Claude Projects, properly — keeping team context in one place
Projects solve the problem of re-explaining your situation at the start of every conversation. Used casually they become a folder of files; used deliberately they become the thing that makes Claude useful for a specific body of work.
AI Tools Hub Editorial · published · updated
What a Project is
A persistent context: custom instructions plus a set of files, under which you can run many separate conversations. Everything in that project inherits the instructions and can reference the files, so the setup cost is paid once.
Setting one up that works
- Write instructions that describe the work, not the AI — what the project is, who it is for, what good output looks like
- Add the documents that would otherwise be pasted repeatedly: specifications, style guides, past decisions
- Say explicitly what to avoid — formats, phrasings, assumptions that have caused problems
- Keep it narrow. One project per body of work beats one project for a department
A structure that works for teams
| Project | Instructions cover | Files |
|---|---|---|
| Product documentation | Voice, audience, terminology | Style guide, glossary, three good examples |
| Customer support | Tone, escalation rules, what not to promise | Knowledge base export, response templates |
| A specific codebase | Stack, conventions, review standards | Architecture notes, key modules |
What makes the difference
Including examples of good output. Instructions describing a tone produce an approximation of it; three examples of the real thing produce something much closer. This is the single highest-leverage thing in a project setup and the one most often skipped.
How many files can a project hold?
Enough that the practical limit is usually relevance rather than capacity. Fewer, well-chosen documents outperform everything you could find.
Can the team share these?
On Team and Enterprise plans, yes — which is where they stop being a personal convenience and start being shared infrastructure.
How does this compare to a custom GPT?
Projects are better for long-document context; custom GPTs are better for packaging and distribution. If the goal is depth on one body of work, Projects win.
Related
More articles
Opinion
Why AI rollouts fail in Japanese companies — four structural reasons
For business
Enterprise AI rollout checklist — SOC 2, ISO 27001, DPAs and personal data
For business
Japanese-developed LLMs for business — ELYZA, Sakana AI, NTT tsuzumi and PLaMo
Comparison
AI coding tools compared — Cursor, Copilot, Claude Code, Windsurf and Codeium