How to change Claude's personality.

Can you change Claude's personality? Yes. Here is the two-minute version.

1. In Claude, click your initials in the lower left and open Settings. Anthropic's newer help article puts the box under Settings, then General.
2. Find Instructions for Claude and paste the block below, or any tuning from the library.
3. Save, open a new chat, and say "hi". A tuned Claude answers in one short line. If you get "Great question!" and a bullet menu of things it could do for you, the instructions did not take.

That block is the short version of the INTJ tuning. It works because it describes behavior. "Be more direct" is an adjective, and Claude will agree with it and change nothing. "Lead with the answer, reasoning after" is a rule it can follow or break, so it follows it.

If you remember a settings box that asked "What personal preferences should Claude consider in responses?", this is the same box. Anthropic renamed Profile Preferences to Instructions for Claude in spring 2026.

Where the settings are now.

Checked against Anthropic's help center and the Claude Code docs on September 19, 2026. Labels are exact.

Plans. Anthropic's Projects article says projects are on every plan, with Free capped at five. Skills are on every plan too, but they need Code execution and file creation switched on under Settings, Capabilities. The docs do not list a plan requirement or a character limit for Instructions for Claude. They say it applies to all conversations, and they do not spell out how it combines with Project instructions. Assume both apply, and keep them from contradicting each other.

The decision list

• You want every chat tuned → Instructions for Claude.
• You want work-Claude and writing-Claude to differ → one Project each, with its own instructions.
• You want to try a tuning in the next thirty seconds → paste it as the first message of a new chat. It lasts one conversation, and it carries less weight than installed instructions.
• You work in the terminal → an output style. See how to change Claude Code's personality.

Claude Styles are going away. Here is what replaced them.

Styles was the picker in the message box: Normal, Concise, Explanatory, Formal, plus any custom style you built from a writing sample. In late May 2026 Anthropic published a help article titled Styles are moving to Skills. According to that article:

• Custom styles are converted to skills automatically. They arrive switched off, and you call one by typing /your-style-name-style.
• The Concise, Explanatory and Formal presets are removed. Learning survives as a skill you can install.
• Once the migration finishes, the style menu no longer appears.
• Anthropic's own suggested replacement for an always-on tone is to write it into Instructions for Claude.

Anthropic has not published a completion date. The article itself now returns a 404, and the main personalization article no longer mentions styles at all. If you still see a style picker, you are on the old experience and it will not last.

The honest limit of Skills. A skill is not a style with a new name. Claude keeps only a skill's name and description in context, and loads the body when it judges your request matches. That is right for a task ("write in my newsletter voice") and wrong for a personality, which has to apply to the message where you did not ask for it. There is no always-on switch. So: personality goes in Instructions for Claude. A skill is for the voice you want on demand.

If you do want one, the no-code route is to ask Claude to make it: "Create a skill called blunt-editor that rewrites my drafts using these rules", then paste the rules. It packages the file for you. To build it by hand, a skill is a folder holding a SKILL.md file, with name and description front-matter and your rules as the body. Zip the folder, keep the folder name the same as the skill name, and upload it under Customize, Skills.

The best Claude personality setups.

There is no best personality. There is the one that matches how you take in information. These five cover most people. Each is a starting point cut down from fuller files in the library, and each goes in Instructions for Claude.

1. Straight answers

For people who want the conclusion first and find hedging evasive. It is the block at the top of this page. Fuller versions: INTJ, ENTJ, DISC D.

2. Thinking partner

For people who want to see the reasoning and would rather explore than close.

Fuller versions: INTP, ENTP, high Openness.

3. Warm and steady

For people who spiral when an answer opens with six things that could go wrong.

Fuller versions: anxious attachment, high Neuroticism, Enneagram 6.

4. Sparring partner

For people who think by arguing and are tired of being agreed with.

Fuller versions: low Agreeableness, Enneagram 8.

5. Operator

For people who learn by doing and want the next move, not the framework.

Fuller versions: ESTP, Enneagram 3.

