How to Set Up Preferences for Working With Claude
Setting up your Claude preferences takes about five minutes and changes every conversation you...
Setting up your Claude preferences takes about five minutes and changes every conversation you have afterward. Most people either don't know the setting exists or open the empty box and don't know what to write. This is the practical walkthrough: where to find it, what to put in it, and how to avoid the common mistakes.
Where do you set Claude preferences?
In Claude.ai: click your initials in the lower-left corner → Settings → Profile. Look for the field asking what preferences Claude should consider in responses. That box is your preferences (also called profile preferences or custom instructions). Type into it, save, and it applies to all your standalone conversations from then on.
It's available on every plan, including free. There's nothing else to configure in Settings — no toggle, no setup flow. Once it's filled in, every new chat starts with that context loaded.
What should you write?
Treat it like briefing a sharp new assistant. Cover four things:
1. Who you are. Your role, field, and level. "I'm a senior backend engineer working mostly in Go and Postgres." "I run marketing at an early-stage B2B SaaS." This sets the baseline for how technical and how detailed Claude should be.
2. How you want Claude to communicate. Be specific enough that it changes behavior. "Start with the answer, then explain only if I ask" beats "be concise." "Give me direct feedback, including what's weak" beats "be helpful."
3. What to do differently from default. This is often the highest-impact part. Name the specific behaviors that annoy you: "Don't open with a preamble." "Don't hedge every sentence." "When editing my writing, fix problems but don't rewrite my voice." "Don't ask for confirmation on every step."
4. Standing context. Recurring tools, terms, or situations: "I usually deploy on Vercel." "When I say 'the app' I mean our iOS client."
A template you can adapt
I'm a [role] working in [field/stack]. My experience level is [beginner/
intermediate/expert] in [area].
Communication:
- Start with the answer, then explain if needed
- Be direct — tell me what's weak and why
- Keep preamble to zero
Don't:
- Rewrite my voice when editing my writing
- Hedge every statement
- Ask for confirmation on obvious next steps
Context I reuse: [tools, recurring terms, typical scenarios]
Keep it to the 20% of preferences that cover 80% of your conversations. If something only matters for one kind of task, it belongs in a Project, not here.
Common mistakes to avoid
- Too vague. "Be helpful and concise" tells Claude almost nothing. Every line should be specific enough to change behavior.
- Too long. Preferences load as tokens on every message. A 2,000-word manifesto eats context budget before you've asked anything. Aim for 150–500 words; if you can't read it in 90 seconds, it's too long.
- Putting project-specific context here. A project fact like "this is a React app using Supabase" — the kind of thing you'd otherwise put in an
AGENTS.md— doesn't belong in account-wide preferences — it'll show up when you ask Claude to plan a trip. That goes in Project instructions. - Set and forget. Your role and priorities change. Review every month or two, and whenever you catch yourself correcting the same thing repeatedly, add a line. Stale preferences are worse than none — Claude applies outdated context confidently.
After preferences: Projects and cross-tool context
Once preferences are set, two next steps depending on your needs:
For project-specific context, use Claude Projects — each gets its own instructions and knowledge base, layered on top of your preferences, so you don't repeat universal stuff per project.
For context across other AI tools, note that everything above lives only in Claude. If you also use ChatGPT, Cursor, or other tools, you'd re-enter equivalent instructions in each. A context layer solves that by keeping your structured context in one place you own and serving it to every tool through MCP — set your Claude preferences first, then add a layer when you're tired of duplicating them everywhere.
→ Why preferences matter and where they stop: Preferences for Working With Claude
→ Persisting context across conversations: How Do I Give Claude Persistent Context Across Conversations?
→ Set your working context once, use it everywhere — Unabyss →
