What you will be able to do
- Tell durable account-wide and project instructions apart from a one-off prompt, and choose the right scope for an instruction
- Set a role or perspective that shapes every chat in a project
- Write instructions that are specific, explain their purpose, and say what to do rather than what to avoid
- Use structure and well-chosen examples so instructions are read the way you intended
Key concept
Scoped, durable instructions — A system-level instruction is written once and then applies to every conversation within its scope. Account instructions cover all your chats, and project instructions cover only the chats inside that project. A one-off prompt shapes a single reply.
1.Instructions that outlast a single prompt
When you type a request into a chat, you shape one answer. System-level instructions do a different job. You write them once, and Claude applies them to every conversation inside their scope. Tone, role, audience and house rules belong here, because those don't change from one request to the next.
You can set two levels of instruction yourself. Instructions for Claude live under Settings and apply to your whole account. Project instructions belong to a single project. The difference that matters is scope, so before you write an instruction, decide where it should apply.
| Feature | Where it applies | Typical use |
|---|---|---|
| Instructions for Claude (profile) | All of your conversations with Claude | Your preferred approaches, common terms you use, general communication instructions |
| Project instructions | Only chats within that project | Project-specific context, workflow guidelines, roles or perspectives for the project |
| Skills | Activated on demand in your conversations | Customizing how Claude formats and delivers its responses |
The personalization guide lists four jobs for project instructions: giving project-specific context, setting guidelines for a particular workflow, establishing requirements for a specific set of tasks, and defining roles or perspectives Claude should adopt. It also names the payoff: they help when you need Claude to maintain consistent context across multiple conversations within the same project. Without them, every new chat in the project would start from nothing, and you would retype the same framing each time.
Checkpoint 1 of 6· Match them up
Match each personalization feature to its scope or purpose
Tap a term, then the definition that fits it.
Profile instructions cover everything you do. Project instructions are limited to one project. Skills add behaviours or capabilities, such as formatting, that you activate when you need them.
“Use profile instructions for account-wide settings that affect all your interactions with Claude.”Source: support.claude.com
Sources1
2.Give Claude a role or a perspective
A role is often the most useful thing a project instruction can set. The projects documentation gives two examples of tailoring: asking for a more formal tone, and asking Claude to answer from the point of view of a particular role or industry. Anthropic's prompting guidance explains why this helps. A role in the system prompt focuses Claude's behaviour and tone on your use case, and even a single sentence makes a difference. You don't need a long persona.
For example, a team checking marketing copy might write: *You are a compliance reviewer checking marketing copy against our published claims policy.* Every chat in that project now starts from that stance, and nobody has to restate it.
Remember that the role applies to every chat in the project. If one project serves two very different audiences, a role that suits one of them will still shape the answers written for the other.
Checkpoint 2 of 6· Exam question
A communications team wants every press-release draft produced in a shared project to come out in the company's house voice, without anyone having to explain the voice again in each chat. What should the team do first?
Correct answer: A — Write down the house voice rules and audience the team already applies, then put them in the project instructions.
- A. Correct. The durable direction has to exist as agreed wording before it can be configured; defining what good looks like comes first.
- B. This is the repeated one-off prompting the team is trying to escape, and it produces different results per person.
- C. Samples are useful background, but they do not state the rules, and inference from mixed past work is unreliable.
- D. This guarantees inconsistency, which is the opposite of the shared house voice the team asked for.
Checkpoint 3 of 6· Check yourself
A procurement team wants every answer in its project to read like a sceptical procurement officer reviewing a supplier bid. Which setup does this most reliably?
Defining a role or perspective is one of the listed jobs of project instructions, and it then applies to every chat in that project. Repeating it per chat is fragile, and account-wide instructions would apply the role to unrelated work too.
“Define roles or perspectives Claude should adopt within the project”Source: support.claude.com
3.Be specific, explain why, and say what to do
Most weak instructions are vague, unexplained, or written only as prohibitions. Anthropic's prompting guidance gives a fix for each.
Be specific. State the output format and constraints you want. "Create an analytics dashboard" leaves Claude guessing how ambitious to be. Adding that it should include as many relevant features as possible removes the guess.
Give the reason. An instruction with its purpose attached works better than a bare rule. The guidance contrasts "NEVER use ellipses" with explaining that the response will be read aloud by a text-to-speech engine that can't pronounce them. Once Claude knows the reason, it can apply the same thinking to cases the rule never mentioned.
Say what to do, not what to avoid. Rather than "Do not use markdown", describe the output you want: smoothly flowing prose paragraphs. Formatting in the instructions themselves also has an effect, because the formatting style of your prompt may influence the style of Claude's response.
| Less effective | More effective | Principle |
|---|---|---|
| Create an analytics dashboard | Create an analytics dashboard. Include as many relevant features and interactions as possible. | Be specific about the desired output |
| NEVER use ellipses | Your response will be read aloud by a text-to-speech engine, so never use ellipses | Give context or motivation |
| Do not use markdown in your response | Your response should be composed of smoothly flowing prose paragraphs. | Tell Claude what to do instead of what not to do |
Here is how the three habits combine. Suppose a project keeps producing figures that aren't in its source material. "Don't make up numbers" is a bare prohibition. A stronger instruction names the behaviour you want, sets its limit, and gives the reason: *Use only figures that appear in the uploaded reports. If a figure isn't there, say it isn't available, because these summaries are checked against the reports.*
Checkpoint 4 of 6· Exam question
An operations lead runs a project whose instructions describe the team's standard weekly status report. This week only, a board member has asked for a one-page version. What is the best way to handle the request?
