarding doc
and summarize the first-week checklist."

• Naming the source in your request (Notion, Slack, Drive) helps Claude know where to look rather than guessing.
• Results pulled from a connected source are typically cited back to the specific document or message they came from.
• This works whether you're in a normal conversation or a Research mode query that includes internal sources.
• Access is scoped - Claude can only see what your account is permitted to see through that connector.

Intermediate Examples

8. Combining Research Mode With a Connected Source

Research mode isn't limited to public information - when a connector is set up, its multi-source search can include your company's own knowledge.

Example message:
"Research how our team has historically handled Q4 budget
overruns. Use our Slack and Drive connectors, plus any public
best-practice context that's relevant."

• This mixes internal and external sources in a single investigation, which is more powerful than either alone.
• Citations from internal sources point back to the actual internal document or thread, the same as public citations point to a URL.
• This only works once the relevant connectors are already linked - check first rather than assuming.
• It's a good pattern for questions that need both "what does the outside world do" and "what have we actually done before."

Related: Connectors and Enterprise Search: Bringing Outside Data into Claude - how this extension of Claude's reach actually works

9. Letting a Skill Shape How Research Mode Reports Back

A Skill can standardize the format of a recurring research task, even though the underlying Research mode process stays the same.

Example: a "competitor-pricing-report" skill might specify that
every research report on competitor pricing should end in a
table with columns for plan name, monthly price, and source date -
so results are consistent every time someone runs that research.

• This is a good pattern for any research task your team runs more than once - a Skill keeps the output format consistent across runs and people.
• The Skill shapes formatting and structure; Research mode still does the actual planning, searching, and citing.
• Building a Skill like this is covered in a different section of this site, focused specifically on designing and packaging Skills.
• Once built, the Skill is reusable by anyone on the team with access to it.

10. Reviewing a Report Before Sharing It Onward

Before forwarding a Research mode report to a decision-maker, a short review pass catches the most common problems.

A quick pre-share review:
1. Do the citations point to real, independent sources - not just
   one source repeated?
2. Are any claims dated or time-sensitive in a way the report
   should flag?
3. Does anything in the report surprise you enough to double-check
   it yourself before it goes further?

• This is the human step that makes Research mode's citation trail worth having in the first place.
• A report that looks thorough can still be built on a narrow or repetitive set of sources - the review catches that.
• Treating this as a habit, not a one-off, is one of the practices covered in this section's best-practices page.
• The full framework for this kind of review lives on the page about evaluating Research mode output.

Related: Research Mode, Skills & Tools Best Practices - ten habits that build on these basics

Stack versions: Written against the Claude model lineup current as of ~June 2026 - Claude Fable 5, Claude Opus 4.8, Claude Sonnet 5 (the default), and Claude Haiku 4.5. Model names, pricing, and product features move quickly - verify current specifics at platform.claude.com/docs before relying on them.