How to Use Claude Projects in 2026

Open Projects in Claude, create a project, add the documents and instructions that should apply across its chats, and test the setup with a fresh conversation. Put recurring context in project knowledge rather than assuming a detail from one chat will appear in another. Check who can access the project before uploading files.

This guide follows Anthropic's project creation and management help and project overview, checked October 30, 2026. Anthropic also describes a newer project version in beta with staged availability. The instructions below focus on the documented project-knowledge workflow; check which version and controls your account actually has.

Toolradar data: its AI assistants snapshot, dated October 25, 2026, tracked 1,572 products across a broad category. That is not a count of project features or a measure of Claude's accuracy. A useful project still depends on the quality of the files and instructions you supply.

What is a Claude project?

Anthropic describes projects as self-contained workspaces with their own chat histories and knowledge bases. A project can hold documents and standing instructions for a recurring task, such as reviewing a product brief, drafting a weekly update or comparing research notes. Each new chat can then use the project knowledge instead of requiring you to paste the same approved background every time.

A project is not an automatic guarantee of factual accuracy or complete recall. In the standard workflow described by Anthropic's management guide, context is not shared across chats unless it is added to project knowledge. Put a fact that must be available in the shared knowledge or instructions, then verify the answer against the source. Do not assume a decision discussed only in another chat has become shared project knowledge.

Step 1: Create a project

Anthropic's help says projects are available to all users, including free accounts, with a maximum of five projects for free users. In Claude, open Projects in the left sidebar or visit claude.ai/projects, then select New Project. Give the project a name and description.

The name is useful to people organizing the work, but Anthropic's guide says Claude does not have access to the project name and description as task context. If a fact matters to the response, add it to the actual project instructions or knowledge rather than relying on the title.

If your organization uses Team or Enterprise, the setup may offer visibility choices. Anthropic describes a private option and, where an administrator enables it, an option to share more broadly with the organization. Check the current account and administrator settings before choosing an audience.

Step 2: Add project knowledge

On the project's main page, Anthropic describes a project knowledge area with a + control. Use it to add relevant documents, text files or code snippets. The guide says uploaded project knowledge is used across chats within that project.

Start with a small set of documents that are current and appropriate to share. For a product-brief project, that could be an approved product description, a current pricing sheet and a style guide. Give files clear names and dates so a user can tell an active document from an old draft. Do not upload confidential files unless the account, recipients and organizational policies permit it.

Before relying on the setup, open a new chat in the project and ask a question whose answer is in one file. Check that the answer matches the document. Then ask what is missing from the file; a useful workflow should leave an unsupported answer unresolved rather than inventing it. We did not perform this product test for the article; it is a suggested check for your own account.

Step 3: Write standing instructions

Anthropic's management guide says to select Set project instructions, write the desired behavior and save it. The instructions are intended for chats in the project. Keep them specific enough to evaluate but general enough to apply to more than one request.

Here is an example for a team reviewing product briefs:

Use the project's current product description as the primary context.
When I provide a draft, identify unsupported claims and unclear terms.
Separate confirmed facts from suggestions. If a price or date is missing,
write "not specified" rather than guessing. Return a short issue list with
an excerpt and a suggested revision for each issue.

This instruction defines the job and a failure condition. It does not supply the actual price or guarantee that Claude will find every issue. Add the current source document, ask for a review, and check the proposed changes yourself.

Step 4: Understand chat boundaries

Anthropic's project-management guide says context is not shared across chats unless information is added to the project knowledge base. A practical implication is that a decision made only in Chat A should not be assumed to be available in Chat B. If the decision becomes an approved project fact, add it to the appropriate shared document or instructions and start a fresh chat to check that the relevant source is available.

That guidance should be read alongside Anthropic's newer project beta. Its overview describes a staged rollout of a new version in which project threads start with files, repositories, instructions and memory. Availability is limited during rollout. Do not assume your account has the beta or that the older guide's context rules describe every new beta behavior; check the version and current settings shown in your account.

For a comparison of the underlying assistants, see Claude vs ChatGPT and Claude vs Gemini. For plan comparisons, see Claude Pro vs Max. A project is an organization and context feature, not a substitute for reviewing the assistant's output.

Step 5: Review access before sharing

Anthropic says Team and Enterprise users can share projects with people in their organization unless an administrator has disabled sharing. Its guide describes a Share project control and two roles: Can view and Can edit. Viewers can see project contents, knowledge and instructions and chat in the project; editors can modify instructions and knowledge and manage member settings.

Those permissions make the project knowledge base a shared surface. Do not put a private customer file or secret in a project before checking who can see it. Give edit access only to people who should be able to change shared instructions or documents, and review access when a person changes roles.

Anthropic also notes that project sharing is distinct from sharing individual chats. Check the current sharing guide for what a recipient can see, and verify the actual members and permissions shown in the project. Do not assume a private chat stays private merely because you did not send its link if the account's project-sharing model says otherwise.

What happens as the knowledge base grows?

Anthropic says paid plans can automatically enable retrieval-augmented generation, or RAG, when project knowledge approaches the context-window limit. In practical terms, ask the current product documentation how material is selected when the knowledge base grows. Do not treat a larger capacity as proof that every sentence of every document is present in each answer.

Keep source documents organized and remove obsolete versions when appropriate. If two documents disagree, identify which is authoritative and ask Claude to explain the conflict rather than silently choosing one. For important facts, open the original file and verify the passage yourself.

A concise knowledge base is often easier to maintain than many near-duplicate documents. Give each source an owner and review date, and update the shared file when a policy or price changes. A project can reduce repeated setup, but it cannot make an outdated document current.

Common mistakes

• Putting critical context only in the project name. Anthropic says the name and description are not available to Claude as task context.
• Assuming every chat shares its history. The management guide says to add cross-chat context to project knowledge; check the separate beta rules if applicable.
• Uploading too broadly. Project viewers can see knowledge and instructions under the documented sharing roles.
• Treating an old file as current. Date and maintain the source that answers a time-sensitive question.
• Giving edit access by default. Editors can modify shared instructions and knowledge.
• Accepting a plausible answer without checking. Compare important claims with the actual source.

FAQ

Method

We reviewed Anthropic's public project-management and project-overview help pages on October 30, 2026, and used Toolradar's dated AI-assistants CSV for category context. We did not sign in or test the feature. The beta rollout, plan limits and sharing controls may change; check the current account and documentation.

Cite this: Dupple, "How to Use Claude Projects in 2026", sources checked October 30, 2026.