You told Claude about your writing style last week in a chat you have since closed. Today you open a new chat and want that style applied again, without turning memory on. What works?

The principle — continuity is either automated by memory, held by a workspace, or supplied by you. With memory off, the last two are what remain.

Why the key is correct — a new chat starts without the others, so the instruction has to enter the conversation somehow. Typing it again does that directly. Working inside a project does it structurally, because the workspace carries its own knowledge base and its own chat history, so a style guide placed there applies to every conversation inside it.

Why the others are wrong — memory is not the only mechanism, which matters because plenty of people prefer to leave it off. Instructions do not leak between conversations. And an exported chat is a file you hold, not something Claude reads unprompted.

Remember this — with memory off, put recurring instructions somewhere structural or expect to repeat them.

Sources — Anthropic's chat search and memory article, and its projects documentation.