How to Give ChatGPT Your Personality (Custom Instructions Guide)
ChatGPT gives generic advice for one reason: it's answering the average person, because it doesn't know who it's talking to. Custom instructions, memory, and project context exist to fix exactly that — but most people fill them with a job title and a sentence about tone, which barely moves the needle. This guide covers what personality context actually helps, how to set it up in each major AI tool, and the difference between describing yourself and measuring yourself.
(Disclosure up front: we build Plexality, a personality assessment with a one-click export for exactly this use case. This guide covers the do-it-yourself route fully — the shortcut is at the end, clearly labeled.)
What actually helps an AI know you
Three properties make personality context useful to a language model:
- Traits, not adjectives. "I'm an introvert who overthinks" gives the model a vibe. "Low assertiveness, high politeness, high orderliness, moderate volatility" gives it dimensions it can reason along — when to suggest directness, when to warn you about rumination, how much structure to put in a plan.
- Spectrums, not types. Positions on continuous scales ("about 70th percentile conscientiousness") carry more information than a four-letter box, and they let the model calibrate rather than stereotype.
- Stable vocabulary. Use the same terms every time. If your context says "high industriousness" and your questions say "I'm lazy lately," the model can connect the two — that contrast is exactly where personalized advice comes from.
Option 1: write it yourself (a template)
You don't need any product to do this. A solid self-written block covers:
- How I decide: fast vs deliberate, evidence-first vs gut-first, how much risk feels right
- How I communicate: direct vs diplomatic, how I handle conflict, what tone lands with me
- How I work: planned vs improvised, detail vs big-picture, what drains and what energizes
- Under stress: what I do when overwhelmed (withdraw, ruminate, over-control, avoid)
- What to do with this: "Tailor advice to these patterns. Challenge me where I said I avoid things. Don't soften direct feedback."
The honest limitation: self-description is a self-portrait, and self-portraits flatter. People tend to describe the person they're trying to be — the model then advises that person instead of you. It's still better than nothing; it's just not measurement.
Option 2: paste a measured profile
The alternative is context built from an actual assessment — scored trait spectrums rather than self-description. This is what Plexality's Context Pack is: you take the free assessment, and one click copies your archetype, aspect-level Big Five scores, and profile as an AI-ready block (a compact version is sized for the custom-instructions box). Measured scores, stable vocabulary, formatted for pasting — the three properties above, without writing anything. Since we make it, weigh this paragraph accordingly; the DIY template genuinely works too.
Setting it up, tool by tool
ChatGPT — custom instructions: Settings → Personalization → Custom instructions → paste your block into "What would you like ChatGPT to know about you?" Applies to new chats automatically. Box size varies by plan; keep your block lean or use a compact version sized to fit.
ChatGPT — memory: memory is built for distilled preferences, not long verbatim documents — ask it to remember the highlights ("remember these key traits about me: …") rather than the full block. For the complete pack, use custom instructions or paste it into the conversation where you want it.
Claude: paste the block into a Project's instructions (best for ongoing use; Projects are a paid feature) — or into your profile preferences, which works on any plan.
Grok: paste it at the start of a conversation, or into its personalization settings where available.
Gemini: paste it into Saved Info ("Tell Gemini about yourself").
None of these sync to anything — if your self-understanding changes (or you retake an assessment), update the pasted block yourself.
What changes once it's in
Ask the same question with and without context and compare. Without it: advice for the average person. With it: advice that can reference your actual patterns — why a confrontation script that works for assertive people will backfire for you, or why your planning advice should account for the fact that you abandon rigid systems. How much each tool leans on the context varies, but the direction is consistent: specificity replaces boilerplate.
Frequently Asked Questions
Does ChatGPT remember my personality between chats?
Custom instructions apply to every new chat automatically. Memory (when enabled) persists what you've asked it to remember. Per-conversation pastes last only that conversation — use one of the first two for anything long-term.
Is it safe to put personality information into ChatGPT?
That's your decision to make deliberately: anything you paste is governed by the AI tool's privacy policy, not by whoever produced the content. Share what you're comfortable with — a trait profile is personal, so treat it with the same judgment as anything else you tell your AI.
Will personality context box the AI into stereotyping me?
Less than you'd think, if the context uses spectrums. Continuous scores invite calibration ("high but not extreme conscientiousness") rather than the stereotype-shaped reasoning that type labels can trigger. And you can always override in the moment — context informs, it doesn't lock.
Do I need a paid account or a paid test for this?
Mostly no. ChatGPT's custom instructions work on every plan, the DIY template costs nothing, and Plexality's assessment and free Context Pack are free. Some equivalents are plan-gated — Claude's Projects are a paid feature, and Gemini's Saved Info depends on account type — but every tool accepts pasted context in a conversation, which needs nothing.
References
- Plexality Context Pack and setup guide: plexality.com/your-ai
- What makes trait measurement different from self-description: our Big Five test guide and the 12-platform comparison.