What You Need to Know
Claude offers four ways to handle work. Ordinary chat is a single conversation. The Help Center describes Projects as self-contained workspaces with their own chat histories and knowledge bases. Research is for questions whose answer must be assembled from several outside sources, and it returns a cited report. An artifact is standalone content displayed in its own window, beside the conversation. Most people pick a feature by habit. Identify the kind of task instead, before writing any prompt.
This lesson uses one made-up example throughout: a recruitment agency. One recruiter there has four jobs. She needs one reply to a client's complaint. Each week she writes client updates from the same fee list and house style. She wants to know how other agencies handle missed interviews, with sources her manager can check. She revises an interview guide for her colleagues every quarter.
Four questions sort the jobs:
- Is the work one-off or recurring?
- Does it stand alone, or does it need the same standing context (background information that stays the same) again and again?
- Is it consumed once, or revised and reused?
- Can Claude answer from what you supply? If not, does the task need one fact from outside, or a report built from many outside sources?
Each feature exists for one kind of task.
Ordinary Chat Is the Default
A single question, a single draft or an analysis of a document attached during the chat needs no workspace. It also needs no agentic search (Research, which builds a cited answer from several sources) or separate artifact window. Ordinary chat handles such a task immediately. The complaint reply is this kind of task.
Projects: Recurring Shared Context
A Project has two assets that persist across chats. Its instructions shape Claude's responses in every chat inside the workspace. Its knowledge base holds uploaded material Claude uses as background for those chats. You configure both assets once, and then both apply to every chat in the Project. The signal is a need for the same background documents, standing preferences and working material across many conversations. Re-pasting the same context into chat after chat is a Project's work done by hand. Projects exist to end that workaround. A Project would hold the fee list and house style the recruiter re-pastes each week.
Research: Cited Multi-Source Synthesis
In Research, Claude works agentically: it runs multiple searches, and later searches use what earlier searches found. Its report is synthesised (combined from all the findings) and includes citations so that the findings can be checked. The signals are a survey of a whole field, a brief on how others do something, and an overview of the current state. Another signal is an audience who will want to see where the findings came from. The missed-interview question fits Research. Research is not a quality upgrade for ordinary questions. A task answerable from material you already hold gains nothing from Research, because nothing needs gathering.
Web Search, Research or Extended Thinking
Three tools can look alike, but each one does a different job. Choose the tool by what the answer needs.
| Tool | What it does | When to use it | Agency example |
|---|---|---|---|
| Web search | Claude searches the web in an ordinary chat and answers with the sources it found. It suits queries that one or two lookups can answer. | You need one current fact from outside. | Has the government changed the minimum wage this month? |
| Research | Claude runs five or more searches. It searches the web and the apps you connect to Claude, such as your email and calendar. Then it writes a report with citations. | You need a report built from many sources. | How do other agencies handle missed interviews? |
| Extended thinking | Claude works through the problem at more length before it answers. It does not search. | You need careful reasoning about material you already have. | From her own interview records, which reminder schedule would have reduced last year's missed interviews the most? |
Research is available only on paid Claude plans. On the newest models (checked 13 November 2026), thinking is always on in Claude. The effort setting controls how much Claude thinks. Lesson 3.3 explains effort.
Artifacts: Work with Its Own Life
Claude creates an artifact when the output is significant, self-contained and likely to be edited, iterated on or reused outside the conversation. That last condition is the dominant signal: the work is used beyond the chat that produced it. Examples are a reference sheet revised every quarter and a plan that will be shared and updated. Each should exist as one evolving version, not as fragments scattered through a conversation history. You iterate an artifact in one of two ways. You ask Claude to change it, or you edit an earlier message. Editing an earlier message creates a branch of the chat (a second line of the conversation from that message) with artifacts of its own. You can return to an artifact and share a finished one. The interview guide fits this signal.
One Dominant Signal
Real work mixes the features. A Project can contain chats that produce artifacts. A Research report can be the basis of an artifact document that is revised each quarter. The agency's missed-interview report might be the basis of a missed-interview policy artifact.
