AI Guides › Claude Mastery
The Difference Between A Chat And A Project (Claude Edition)
By Nigel Guy · 2 min read
Most people use Projects the same way they use a plain chat — as a place to start typing — and never notice the one thing a Project is actually for: it holds context so you stop retyping it. That's not a filing preference. It's the entire point of the feature, and skipping it means paying the "re-explain everything" tax on every single conversation you have about the same piece of work.
The rule: a Project is a standing memory for one piece of work, not a folder — put the facts that don't change conversation-to-conversation into it once, and every chat inside inherits them for free.
The mechanism
- Ask what's stable before you ask what's urgent. Before opening a chat, ask: is there background here — a style guide, a client brief, a codebase, a set of house rules — that will still be true next week? If yes, that's Project material, not chat material.
- Put custom instructions in the Project, not in your first message. Anything you'd otherwise paste at the top of every new chat — tone, format, constraints, who the audience is — belongs in the Project's instructions field. Write it once.
- Add reference documents to the Project's knowledge, not to individual chats. A pasted document in a single chat helps that one conversation. The same document added at the Project level helps every conversation you'll ever have in it.
- Start a new chat inside the Project for each new task, rather than one enormous thread. The Project supplies continuity; the chat inside it should still stay focused on one thing at a time — see "When To Start A New Conversation" for the reasoning.
- Prune the Project's knowledge occasionally. A Project that's accumulated eighteen months of outdated briefs is worse than an empty one — old material can crowd out or contradict the current version of the truth.
What to skip
Don't build a Project for something you'll only ever ask about once — a one-off question doesn't need standing infrastructure, it needs a chat. And don't try to use a Project as a substitute for the separate memory feature that already runs across your ordinary chats; they solve different problems; a Project is scoped and shared with anyone you invite into it, while memory is personal and follows you everywhere. Using one to replace the other means neither works quite right.
Guardrails
- Exactly what a Project can hold — file types, size limits, whether it supports connectors — moves with product updates. Check the current Claude.ai documentation rather than assuming last year's limits still apply.
- Anything you put in a Project's knowledge is available to everyone with access to that Project. Don't treat it as a private notebook if you've shared it with a team.
- A well-stocked Project reduces re-explaining; it doesn't guarantee correctness. Review anything the Project's material feeds into an answer, the same as you would for any other source.
All 751 AI guides · JulieMango plans from £17/mo