Correct answer: B — Ask for the one-page version in that chat, leaving the project instructions describing the normal report.
- A. Knowledge is background material, not direction, and this still makes a temporary need permanent.
- B. Correct. A one-off need belongs in the individual request; durable instructions should keep describing the standing format.
- C. Baking a one-off requirement into durable direction changes every future report for a single exception.
- D. A whole parallel workspace is heavy maintenance for a single request, and the two projects will drift apart.
Checkpoint 5 of 6· Check yourself
Which project instruction best follows Anthropic's guidance for effective instructions?
This option says what to do, names the audience, and gives the reason, which lets Claude generalize to terms the instruction never lists. The others are vague, unexplained, or only prohibitions.
“explaining to Claude why such behavior is important, can help Claude better understand your goals and deliver more targeted responses.”Source: platform.claude.com
Sources3
4.Structure the instructions and show examples
Instructions get harder to read as they grow, especially once they mix rules, background context and sample outputs. Two techniques help.
Separate the different kinds of content. Anthropic recommends XML tags for prompts that combine instructions, context, examples and inputs. Wrap each kind in its own tag, such as <instructions> or <context>, and use the same tag names every time. Its guidance on writing instruction files makes a related point: related instructions grouped under headers and bullets are easier for Claude to follow than dense paragraphs.
Show, don't only tell. According to the guidance, examples are among the most reliable ways to steer format, tone and structure. Good examples are *relevant* (close to the real use case), *diverse* (covering edge cases and varied enough that Claude doesn't copy an accidental pattern), and *structured* (wrapped in <example> tags so Claude can tell them apart from the instructions).
Checkpoint 6 of 6· Check yourself
A support project's instructions include three sample answers, and all three happen to be about refunds. Claude now steers unrelated questions toward refund policy. Which guidance addresses this?
Examples are a reliable way to steer output, but examples that all look alike teach the shared feature as a rule. The guidance calls for examples that are diverse.
“Diverse: Cover edge cases and vary enough that Claude doesn't pick up unintended patterns.”Source: platform.claude.com
Exam traps
Each one states something that sounds right. Open it to see what is actually true.
1.The strongest instruction is a capitalised prohibition such as "NEVER do X".Why is that wrong?
Anthropic's guidance prefers telling Claude what to do and explaining why. With the reason, Claude can apply the instruction to cases it never named.
Covered in Be specific, explain why, and say what to do
2.Setting a role only works if you write a long, detailed persona.Why is that wrong?
A short role statement already focuses Claude's behaviour and tone. Anthropic notes that even one sentence makes a difference.
Covered in Give Claude a role or a perspective
Sources
Every claim above is drawn from one of these pages, quoted as it was written on the date shown.
- 1.https://support.claude.com/en/articles/10185728-understanding-claude-s-personalization-featuresOfficial docs
“Any instructions you add here will be applied to all of your conversations with Claude.”
↩︎ Instructions that outlast a single prompt“need Claude to maintain consistent context across multiple conversations within the same project”
↩︎ Instructions that outlast a single prompt“These instructions only apply to chats within that project.”
↩︎ Key concept“These instructions only apply to chats within that project.”
↩︎ Prediction“Use profile instructions for account-wide settings that affect all your interactions with Claude.”
↩︎ Checkpoint“Define roles or perspectives Claude should adopt within the project”
↩︎ Checkpoint - 2.
“instructing Claude to use a more formal tone or answer questions from the perspective of a specific role or industry.”
↩︎ Give Claude a role or a perspective - 3.https://platform.claude.com/docs/en/build-with-claude/prompt-engineering/claude-prompting-best-practicesOfficial docs
“Setting a role in the system prompt focuses Claude's behavior and tone for your use case.”
↩︎ Give Claude a role or a perspective“Be specific about the desired output format and constraints.”
↩︎ Be specific, explain why, and say what to do“Claude is smart enough to generalize from the explanation.”
↩︎ Be specific, explain why, and say what to do“The formatting style used in your prompt may influence Claude's response style.”
↩︎ Be specific, explain why, and say what to do“Wrapping each type of content in its own tag (for example, <instructions>, <context>, <input>) reduces misinterpretation.”
↩︎ Structure the instructions and show examples“Examples are one of the most reliable ways to steer Claude's output format, tone, and structure.”
↩︎ Structure the instructions and show examples“Tell Claude what to do instead of what not to do”
↩︎ Exam trap 1“Even a single sentence makes a difference”
↩︎ Exam trap 2“explaining to Claude why such behavior is important, can help Claude better understand your goals and deliver more targeted responses.”
↩︎ Checkpoint“Diverse: Cover edge cases and vary enough that Claude doesn't pick up unintended patterns.”
↩︎ Checkpoint - 4.https://code.claude.com/docs/en/memoryOfficial docs
“Organized sections are easier for Claude to follow than dense paragraphs.”
↩︎ Structure the instructions and show examples