What you will be able to do
- Break a work task into its deliverable, its information needs and where that information lives before choosing any Claude capability
- Choose between web search, extended thinking and research based on how much information gathering a question needs
- Decide whether context belongs in a pasted chat, a Project, a connector or memory
- Combine code execution, artifacts, Skills and connectors for recurring or deliverable-heavy work
- Recognise when a task needs code-executed analysis of an uploaded data file rather than a written answer
Key concept
Capability fit — Each Claude capability is built for one kind of need: finding public information, reaching private data, holding background context, running a repeatable procedure or producing an editable deliverable. To choose well, break the task into those needs and give each one the capability built for it. The most powerful option is not automatically the right one.
1.Analyse the task before picking a capability
Capability selection starts with the task. Picking a tool comes after. Claude offers chat, Projects, artifacts, research, web search, code execution, Skills, connectors and memory. Each one solves a different problem, and each has its own cost: research uses your plan's limits faster, and a Skill takes time to build. If you choose a capability first and then fit the work to it, you usually end up with too much (a ten-minute research report for a one-line fact) or too little (a plain chat asked about last quarter's pricing).
A reliable approach has four steps, in this order:
- Define the finished deliverable: its format, its audience, and whether it is needed once or every month.
- List the information the task needs and where each piece lives: text you can paste, files on your machine, private tools like email or a CRM, or the public web.
- Match each need to the capability built for it, such as research for public sources, a connector for an internal tool, or code execution for a spreadsheet.
- Run a small first attempt and check the result before you commit to a bigger setup, such as a shared Project or a custom Skill.
The order matters because each step depends on the one before it. You can't say where information lives until you know what the deliverable needs. And it is wasteful to build a reusable Skill for a procedure you haven't yet seen Claude carry out once. The deliverable also tells you whether the work repeats. A one-off briefing and a monthly reconciliation can use the same data sources and still call for very different setups.
Code-executed analysis is the match when the answer has to be computed from data rather than recalled or reasoned out. Say a support lead uploads a CSV of 40,000 tickets and asks for median resolution time by region, plus a chart. Claude writes Python scripts for the analysis, runs them, and can return the visualization as an image file such as a PNG. Reading the file as pasted text and estimating would be slower and less reliable. The reverse also holds: simple arithmetic, a unit conversion or a well-known fact is answered directly, with no code run. For a borderline request, ask explicitly, for example "run code to verify this". Code execution and file creation must be turned on in Settings > Capabilities, and uploading XLSX files needs it as well.
Checkpoint 1 of 9· Check yourself
A colleague needs one fact before a meeting: the name of a competitor's current CEO. They turn on research because it is the most powerful option. Based on analysing the task first, what is the best assessment?
If you define the deliverable first (one fact, needed once), the task needs only a light capability. Choosing research because it is the most powerful produces more than the task needs and uses plan limits faster.
“research sessions can use up your limits faster due to Claude retrieving multiple sources and providing comprehensive responses”Source: support.claude.com
Checkpoint 2 of 9· Exam question
A project coordinator is planning how to approach a new work task with Claude and wants to choose the right capabilities for it. Which sequence best supports that choice?
Correct answer: A — Define deliverable, audience and frequency → List information needs and where they live → Match each need to a capability → Run a small first attempt
- A. The deliverable, its audience and how often it is needed set what any capability must achieve. Next, knowing where the information lives narrows the relevant capabilities, so each need can then be matched to one. A small trial comes last, to confirm the combination works before heavier investment such as a Skill.
- B. Listing information needs before defining the deliverable is misplaced. Without the finished format, audience and frequency, there is no basis for deciding which information matters, so the list may cover the wrong sources.
- C. Matching capabilities before listing where the information lives is misplaced. Whether research, a connector or a file upload fits depends on whether the data is pasted, private or public, so the match is made on guesses and may need redoing.
- D. Running the trial before matching each need to a capability is misplaced. The small attempt exists to confirm the chosen combination of capabilities, so trying it first tests nothing deliberate and may miss needs that a connector or research would have covered.
2.Finding information: web search, extended thinking or research
Claude has three capabilities for questions you can't answer by pasting something in. The choice between them depends on two things: how much gathering the question needs, and whether it needs current information at all.
| Capability | Best for | Example |
|---|---|---|
| Web search | Straightforward factual queries answered with one or two tool calls; you steer what to search | Recent news headlines, details about one specific company, last night's game result |
| Extended thinking | Complex reasoning that doesn't need recent information from the web | Optimizing an algorithm, debugging code, analysing an ethical question from several perspectives |
| Research | Comprehensive gathering needing five or more tool calls over 1-3 minutes, synthesized into an in-depth report | Comparing business competitors, updating outdated internal documents with web information |
Research is more than a longer web search. Claude runs multiple searches that build on each other and decides what to look into next based on what it finds. Then it brings the results together across sources, with proper citations, into a longer final report. Take a strategy associate briefing a COO on how three competitors changed their pricing and partner programmes over a year. That is a research job: many public sources, a time span, and a synthesis the associate can check against the citations. Other options fall short. Memory only holds what came up in earlier chats. A Project's instructions only contain what someone typed into them. And a chat with no web access can't see changes that happened recently.