In the exam
This objective tests whether you pick a feature by reading the task. The exam does not test the combinations. Each question presents one piece of work with one dominant signal: recurring context, multi-source citation, iterative deliverable or simple one-off. The wrong answers are always real features applied to the wrong kind of task. Do not dismiss ordinary chat, because the exam treats it as a right answer. When a question's dominant signal is a simple one-off, ordinary chat is the professional answer, however capable the other features are. The exam is willing to make the plainest option the right answer.
Writing a Project's instructions and curating its knowledge base belong to Domain 5. In this objective, the exam tests whether you recognise the recurrence. Checking a Research report's citations belongs to Domain 2. Choosing between an inline answer, a table and a document is Domain 2's format question. In this objective, an artifact is the place for iterative, reusable work.
Key Concept
Find the dominant signal first, and then choose the feature that matches it. Recurring shared context means a Project. A cited multi-source answer means Research. One current fact from outside means web search in an ordinary chat. An output that is revised and reused means an artifact. A self-contained one-off means ordinary chat.
Exam Traps
Treating an artifact as the place to keep standing context for new chats
An artifact is a container for one evolving output; it does not give new conversations standing context. A Project's instructions and knowledge base apply from the start of every chat, so nothing needs pasting in.
Running everything through a Project or Research to maximise quality
The most elaborate feature is not the safest choice. A simple one-off gains nothing from a workspace or a multi-source report. Ordinary chat is a legitimate answer in this objective. The dominant signal decides, not the feature's prestige.
Treating re-pasting the same context into every chat as diligence
The manual workaround repeats a Project's setup work every single time instead of once. When a scenario describes the same documents and preferences supplied chat after chat, the answer moves that work into a Project.
Practice Scenario
An office manager at a software company has four pieces of work: the monthly agenda-and-minutes cycle for the management meeting, which always needs the company's meeting rules, attendee list and format conventions; a one-off briefing on how five other software companies run remote working, with sources the managers can check; a single employee's question about an expenses deadline; and an emergency-contacts sheet that is revised every quarter and shared with all staff. She wants nothing more elaborate than each piece needs, no Project for work that never recurs, and nothing re-supplied by hand. Which plan matches feature to work?
Build Exercise
Audit Your Feature Choices
What you'll learn
- How to identify the kind of task before choosing a feature
- How to name the dominant signal that picks each feature
- How to spot the manual workaround in your own working pattern
- How to decide when a recurring task deserves a Project
- List your five most recent Claude conversations from your own work
Why: Real conversations expose real habits; an invented list makes the results look better than they are.
You should see: Five one-line descriptions of actual tasks you ran recently.
- Label each with the feature it should have used: ordinary chat, a Project chat, Research, or an artifact for the output
Why: The labelling forces the selection decision this objective tests, one task at a time.
You should see: A feature label against each of the five tasks.
- Name the dominant signal that decides each label: recurring shared context, multi-source cited synthesis, iterative shareable deliverable, or simple one-off
Why: Signals, not feature names, are what exam questions give you; practising how to match signals to features is the point.
You should see: One named signal beside each feature label.
- Find one manual workaround in your own pattern, such as context re-pasted across chats or a document that is revised over time, hidden in an old conversation, and name the feature that ends it
Why: The workaround is the bad practice that the exam most often presents as careful work; finding your own makes it easier to see in exam questions.
You should see: One workaround described in a sentence, with the feature that replaces it.
- If a recurring piece of work deserved a Project, note it for setup, then stop; configuring it well is a later skill
Why: This objective ends at the selection decision; instructions and knowledge curation belong to Domain 5.
You should see: A note naming the Project you would create and the recurring work it serves.
Sources
- What are projects? — Anthropic
- Use research on Claude — Anthropic
- Analyze campaign performance — Anthropic
- What are artifacts and how do I use them? — Anthropic
- When should I use web search, extended thinking, and research? — Anthropic
- Change the model, effort, and thinking settings — Anthropic
- Claude Certified Associate – Foundations Exam Guide — Anthropic