Choosing the right Claude feature

These are not mutually exclusive, which is what makes the exam items feel close. A single afternoon can involve a Project that contains the source documents, a Research run inside it, and an artifact as the deliverable. The question is always which feature the scenario turns on — and that is decided by one property each.

Chat: the default, and often the right answer

An ordinary conversation is the right tool for anything self-contained: a question, a rewrite, a calculation, a one-off summary of a file you will not need again. It is worth saying plainly because a common exam distractor is to create a Project for a task that happens once. A Project is organisational overhead — a workspace, a knowledge base, instructions — and overhead only pays back when something repeats.

Chat is not memoryless, though, and that changes the calculation slightly. On paid plans Claude can search and reference your previous conversations, so you can ask what you discussed about a topic earlier. Separately, memory builds a set of topics as you chat — your role, your ongoing work, your preferences — and is on by default for Free, Pro and Max plans, while Team and Enterprise have it off by default until an owner enables it. Chats outside a project and chats inside one are searched separately, and each project has its own memory space. Incognito chats are excluded from both: they are not saved to history or memory and are not referenced later.

Projects: when the same context comes back

A Project is described as a self-contained workspace with its own chat history and knowledge base. You upload the documents, text or files the work depends on, and you write project instructions that tailor Claude’s responses for every conversation inside it. That second part is what people under-use: the standing brief — who the audience is, what format the output takes, what must never be claimed without evidence — lives in the instructions instead of being retyped into every message.

Two practical facts follow. First, capacity: on paid plans, as a project approaches the context limit, retrieval-augmented generation expands capacity by up to ten times, so a knowledge base can be larger than a single conversation could hold, with the relevant parts retrieved rather than everything loaded. Second, collaboration: sharing projects is a Team and Enterprise capability, with “Can view” and “Can edit” permission levels and a “Shared with me” tab. Free accounts can create projects, up to five. Setting them up properly — instructions, knowledge sources, connectors, keeping them current — is Domain 5; this objective is only about recognising when a Project is the right answer.

Web search and Research: when the answer is outside

Both of these exist because a model’s training data has a cutoff and your files do not contain the outside world. They differ in depth. Web search is enabled from the plus button in the chat window; Claude invokes it automatically for topics that benefit from current information, and you can ask for it explicitly or tell it not to. Responses come back with citations and source links, and on Team and Enterprise accounts an owner must enable it for the organisation first.

Research is the heavier instrument. Selected from the same plus menu, it runs multiple searches that build on each other, exploring different angles of the question rather than answering from one lookup, and returns a thorough answer with citations designed to be easy to check. With Google Workspace connected it can draw on internal context — Gmail, Calendar, Docs — alongside the public web. It requires web search to be on, is available on paid plans, and while it counts towards usage limits in the same way as a normal conversation, it can consume them faster because it retrieves many sources.

Artifacts: when the output is a thing, not a reply

An artifact is described as anything Claude makes that you would put in front of someone — a document, a design, a deck, a dashboard or a small interactive tool. It opens beside the conversation, and you can edit it, come back to it and share it with a link. Claude creates one automatically when the content is significant and self-contained (typically over fifteen lines), when it is something you are likely to edit, iterate on or reuse outside the conversation, when it stands on its own without the surrounding chat, and when you are likely to refer back to it.

Those criteria are worth reading as a test you can apply yourself. A three-sentence answer to a question belongs inline; a policy page that four people will comment on belongs in a document you can version and link. Artifacts support editing through chat or, for Markdown, directly in place, switching between versions, having several open at once, and exporting by copy, download or link. Deciding between an artifact, an inline answer and a structured format for a given audience is objective 2.6 — here the point is simply that “the output is an editable, shareable object” is a feature choice.

Traps the wrong answers are built from

You should now be able to

• Decide between an ordinary chat and a Project from whether the context recurs.
• Describe what a Project holds — knowledge base, instructions, its own history and memory — and who can share it.
• Choose between web search and Research based on whether the question needs one lookup or several angles.
• Recognise when a question belongs to your own documents rather than the web.
• Apply the documented criteria for when content becomes an artifact rather than an inline reply.
• Combine features in one workflow without treating any of them as the default answer.