Claude custom instructions live in three places: Instructions for Claude in your settings apply to every chat, project instructions apply only to chats inside one project, and on Team and Enterprise plans an owner can add organization instructions that apply to everyone. This guide shows where each one is set, which wins when they disagree, and how to write instructions Claude follows.
On this page
Where each kind of instruction is set
Anthropic’s personalization overview and its related help pages describe the three levels:
| Level | Where to set it | Applies to | Who can set it |
|---|---|---|---|
| Instructions for Claude | Your initials (lower left) → Settings → Instructions for Claude | All of your conversations | Every user |
| Project instructions | Open a project → Set project instructions | Chats inside that project only | Anyone who can edit the project |
| Organization instructions | Organization settings → Organization and access | Every conversation in the organization | Owners and Primary Owners on Team and Enterprise |
Set your account-wide instructions
Click your initials in the lower left corner, choose Settings, and fill in Instructions for Claude. Anthropic suggests using the field for your preferred approaches or methods, terms you use often, situations you typically deal with, and general communication preferences. Whatever you save there is applied to all of your conversations.
A short example that follows those categories:
I'm a secondary-school science teacher in the UK. Use British spelling.
When I ask for lesson material, give a 3-point outline first and wait before writing the full version.
Keep answers under 250 words unless I ask for more. Skip opening compliments and closing summaries.
Use project instructions for one job
Project instructions are the right place for context that only matters to one piece of work: a client’s style guide, a course syllabus, a role you want Claude to take. According to Anthropic’s projects guide, you open the project, click Set project instructions, write how Claude should behave and respond, and click Save instructions. Claude then uses them in every chat in that project.
Projects are available on every plan; free accounts can create up to five. Two details from the same page are easy to miss:
- Files you add to the project knowledge base are used across all chats in the project, but context from one project chat is not shared with the next unless it’s in that knowledge base.
- The project’s name and description are not visible to Claude, so put anything Claude needs to know in the instructions or the knowledge base, not the description.
When organization and personal instructions conflict
On Team and Enterprise plans, owners can add organization instructions of up to 3,000 characters. Anthropic says they take precedence: if an organization instruction says “Always respond in formal English” and yours says “use a casual tone”, Claude answers formally. Your own instructions still apply to anything the organization’s text doesn’t cover. Changes can take up to an hour to reach all Claude products, and members can’t see what the organization instructions say; only Owners and above can.
How to write instructions Claude follows
Anthropic’s advice for organization instructions works for personal and project ones too:
- Be concrete. “Respond in formal English. Don’t use contractions, slang, or emojis” works better than “be professional”.
- Keep it short. Instructions are sent with every message, so a tight list keeps conversations efficient.
- Stick to habits that should always hold. Formatting, tone, terminology and standing context belong here; one-off requests belong in the chat.
- Check for contradictions. If two lines conflict, Claude may follow neither reliably.
- Test after saving. Start a new conversation and try a few different questions to confirm the result.
For the wording of individual prompts, our Claude prompting guide covers clear tasks and examples, and the system prompt guide explains how standing instructions differ from a single request.
Instructions, memory and past-chat search are separate
Instructions are text you write and control. Memory is what Claude saves on its own from your chats: Anthropic’s memory article says it is on by default for Free, Pro and Max plans, that each project keeps its own memory, and that you can switch memory off for a single chat from the + menu before the first message. If Claude keeps bringing up something you’d rather it didn’t, check Settings → Memory as well as your instructions.