Research has some practical limits:
- It needs web search turned on.
- It is available on paid plans.
- It uses your plan's limits faster than an ordinary chat.
- It runs on Claude Opus 5.5 whatever model you selected for the chat, falling back to Opus 4.8 when needed.
- In Claude Mobile it searches only the web and silently skips your connectors, so research that needs Gmail or Google Docs should start on the web or in Claude Desktop.
Checkpoint 3 of 9· Match them up
Match each capability, or combination, to the information task it fits best
Tap a term, then the definition that fits it.
The choice depends on how much gathering the question needs and whether it needs current information. Web search fits one or two lookups. Extended thinking fits reasoning that needs nothing from the web. Research fits gathering that takes many tool calls and ends in a synthesis. Using thinking and research together lets Claude plan before it gathers.
“Research is optimal for comprehensive information gathering requiring five or more tool calls over 1-3 minutes, creating in-depth reports by synthesizing information from multiple sources”Source: support.claude.com
Checkpoint 4 of 9· Check yourself
On a train, an analyst starts research in Claude Mobile. They ask Claude to compare competitor news with the team's planning notes in their connected Google Docs. The report arrives with no errors or warnings. What should the analyst conclude?
Research started in Claude Mobile searches only the web and skips connectors without warning. When the research depends on Gmail or Google Docs, it has to start on the web or in Claude Desktop.
“When you start research in Claude Mobile, Claude only searches the web and doesn't use your connectors.”Source: support.claude.com
Checkpoint 5 of 9· Exam question
A strategy associate must brief the COO tomorrow on how three competitors have changed their enterprise pricing and partner programs over the past year. No internal files are involved. Which approach gives the most useful first draft?
Correct answer: C — Use research mode so Claude searches many public sources and returns a synthesized, cited report that the associate then verifies
- A. Incorrect: memory only holds what came up in earlier conversations; the stem gives no sign those chats covered competitor pricing.
- B. Incorrect: Project instructions containing only company names add no pricing information, so Claude would have nothing to ground the list in.
- C. Correct: research mode is built for multi-source investigation of current public information and returns a cited synthesis, which the associate can check against the sources before presenting.
- D. Incorrect: training-data recall can be out of date, and the task is explicitly about changes over the past year, so current sources are required.
3.Bringing in your own context: chat, Projects, connectors and memory
For private or organisational information, the question is where it lives and how often it's needed. Small, one-off material belongs in the chat itself: paste it or upload the file. Material that many conversations reuse belongs in a Project. Material that lives in a live system belongs behind a connector.
A Project is a self-contained workspace with its own chat history and knowledge base. Documents uploaded to it give background to every chat inside it, and project instructions can set things like tone or the role Claude answers from. Think of a support team of twelve who ask questions every day about a 35-page escalation handbook and a set of email templates. A shared Project fits them: one place to upload the material, one place to replace it when the handbook changes each quarter, and on Team and Enterprise plans, sharing with view or edit permissions. On paid plans, when project knowledge approaches context limits, Claude switches to retrieval (RAG) mode, which expands capacity by up to 10x.
Connectors handle information that lives in a live system rather than a document you could upload. They let Claude read from apps like Google Drive, Slack or Linear and take actions in them, such as creating an issue or sending a message. Claude inherits each person's permissions in the connected service, so a connector can never reach a file the user couldn't open themselves. Admins can narrow this further by setting tools to Always allow, Needs approval or Blocked. Say a hiring manager wants interview invitations drafted into their email drafts folder, using their calendar availability for the proposed times. They need email and calendar connectors, because both the input (free slots) and the output (drafts) live in those tools.
Memory solves a different problem: continuity about you and your work across conversations. Claude saves memory as individual topics while you chat. If you mention that a deadline moved, your next conversation already knows. Each project keeps its own memory. Memory is not a source of facts about the outside world. It holds only what came up in your chats, so it can't stand in for research on public information.
Only the deadline. Memory holds topics saved from the user's own chats. The competitor's current price is public information that changes over time, so it needs web search.
Checkpoint 6 of 9· Match them up
An operations lead has four different pieces of private context to give Claude. Pair each one with the way of bringing it in that fits where it lives and how often it's needed.
Tap a term, then the definition that fits it.
Small one-off material belongs in the chat. Material that many conversations reuse belongs in a Project's knowledge base. Data that lives in a live system such as email or a calendar needs a connector, which can both read it and act in it. Memory carries what came up in your own chats into later ones.
