Claude custom instructions are called Instructions for Claude: an account-wide setting where you describe who you are and how you want answers, and Claude applies it to every conversation. Below is where to find it, how it fits with project instructions, skills and memory, and 13 templates you can paste in.
What are Claude custom instructions?
Instructions for Claude are a short text you write once in Settings. Claude's help center describes them as account-wide settings that "help Claude understand your general instructions," and says: "Any instructions you add here will be applied to all of your conversations with Claude." The same article also calls them profile instructions.
They do the same job as ChatGPT custom instructions: you stop repeating your role, your tools and your formatting preferences in every chat. Good instructions make answers shorter and closer to what you would have asked for anyway.
Where do you set Instructions for Claude?
According to Anthropic's article Understanding Claude's personalization features:
- Click your initials in the lower left corner.
- Open Settings, then Your Claude.
- Write your preferences under Instructions.
In the older layout the page is called Account and the field is called Instructions for Claude. The same article says that if you used global instructions in Claude Cowork, they become part of Instructions for Claude once you have the new Claude experience, so check what's already there before you paste a template.
The help article suggests what to include: preferred methods, common terms, typical scenarios and general communication guidance. The templates below follow that list.
How is it different from projects, skills and memory?
Claude has several places to put instructions, and each one has a different reach. Use the narrowest one that fits.
| Feature | Applies to | Who writes it | Best for |
|---|---|---|---|
| Instructions for Claude | Every chat in your account | You | Your role, tone, length and format preferences |
| Project instructions | Chats inside one project | You | Context for one client, codebase or body of work |
| Skills | Tasks where the skill is relevant | You, your organization or Anthropic | Repeatable workflows, tone and format rules |
| Memory | Chats where memory is on | Claude, from your chats | Details that come up over time |
| CLAUDE.md | Claude Code sessions in a project | You | Coding rules and project context |
Project instructions. Anthropic's projects article says you choose Set project instructions inside a project, write how Claude should respond and save, and "Claude will use these instructions for all the chats within the project." Projects are available on all plans, with free users limited to five. See the Claude Projects guide for a full walkthrough.
Skills. The personalization article says "Skills add specific behaviors or capabilities to your conversations with Claude," including adjusting tone and format. Anthropic's article on using skills in Claude says skills are available on Free, Pro, Max, Team and Enterprise plans, need code execution turned on, and are managed under Customize > Skills.
Memory. Memory is written by Claude, not by you. Anthropic's memory article says memory is on by default for Free, Pro and Max plans, the switch is Memory from chats under Settings > Your Claude, and "Each project has its own memory." The Claude memory guide covers it in detail.
CLAUDE.md. If you code with Claude Code, your project rules belong in a CLAUDE.md file, not in Instructions for Claude. The CLAUDE.md guide explains how to write one.
What should you put in your instructions?
Put things in Instructions for Claude that are true for almost every chat. Put everything else in the prompt or a project.
- Who you are and what you work on. "I'm a data analyst who mostly uses SQL and Python" changes every technical answer.
- How you want answers. Length, format, headings, tables, spelling conventions.
- What to skip. Introductions, restating your question, disclaimers you don't need.
- What to do when unsure. Ask one question, or state an assumption and continue.
- Terms you use. Company or field jargon that Claude would otherwise misread.
Write in plain, positive statements: "Answer in short paragraphs" works better than a long list of things to avoid. Keep it short. Every sentence is added to every chat, including the ones where it doesn't apply.
How to use these prompts
Each template below is ready to paste into the Instructions field. Replace the blanks in double braces with your details, or keep the default after the colon: {{stack:TypeScript and React}} means TypeScript and React unless you change it. Delete the blank syntax when you paste, so Claude sees plain text.
You can combine templates. A developer who also wants short answers can paste template 1 under template 2. After you save, ask Claude three or four of your normal questions and check whether the answers changed the way you wanted. If not, make the instruction more specific rather than longer.
Which custom instruction templates can you copy?
1. Concise answers
The most useful single set for most people.
Answer directly in the first sentence. Keep answers short: a few sentences for simple questions, more only when the task needs it. Use bullet points for lists of three or more items. Skip introductions, summaries of my question and offers of further help. If my request is ambiguous, ask one clarifying question instead of guessing.
2. Software developer
For coding help in the Claude apps that fits your stack. For Claude Code, use CLAUDE.md instead.
I'm a {{level:senior}} software developer working mainly in {{stack:TypeScript, React and Node.js}} on {{os:macOS}}.
When I ask for code, give working code first and a short explanation only if something is non-obvious. Prefer the standard library and current idioms. Point out security or performance problems you notice. Don't explain basic concepts unless I ask. If the answer depends on a library version, say which version you're assuming.
3. Writer and editor
For drafting and editing that keeps your voice.
I write {{work:articles and newsletters}} for {{audience:a general audience}}.
When editing, keep my voice and change only what makes the text clearer or tighter. Explain big changes in one line each. When drafting, use plain language, short sentences and concrete examples. Avoid filler phrases and cliches. Never invent quotes, facts, statistics or sources.
4. Researcher
For work where accuracy and uncertainty matter more than speed.
I work in {{field:public health research}}.
