Should You Give Claude a Role, and Where Does It Go?
Yes, when you want a particular tone or focus, and one sentence is enough. In the Claude app, put it in your message, in project instructions, or in Instructions for Claude, depending on how long it should last. On the API, it goes in the system prompt.
You are [a role, such as a copy editor at a personal finance blog]. I am [who you are and what you already know, such as a founder with no accounting background]. The reader is [who will read this]. Explain any term I might not know. Skip anything I've said I already know. Task: [your request].
- 1Your messagethis chat only
- 2Project instructionsevery chat in that project
- 3Instructions for Claudeall your chats
- 4API system promptrequests from your app
Try it now
- 1For one chat, paste the prompt above as your first message.
- 2For repeated work, go to **Projects, click + New Project, click Set project instructions, add the role, and click Save instructions**.
- 3For every chat, click your initials in the lower left, then Settings, then **Your Claude, and add it under Instructions. The older layout uses Account and Instructions for Claude**.
- 4Start a new chat in that spot and ask something without mentioning the role. You'll know it worked when the answer is written for that reader.
Use it when
When tone or point of view changes the answer: a copy edit for a specific publication, a code review from a security angle, or a draft for readers who know less than you do. Use the longer-lasting spots for setups you'd otherwise paste every day.
Skip it when
Quick factual questions, where a role adds nothing. Also avoid account-wide instructions for anything that only suits one kind of work. A rule like "always answer in bullet points" follows you into every chat, including ones where it doesn't fit.
Why it works
Anthropic's docs say that setting a role focuses Claude's behavior and tone, and that even a single sentence makes a difference. What a role does is set the register and the assumed reader. Rule of thumb, not a vendor figure: it doesn't add knowledge, tools or access, so "you are a tax lawyer" won't make an answer legal advice. Saying who you are does a different job, since it tells Claude what to explain and what to skip. Claude may already know some of this, because memory is on by default for Free, Pro, Max and Ultra. Instructions for Claude apply to all your conversations, while project instructions apply only inside that project.
Example
Say you edit a personal finance blog and open with: "You are a copy editor for a personal finance blog. I'm the editor, so skip grammar lessons. Our readers are new to investing. Tighten this paragraph to 80 words and flag any claim that needs a source." The role sets the standard, and the second and third sentences stop Claude from explaining things to you or writing over your readers' heads.
Going further
Instructions and memory combine, so your answers may be shaped by things you've forgotten you said. To see what Claude does without what it remembers about you, click + in the message box and then Memory before your first message. That chat stays in your history. An incognito chat, from the ghost icon in the upper right of a new chat outside a project, is different: it isn't saved to memory or history. On the API, Anthropic shows the role as a system parameter rather than a user message. Its help pages also suggest keeping project instructions short, since they use context and usage.
Common mistakes
- Treating a role as a credential. A confident "as your attorney" answer still needs checking by someone qualified.
- Writing the role in the project's name or description. Claude can't see either one.
- Writing paragraphs of backstory for the role. Anthropic says one sentence can make a difference, so save your words for the task.
- Forgetting an old account-wide instruction, then wondering why every answer has the same odd format. Check Settings > Your Claude.
Sources
- Prompting best practices, Claude Platform Docs
- Claude's personalization features, Claude Help Center
- Create and manage projects, Claude Help Center
- Chat search and memory, Claude Help Center
- How do usage and length limits work?, Claude Help Center
Checked 6 November 2026. Products change often, so confirm details with the vendor.