“Connectors let Claude access your apps and services, retrieve your data, and take actions within connected services.”Source: support.claude.com
4.Producing deliverables and repeating procedures: code execution, artifacts and Skills
The deliverable you defined at the start points to the output capabilities. Code execution lets Claude create and work with files directly in the conversation. It can build Excel models with working formulas, perform advanced analyses on uploaded data, and produce Word, PowerPoint and PDF files you can download. Choose it whenever the answer is a computation or a file, not a paragraph.
An artifact is anything Claude makes that you'd put in front of someone: a document, a deck, a dashboard, a small interactive tool. It opens beside the conversation, where you can edit it, return to it and share it. Claude creates one when the content is substantial and self-contained, typically over 15 lines, and likely to be reused outside the chat. Templates such as Docs, Slides, Data and Sheets give a starting point for shared work. Artifacts need Cloud code execution and file creation turned on, so if that setting is off, no artifacts appear.
That capability is Skills. A Skill is a folder of instructions, scripts and resources that teaches Claude how to do a specific task the same way every time. When you ask for something, Claude reviews the available Skills, loads the relevant ones and applies their instructions. Only the Skills it needs get loaded, which keeps the context window from overloading. Anthropic provides Skills for creating Excel, Word, PowerPoint and PDF files. Your organisation can write custom ones in Markdown, for example for applying brand guidelines or following a team's data-analysis workflow. Owners on Team and Enterprise plans can provision Skills to everyone. Like artifacts, Skills require code execution.
| Capability | What it provides | When it applies |
|---|---|---|
| Projects | Static background knowledge | Always loaded when you start chats within the project |
| MCP connectors | Access to external services and data sources | Whenever the connected tool is relevant |
| Custom instructions | Broad preferences | All your conversations |
| Skills | Procedural knowledge: how to complete a specific task or workflow | Load dynamically only when relevant, everywhere across Claude |
Artifacts are what Claude produces. Skills are how it produces them. A branded deck is an artifact, and the Skill is what makes every deck come out branded.
Checkpoint 7 of 9· Check yourself
Every month, a finance lead reconciles vendor invoices following a fixed eight-step procedure, and they have written the steps into a Project's instructions. Colleagues who ask for the reconciliation in ordinary chats outside that Project get inconsistent results. What change best fits the need?
A Project supplies static background knowledge only to chats inside that Project. A Skill supplies a procedure that loads when a matching task appears, anywhere in Claude. Custom instructions hold broad preferences for every conversation, and connectors give access to data and tools, not the procedure for using them.
“Skills provide specialized procedures that activate dynamically when needed and work everywhere across Claude.”Source: support.claude.com
Checkpoint 8 of 9· Exam question
A team lead asks which capability decides whether Claude can follow a reusable, packaged set of instructions and files for a specialized task, such as producing a branded slide deck, when a request matches it. What is it?
Correct answer: B — Skills
- A. Incorrect: artifacts are the rendered outputs shown beside the chat, not packaged procedures.
- B. Correct: Skills bundle instructions and resources that Claude can load when a task matches, which makes specialized, repeatable outputs consistent.
- C. Incorrect: connectors give Claude access to external tools and data sources; they do not define how a task should be done.
- D. Incorrect: memory stores user preferences and context between chats rather than packaging task procedures and files.
5.Combining capabilities for a real task
Most real work needs more than one capability, and the right combination comes from the four-step breakdown. Take a finance analyst's monthly task: pull the expense export from Drive, run the company's reconciliation checks using a fixed procedure and template, and produce a formatted workbook. The data lives in Drive, so it needs a connector. The procedure is fixed and repeats every month, so it belongs in a Skill. The output is a spreadsheet with working formulas, so it needs code execution. A Project full of reconciliation notes would give background but would not run the procedure the same way each time. And research has nothing to do here, because none of the information is on the public web.
Connectors and Skills complement each other: the connector gives Claude access to a tool, and the Skill teaches Claude how to use that tool effectively. That is why the Skills Directory includes partner Skills built to work alongside their own connectors. Following the fourth step, the analyst would first run one month's reconciliation in a normal chat and check the workbook by hand. Packaging the procedure as a Skill only makes sense once that first run is right.
Checkpoint 9 of 9· Check yourself
Every week, a sales-ops manager pulls pipeline records from the team's CRM. They apply the same fixed summarising and branding procedure each time and deliver a PowerPoint deck to download. Which setup covers every need in this task?
Each part of the task needs its own capability. The records are in a live system, so they need a connector. The fixed procedure repeats every week, so it belongs in a Skill. The output is a file, so it needs code execution. Project notes give background but don't run the procedure the same way each time. Memory holds only what came up in earlier chats, not live CRM data.
