User preferences
User preferences on claude.ai are account-wide settings that let you control how Claude behaves and communicates across all your conversations. They include profile instructions (a persistent text field where you describe your needs, role, or rules Claude should always follow), response styles (preset or custom formatting choices), appearance settings, and privacy/data controls. Any instructions you save in your profile are automatically included in every new conversation you start, acting like a standing background briefing Claude receives before you type a single word.
Beyond your personal profile, the preferences system has multiple layers. Project instructions apply only to conversations inside a specific project, letting you keep task-specific context separate from your global rules. On Team and Enterprise plans, organization administrators can set org-wide instructions that apply to every team member's conversations and take precedence over individual settings.
Preferences do not rewrite ongoing chats retroactively — changes take effect only in conversations you start after saving. They also cannot override Claude's built-in safety behaviors; instructions asking Claude to ignore safety guidelines or policies are simply disregarded.
When you’d use it
- ◆Consistent tone and formatting — A content writer who always needs UK English spelling, the Oxford comma, and no unsolicited bullet lists sets these as global profile instructions so every response is ready to paste without manual cleanup.
- ◆Professional context without repeated disclaimers — A licensed attorney adds their jurisdiction and professional status to profile instructions so Claude stops appending generic 'this is not legal advice' boilerplate on every response, saving editing time.
- ◆Project-specific coding standards — A developer working in a TypeScript monorepo creates a project with instructions specifying ESLint rules, naming conventions, and preferred patterns so Claude's code suggestions are already lint-compliant.
- ◆Organization-wide compliance language — An Enterprise admin at a financial services firm sets org instructions requiring Claude to flag any response touching PII and to cite applicable regulations, ensuring every employee's interactions meet compliance standards automatically.
- ◆Learning mode with structured explanations — A student switches to the 'Explanatory' style and adds a profile instruction to break down new concepts step-by-step, turning Claude into a consistent tutor rather than a quick-answer machine.
What changed recently
- ◆2025-10 — Anthropic updated its data retention and model-training opt-in policy. Data retention was extended to five years for users who allow their data to be used for model training. Free, Pro, and Max users (including Claude Code on those plans) are affected. Existing users received pop-up notifications and were given a deadline to choose their preference. The setting can be changed at any time in Privacy Settings.
- ◆2025 — Organization instructions were formally documented with clearer best-practice guidance. Anthropic clarified that org instructions should be short and specific because they are injected into every team member's conversation, consuming context window space at scale.
- ◆2026-01 — Past chat search began rolling out to Max, Team, and Enterprise plan users, with expansion to other plans anticipated. This feature allows Claude to search through a user's previous conversations to retrieve relevant context in new chats.
This is the short version
The full chapter has three worked examples, the common pitfalls, and the workflow that makes it pay — plus the other 84 features, kept current.
Get Claude Master — $97 →