What Context Does Claude Need From You?
Claude needs what it can't look up: who the result is for, what has already happened, and what the result has to achieve. Anthropic's docs compare Claude to a brilliant new employee who doesn't know your norms or workflows. Tell it what you'd tell that person on day one, and skip what it can already find.
This is part of [the larger task], for [who it's for]. They'll use it to [what the result lets them do]. What already happened: [the last email, meeting or decision, in a few lines] What they care about: [their priorities or likely objections] Limits: [length, tone, anything that must be included or left out] My request: [what you want]. If a missing detail would change your answer, ask me up to three questions first. Otherwise, list your assumptions at the top and go ahead.
- Memory from earlier chats
- Project knowledge files
- Web search, if it's on
- Who the result is for
- What already happened
- What it has to achieve
Try it now
- 1Find a recent prompt that got a generic answer, and read it as if you were a colleague with no background. Note what you'd have to ask.
- 2Open a new chat, fill in the prompt above with those missing details, and send it.
- 3If you'll reuse the background, go to **Projects, click + New Project, then click Set project instructions** and paste it in.
- 4You'll know it worked when the reply uses your details, and any assumptions it lists are ones you can confirm or correct in a line.
Use it when
Anything shaped by your situation: an email to a specific person, a plan for your team, a decision with a history behind it, or a review of work against a standard only you know. The test from Anthropic's docs is quick: if a colleague with minimal context would be confused by your prompt, Claude will be too.
Skip it when
Questions with one right answer regardless of who asks, such as how a TCP handshake works. Also skip anything in a code repository that Claude Code can read for itself. Its docs say it already finds your file layout and conventions, so writing them out adds little.
Why it works
Anthropic's best-practice docs say that giving the reason behind an instruction helps, because Claude can generalize from the explanation. Their example swaps a bare ban on ellipses for the reason: the text will be read aloud by a speech engine. The Fable 5 prompting guide makes the same point and says Fable tends to do better when a prompt states the intent behind a request. What Claude already has varies. Memory is on by default on Free, Pro, Max and Ultra, project knowledge is available in every chat in that project, and paid plans can search past chats. So the useful question isn't "what's my whole background?" It's "what can't Claude get to?"
Example
Say a client went quiet after a proposal three weeks ago. A bare "write a follow-up email" leaves Claude to invent the relationship. Instead, say you've worked together for two years, that their budget owner changed in September, and that you want a yes or no on a smaller first phase by Friday. Now the draft can name the smaller phase and the date. A generic check-in has no room for either.
Going further
Claude can't see a project's name or description. Anthropic's help page says so directly, so anything meant for Claude has to go in the project instructions or knowledge. Context isn't shared between chats in a project unless it's in the knowledge base, though each project does keep its own memory. To stop your account memory from reaching a client project, turn off Use account memory under Memory on the project's page.
Common mistakes
- Writing the background in the project description, where Claude never reads it.
- Expecting one project chat to know what you said in another. Add anything you want shared to project knowledge.
- Pasting a whole inbox or folder. Anthropic's context docs warn that accuracy and recall degrade as the token count grows, which they call context rot.
- Relying on memory for a detail this answer depends on. Check what it holds at Settings > Your Claude, using Manage next to Your memory, or just state the detail.
Sources
- Prompting best practices, Claude Platform Docs
- Prompting Claude Fable 5, Claude Platform Docs
- Context windows, Claude Platform Docs
- Best practices for Claude Code
- Create and manage projects, Claude Help Center
- Chat search and memory, Claude Help Center
Checked 6 November 2026. Products change often, so confirm details with the vendor.