“You can use both together: MCP connections give Claude access to tools, while skills teach Claude how to use those tools effectively.”Source: support.claude.com
Sources
(9)
Every claim above is drawn from one of these pages, quoted as it was written on the date shown.
- 1.https://support.claude.com/en/articles/11095361-when-should-i-use-web-search-extended-thinking-and-researchOfficial docs
“Understanding when to use each option—web search, extended thinking, and research—can help you get the best results for your specific needs.”
↩︎ Analyse the task before picking a capability“The research feature can synthesize information across multiple sources with proper citations and produces a longer final report.”
↩︎ Finding information: web search, extended thinking or research“Extended thinking shines when tackling complex reasoning tasks that don't require recent info from the web”
↩︎ Finding information: web search, extended thinking or research“Understanding when to use each option—web search, extended thinking, and research—can help you get the best results for your specific needs.”
↩︎ Key concept“Research is optimal for comprehensive information gathering requiring five or more tool calls over 1-3 minutes”
↩︎ Exam trap 1“Web search is best for straightforward, factual queries that can be answered with one or two tool calls”
↩︎ Prediction“Research is optimal for comprehensive information gathering requiring five or more tool calls over 1-3 minutes, creating in-depth reports by synthesizing information from multiple sources”
↩︎ Checkpoint - 2.
“research sessions can use up your limits faster due to Claude retrieving multiple sources and providing comprehensive responses.”
↩︎ Analyse the task before picking a capability“It runs multiple searches that build on each other and decides what to look into next based on what it finds.”
↩︎ Finding information: web search, extended thinking or research“You must have web search turned on for research to function.”
↩︎ Finding information: web search, extended thinking or research“When you start research in Claude Mobile, Claude only searches the web and doesn’t use your connectors.”
↩︎ Exam trap 2“research sessions can use up your limits faster due to Claude retrieving multiple sources and providing comprehensive responses”
↩︎ Checkpoint“When you start research in Claude Mobile, Claude only searches the web and doesn't use your connectors.”
↩︎ Checkpoint - 3.
“Claude can create Python scripts for data analysis.”
↩︎ Analyse the task before picking a capability“Claude can execute code to create and work with files directly in your conversations.”
↩︎ Producing deliverables and repeating procedures: code execution, artifacts and Skills“You can create financial models in Excel with working formulas”
↩︎ Combining capabilities for a real task - 4.
“Simple arithmetic and well-known math facts”
↩︎ Analyse the task before picking a capability - 5.
“Projects allow you to create self-contained workspaces with their own chat histories and knowledge bases.”
↩︎ Bringing in your own context: chat, Projects, connectors and memory“When your project knowledge approaches context limits, Claude seamlessly enables RAG mode to expand capacity by up to 10x”
↩︎ Bringing in your own context: chat, Projects, connectors and memory - 6.https://support.claude.com/en/articles/11176164-use-connectors-to-extend-claude-s-capabilitiesOfficial docs
“Connectors let Claude access your apps and services, retrieve your data, and take actions within connected services.”
↩︎ Bringing in your own context: chat, Projects, connectors and memory“Claude inherits each person's permissions from the connected service.”
↩︎ Bringing in your own context: chat, Projects, connectors and memory - 7.https://support.claude.com/en/articles/11817273-use-claude-s-chat-search-and-memory-to-build-on-previous-contextOfficial docs
“Each project has its own memory.”
↩︎ Bringing in your own context: chat, Projects, connectors and memory“Claude saves memory as a set of individual topics as you chat, rather than summarizing conversations after they end.”
↩︎ Exam trap 4 - 8.https://support.claude.com/en/articles/17153992-what-are-artifacts-and-how-do-i-use-themOfficial docs
“An artifact is anything Claude makes for you that you'd put in front of someone”
↩︎ Producing deliverables and repeating procedures: code execution, artifacts and Skills“Artifacts require Cloud code execution and file creation”
↩︎ Producing deliverables and repeating procedures: code execution, artifacts and Skills - 9.
“Skills are folders of instructions, scripts, and resources that Claude loads dynamically to improve performance on specialized tasks.”
↩︎ Producing deliverables and repeating procedures: code execution, artifacts and Skills“When you ask Claude to complete a task, it reviews available skills, loads relevant ones, and applies their instructions.”
↩︎ Producing deliverables and repeating procedures: code execution, artifacts and Skills“MCP connections give Claude access to tools, while skills teach Claude how to use those tools effectively.”
↩︎ Combining capabilities for a real task“Skills provide specialized procedures that activate dynamically when needed and work everywhere across Claude.”
↩︎ Exam trap 3“Skills provide specialized procedures that activate dynamically when needed and work everywhere across Claude.”
↩︎ Checkpoint“You can use both together: MCP connections give Claude access to tools, while skills teach Claude how to use those tools effectively.”
↩︎ Checkpoint