How do you use Claude Projects effectively?

Short answer

Use a Project for context that keeps coming back: one client, one product, one recurring report. Put the rules that must always apply in the project instructions, keep the knowledge files focused and named clearly, start a new chat for each task, and decide per Project whether it should share account memory. Content in a Project is cached, so reuse also counts less against your usage.

What goes where

A Project has four places to put context, and most disappointing Projects mix them up. Each place has a different job.

Place What Claude does with it Put here
Project instructions Applies them to every chat in the Project Tone, format, role, hard rules, what to always check
Project knowledge Reads it as background, or searches it when it gets large Reference documents, style guides, product sheets, code, templates
Project memory Remembers useful details from chats in this Project Review and edit saved topics; ask Claude to remember important details
Name and description Nothing: Claude does not see them A label for you and your colleagues

The description field is the trap. People write their brief there, and Claude never reads it. Anthropic’s help center says so directly: Claude will not have access to these details.

Set up a Project that works

  1. Open Projects in the sidebar (or go to claude.ai/projects) and click “+ New Project”.
  2. Give it a name per client, product or recurring job, not per person.
  3. Click “Set project instructions” and write the rules that hold for every chat: who the reader is, the format, the words you never use, what Claude must ask before it starts.
  4. Use the “+” button on the Project page to add knowledge: a few focused files with clear names beat one huge export.
  5. Under Memory on the Project page, decide whether “Use account memory” stays on. Turn it off when the Project must stay separate, for example client work.
  6. Test with a question that only the right file can answer, and check that Claude used it.

Work inside it

  1. One chat per task. Chats in a Project do not share context with each other unless the information is in the project knowledge. A long chat that covers five tasks also costs more usage, because conversation length is one of the factors Anthropic lists.
  2. Promote what you learn. When a chat produces a decision or a better version of a template, add it to the knowledge or the instructions. Otherwise the next chat starts without it.
  3. Keep files current. Uploaded files do not update themselves. If the price list changes, replace the file; an old version and a new version side by side make answers inconsistent.
  4. Let caching work for you. Anthropic says content in Projects is cached and counts less against your limits when it is reused. The cache expires after a period of inactivity, so the first message after a long break counts the content in full again.

Large knowledge: what changes

On paid plans, when the knowledge nears the context window, Claude switches to RAG mode: it no longer reads every file in full, it searches the knowledge and pulls the parts that seem relevant. Anthropic says this expands capacity by up to 10x. The practical effect: with little knowledge Claude sees everything; with a lot, it sees what the search finds. A question that needs two facts from opposite corners of a big collection can miss one. That is a reason to keep files focused and to put the rules that always apply in the instructions, not buried on page 40 of a PDF.

When a regular chat is better

A Project is worth it when the same context keeps coming back. For a one-off question, a normal chat is faster and does not pull in knowledge you do not need. And if you mainly want Claude to follow a procedure (how to write a proposal, how to format a report) across many Projects, a Skill fits better than copying the same instructions everywhere. See Claude Skills vs Projects.

The misunderstanding: a Project learns by itself

Project memory does pick up details as you work, but it stores selected topics, not everything. The corrections you make in chat (“no, we never discount the pilot”) only become reliable when you put them in the instructions. Treat the instructions as the place where the Project gets smarter, and review them every few weeks.

What this means for a team

On Team and Enterprise a Project can be shared, so the instructions and knowledge one person curates help everyone who opens it. What does not travel: each member’s own memory, and anything learned outside that Project. Two colleagues working on the same client in two different Projects build two separate pictures. For how sharing works, see sharing Claude Projects with your team, and for the basics, how Claude Projects work.

Where Bombos fits

A well-kept Project is still a folder that someone has to maintain by hand, inside Claude only. Bombos is a plugin that gives every AI tool your team uses, Claude and Claude Code alongside ChatGPT, Copilot and Cursor, one shared memory that reads what already lives in mail, documents and repos and keeps the rules your team taught it, so the correction made in one Project is there in the next. One person can try it first for their own agents.

Related questions

Does Claude read the Project description?

No. The name and description are for you and your colleagues. Anthropic's help center says Claude does not have access to those details, so anything Claude must know goes in the instructions or the knowledge.

Do chats in the same Project know about each other?

Not directly. The help center says chats do not share context with each other unless the information is in the project knowledge. Project memory and chat search inside the Project can bring some of it back.

How long should project instructions be?

Anthropic gives no limit. Keep them to the rules that apply to every chat in the Project; long reference material belongs in the knowledge files.

Should I put everything in one big Project?

No. Separate Projects keep memory and retrieval focused. Free accounts can have up to five Projects; the help center names no cap for paid plans.

Sources

  1. How can I create and manage projects? (Claude Help Center)
  2. What are projects? (Claude Help Center)
  3. Usage limit best practices (Claude Help Center)
  4. Use Claude's chat search and memory (Claude Help Center)
The first half hour · free

What does your AI forget?

  1. 01You tell us which AI you use and where it keeps starting over.
  2. 02We show what a shared memory changes there, on your own examples.
  3. 03You keep the plan, with or without Bombos.
Piet Baudoin
Piet BaudoinYou talk to Piet, from the Bombos teamDutch team from Utrecht and Nijmegen · data in Europe
We reply the same working day.