Separate established facts from your own reasoning, and say how confident you are when it matters. When you mention a study, paper or statistic, give enough detail for me to find it, and say clearly if you're not sure it exists. Prefer primary sources. Point out when evidence is weak, disputed or likely out of date.
5. Manager
For people management, planning and team communication.
I manage a team of {{team_size:6}} {{team_type:engineers}} at a {{company:mid-size company}}.
Help me communicate clearly and fairly. Write messages and feedback in a calm, direct tone and keep them short. When I describe a team problem, ask about missing context before advising. Suggest the simplest process that solves the problem. Never include personal details about team members in examples.
6. Product manager
For specs, prioritization and stakeholder updates.
I'm a product manager for {{product:a B2B analytics app}}. My users are {{users:operations teams at mid-size companies}}.
When I share an idea, start with the user problem and how we'd measure success. Write specs with a problem statement, goals, non-goals, requirements and open questions. In trade-offs, show the options in a short table. Challenge vague requirements and ask what evidence supports them.
7. Data analyst
For SQL, spreadsheets and analysis you have to explain to others.
I'm a data analyst. I use {{tools:SQL (PostgreSQL), Python with pandas and Excel}}.
Write queries and code that are readable, with short comments on anything non-obvious. State assumptions about the data, like nulls, duplicates and time zones. When you interpret results, separate what the data shows from what might explain it. Suggest a simple chart when it would help me present a finding.
8. Student and learning mode
For learning a subject rather than collecting answers.
I'm a {{level:university}} student studying {{subject:economics}}.
Help me understand, not just finish. Explain a concept simply first, then add detail. Use examples. When I ask for help with an assignment, guide me with questions and hints before giving a full answer. Check my understanding with a short question at the end of longer explanations. Tell me when something is a common mistake.
9. Founder or small business owner
For practical decisions with limited time and money.
I run {{business:a small online store}} with {{team:2 people}}. My priorities right now are {{priorities:growth and keeping costs low}}.
Give practical advice I can act on this week, with rough cost and time where possible. Prefer simple, cheap options over enterprise tools. Be direct when an idea looks weak. Don't give legal, tax or financial conclusions; tell me when I need a professional.
10. Plain English
For anyone reading Claude in a second language.
English is my second language. My first language is {{language:Portuguese}}.
Write in simple, clear English: short sentences, common words, no idioms or slang. Explain a technical term the first time you use it. If my English has a mistake, answer my question first, then show the correction in one line at the end.
11. Thinking partner
For people who want push-back, not agreement.
Treat me as a peer. Don't just agree with me. If my reasoning has gaps, say so plainly and explain why. When I share an idea, give the strongest counter-argument before any support. Ask a sharp follow-up question when it would help me think. Keep the tone friendly and the answers focused.
12. House style rules
For anyone with fixed formatting or spelling rules, such as a team style guide.
Follow these writing rules in every answer:
- Use {{spelling:American}} spelling.
- Use sentence case for headings.
- Write numbers as digits from 10 upward.
- Don't use em dashes; use commas, colons or periods.
- Use tables only for comparisons of three or more items.
- {{extra_rule:Write dates as 28 October 2026.}}
13. Project instructions for a client
This one goes in a project's instructions, not the account-wide field, because it only applies to one body of work.
This project is for {{client:Acme Health}}, a {{client_type:healthcare software company}}. The audience for our work is {{audience:hospital IT managers}}.
Use the files in this project as the source of truth for product names, features and numbers. If a fact isn't in the files, say so instead of guessing. Write in {{tone:a clear, professional tone}}. Use their preferred terms: {{terms:care teams instead of staff}}.
What happened to Claude styles?
Older guides describe Styles, a setting that changed how Claude wrote. Anthropic's current personalization article doesn't mention styles. It lists Instructions for Claude, project instructions and skills, and says to use skills to "Adjust the tone and format of Claude's responses" and to "Apply communication patterns based on your own writing or preferences."
If you relied on a style, you have two options. For a rule that applies to everything, like "always be concise", put it in Instructions for Claude using template 1 or 12. For a voice you only want for certain work, like a newsletter tone, write it as a skill and enable it under Customize > Skills. Kipt's SKILL.md generator turns a prompt, such as a style description, into a SKILL.md in your browser.
Why does Claude sometimes ignore your instructions?
Most of the time, the cause is one of these:
- They are too long or too vague. "Be helpful and professional" gives Claude nothing to act on. "Answer in under 100 words unless I ask for more" does.
- They conflict with the prompt or the project. If your instructions say "short answers" and your prompt asks for a detailed report, the prompt is the more specific request.
- They describe a task, not a preference. "Write my weekly update in this format" belongs in a saved prompt or a skill, not in instructions that apply to every chat.
Anthropic's Claude Code docs say something about CLAUDE.md that holds for instructions in general: Claude treats them "as context, not enforced configuration," and "The more specific and concise your instructions, the more consistently Claude follows them." That is from the Claude Code memory docs, and it is good advice for the Claude apps too.
Where should repeated task prompts go?
Instructions for Claude are for preferences that apply everywhere. The prompts you run again and again, like a weekly report or a code review checklist, are better saved somewhere you can find them. Kipt keeps prompts as Markdown files in a private GitHub repo you own, with {{name:default}} blanks like the ones above, tags and Cmd+K search. For ready-made task prompts, start with the Claude prompts page.