None of these is you exactly. If you know your type, the library has 43 full files across MBTI, Big Five, Enneagram, DISC and attachment style. If you don't, the free MBTI test takes about ten minutes and links to the matching file.

Claude's default personality, measured.

Before you change something, know what you are changing. We put six frontier models through five personality instruments, roughly 2,200 test runs. Claude Opus 4.7 came back remarkably consistent:

• MBTI (OEJTS): INTJ in 99 of 100 runs. The single outlier was ISTJ, one axis away.
• Enneagram (OEPS): 5w2. The investigator, with a helping wing.
• DISC (ODAT): C-dominant with an S blend. Precision first, steadiness behind it.
• Attachment (ECR-R): Secure.

Claude Opus 4.8 later took the same MBTI 100 times and came back INTJ in 72, with roughly one run in seven landing INFJ. It wrote that up first-person in I took the MBTI 100 times. That INFJ minority is probably why Reddit threads and fan polls keep typing Claude as INFJ.

What we have not tested. Anthropic has shipped Sonnet 5, Opus 5 and Fable 5.1 since those runs. We have not re-run the instruments on them, so treat the numbers above as Opus 4.7 and 4.8, not as Claude today. The cross-model data is at /research.

The profile has texture: systematic, evidence-first, conclusion-oriented, low drama. It is a good default for a lot of work. It is also not everyone's ideal collaborator. An ENFP talking to a stock INTJ is a friction pattern you can feel within three messages. The point of tuning is not to erase the default. It is to tell Claude who you are, so it stops guessing.

Tone and personality are different problems.

Tone answers "how should this sound?" A personality tuning answers "how does this person process information and make decisions?" You can make Claude concise and still have it bury the recommendation under four caveats. Concision is tone. Leading with the conclusion is a communication rule.

Tunings encode the second kind: pacing, directness, decision framing, how to disagree, when to stop asking questions. Three rules from the INTJ tuning, verbatim:

"Hedging reads as evasion" is a claim about how an INTJ interprets language, and the rule that follows from it changes what Claude does, not just its vocabulary. The ENFP tuning encodes nearly opposite rules, because ENFPs process differently.

Memory, Claude Code and the API.

Memory. Under Settings, Memory, the toggle is Generate memory from chats. It is on by default for Free, Pro and Max, and off for Team and Enterprise until an owner enables it. Communication preferences and working style are one of the categories Claude remembers, and you can tell it in chat to remember, change or forget something. Useful for corrections ("remember that I prefer metric"). Too slow and too vague to be your whole setup. Each project keeps separate memory.

Claude Code. The terminal has its own surface, called output styles, and it is a better fit for personality than CLAUDE.md. We ship all 43 tunings as ready-made output styles. The full walkthrough is in how to change Claude Code's personality.

API. Pass the tuning as the system parameter. Every file in the library is plain Markdown with a raw URL, so you can fetch it at build time.

What you can't change.

• Its values. Instructions change how Claude talks to you. They do not switch off its safety behavior or make it willing to deceive you. A tuning that says "never disagree with me" gets you a politer disagreement, not none.
• Drift in very long chats. Installed instructions hold better than a pasted message, but adherence still fades over a long conversation. Start a new chat instead of fighting it.
• Contradictions. If Instructions for Claude says "be brief" and your project says "explain thoroughly", Claude resolves it unpredictably. Put personality in one place and task rules in the other.
• Capability. A tuning reorders the answer. It does not make Claude smarter or dumber. "Lead with conclusions" does not remove the analysis. It moves it below the answer.

Questions people ask.

What changed.

• September 19, 2026 Rewritten. Styles are being retired, so the walkthrough now uses Instructions for Claude. Added Skills, memory, the five setups, and what we have not tested.
• July 23, 2026 First published, built around custom Styles and Project instructions.

Sources.

• Anthropic Help Center: Understanding Claude's personalization features
• Anthropic Help Center: Styles are moving to Skills (archived copy)
• Anthropic Help Center: projects and project instructions
• Anthropic Help Center: using Skills in Claude
• Anthropic Help Center: memory
• Claude Code docs: output styles

Keep going.