Guide · Getting started

How to give your AI a personality.

Two different goals hide inside that sentence. Here is how to set up both across ChatGPT, Claude, Meta Muse, Perplexity, Grok, Gemini, and coding agents.

Quick answer: Write six to eight checkable rules—covering how replies open, length, formatting, pushback, and follow-up questions—and paste them into your AI's persistent settings box: ChatGPT Custom Instructions (1,500 chars), Claude Personal Preferences, Meta Muse Soul.md, Perplexity Spaces (4,000 chars), Grok Customize settings, Gemini Saved Info/Gems, or CLAUDE.md and AGENTS.md for coding agents.
On this page

"Give my AI a personality" means two different things.

People use that phrase for two distinct goals:

  • Direction A — Match how you work (communication tuning): Control whether the AI leads with the answer, challenges weak assumptions, skips cheerleading intros, or checks in before acting.
  • Direction B — Play a character (persona prompt): Give the assistant a named role, voice, or backstory (such as in Meta Muse Identity.md, a Gemini Gem, or a Grok Bot).

In both cases, checkable rules beat adjectives. “Be concise and candid” is vague; “Lead with your recommendation in sentence one, keep routine replies under 150 words, and name the weakest assumption in my plan” is easy for a model to follow on every turn. Copy this starter block or customize it in the Custom Instructions Generator:

# How to work with me - Lead with the direct answer or recommendation, then the main reason. - Keep simple answers concise; expand only when trade-offs or edge cases matter. - Use plain prose by default; use bullets or tables when comparing 3+ items. - Push back with evidence when my premise has a flaw instead of agreeing. - State material uncertainty in one plain sentence rather than hedging. - Skip praise, prompt recaps, and generic closing offers.

Four steps. About ten minutes.

  1. Pick your preferences (or take a 5-minute test): Toggle six plain-English sliders in the Preferences Builder, use the 20-second 5-System Matcher, or take a free quiz on the Tests Hub.
  2. Grab a template: Browse all 43 MIT-licensed tuning files or build a custom export in the generator.
  3. Paste it into your app’s settings box (web & mobile):
App / AgentWhere to paste (Web & Mobile)Walkthrough
ChatGPTSettings → Personalization → Personality presets & Custom Instructions (≤1,500 chars) or Project InstructionsChatGPT Personality guide · 20 examples
ClaudeInitials / Profile icon → Settings → Profile → Personal preferences (or Projects / Skills)Claude setup · Mobile & Voice
Meta MuseAssistant icon → Identity → Soul.md (keep avatar/name in Identity.md)Soul.md guide · 20 templates
PerplexityPerplexity Spaces → Space Settings → Custom Instructions (4,000 chars) or AI ProfilePerplexity Spaces guide
Grok & Grok BotX / grok.com: Settings → Customize Grok · Grok Bot: Edit Profile → DescriptionGrok settings · Grok Bot
Google GeminiSettings → Saved info, custom Gems, or ~/.gemini/GEMINI.mdGemini guide
Coding Agents~/.claude/output-styles/*.md, CLAUDE.md, ~/.codex/AGENTS.md (32 KiB), .cursor/rules/*.mdcAll coding rules files

Step 4 — Check for conflicts: Delete any line that doesn’t fit your work, and paste your rules into the Instruction Checker to catch contradictions.

Five systems you can tune from.

You can pick a tuning file using any of our five personality frameworks—each one focuses on a different communication layer (compare all five in our 5-system matcher):

MBTI — cognitive framing (16 types)

Maps how you take in information and make decisions. For example, an INTJ tuning leads with system architecture and tradeoffs; an ESFP tuning leads with the immediate concrete action. Take the test at /tests/mbti.

Big Five (OCEAN) — single-trait sliders (10 poles)

Five continuous dimensions—Openness, Conscientiousness, Extraversion, Agreeableness, and Neuroticism—with High and Low templates for each trait. Ideal when you want to adjust one specific habit without adopting a four-letter type. Take the test at /tests/big-five or compare MBTI vs Big Five for AI.

Enneagram — core motivation & pushback (9 types)

Focuses on what drives you and how you want critique delivered. Take the test at /tests/enneagram or browse all nine profiles in the Enneagram AI prompts guide.

DISC — workplace speed & structure (4 styles)

Four workplace styles (Dominance, Influence, Steadiness, Conscientiousness) and the fastest test on the site (16 questions, ~3 minutes at /tests/disc).

Attachment — autonomy & check-in cadence (4 styles)

Controls how often the assistant pauses for confirmation versus executing end-to-end, and how gently it flags uncertainty. Take the test at /tests/attachment.

What actually changes (and what our data shows).

Custom instructions change how a model delivers its work, not its raw intelligence. In our 60-reply controlled pilot, adding a short “Straight answers” tuning block raised top answer-first scores from 14/20 to 19/20, cut hedging from 0.35 to 0.05 per reply, and dropped trailing filler questions from 0.90 to 0.25 without hurting accuracy.

Treat any personality template as an editable starting draft. Keep the bullets that save you time on real tasks and delete the rest.

Questions people ask.

Can I really give ChatGPT a personality?

Yes, via Settings → Personalization → Custom Instructions (up to 1,500 characters) or per-workspace Project Instructions. Concrete rules like "lead with the recommendation" hold up across conversations far better than vague adjectives.

What's the difference between a persona prompt and a tuning file?

A persona prompt gives the AI a character to perform. A tuning file adjusts the AI to your working preferences: how it delivers disagreement, how many options it offers, and how often it checks in.

Which personality test should I use for AI tuning?

A test is completely optional—you can pick plain-English toggles directly in the generator. If you want a questionnaire, DISC (16 items, ~3 min) is fastest for workplace habits, MBTI (32 items) maps cognitive framing, Enneagram (36 items) maps motivation, Attachment (36 items) tunes check-in cadence, and Big Five (50 IPIP items) scores five continuous traits.

Do personality prompts actually change AI behavior?

Specific behavioral rules do; standalone type labels or adjectives mostly don't. In our 60-reply pilot, explicit rules shifted answer-first openings, hedge counts, and trailing questions measurably. Test any block on your own tasks with the free A/B test bench.

Doesn't my AI already have a personality?

Yes—default assistants share a similar RLHF baseline: structured, agreeable, and cautious (scoring 44–47/50 on Agreeableness and predominantly INTJ/ISTJ in our 2,000-session questionnaire archive). A tuning tells the model where your preferences differ from the average user it was trained for.

What if the tuning file gets things wrong about me?

Edit it; that's the design. The type is a starting hypothesis and the file is plain Markdown under an MIT license. Delete rules that don't sound like you, sharpen the ones that do, add what's missing. A tuning you've edited for ten minutes beats any unedited type match.

Keep going.