How do you give ChatGPT a personality?
Go to Settings → Personalization, pick a built-in preset (Default, Friendly, Efficient, Professional, Candid, Quirky, Cynical, Nerdy), then add Custom Instructions — up to 1,500 characters per field — describing the traits you want. As of 2026-07 this is free on every plan, applies to all new chats instantly, and beats re-typing "be blunt" every session.
Why — the first-principles explanation
To steer a personality reliably you have to know what one is here. ChatGPT predicts the next token given everything in its context window. Your custom instructions get placed into that context before your message — so they aren't a setting the model "obeys," they're evidence about what kind of conversation this is. The model then continues in the style that best fits the evidence. You aren't flipping a switch; you're establishing a role, and the model plays along.
That single fact explains every rule of thumb that follows. Describe, don't command. "Be interesting" is an instruction with no statistical pull — text preceded by "be interesting" doesn't look like anything in particular. "You are a skeptical editor who has read the paper twice and finds one real hole" describes a person, and there is a huge amount of text in the world written by people like that. The second prompt gives the prediction loop something to imitate. The first gives it nothing.
Negatives work poorly; positives work well. Telling a model "don't be verbose" still fills the context with the concept of verbosity, and leaves the actual target unspecified — there are a thousand ways not to be verbose. "Answer in three sentences, then stop" names the target directly. Where a trait has a natural opposite, always state the side you want.
Personality drifts back over long chats. Your instructions sit at the front of the context; by turn forty they're competing with thousands of tokens of ordinary-sounding conversation, and the model's default trained voice reasserts itself. This is why saved Custom Instructions beat a one-off prompt — and why, in a long session, the fastest fix is often just starting a fresh chat rather than arguing with the drift.
One honest limit: the preset and your instructions sit underneath OpenAI's own system prompt and safety training. You can make ChatGPT terse, sardonic, or relentlessly Socratic. You cannot instruct the personality layer into removing its guardrails — those aren't part of the persona.
An example that makes it click
It's the difference between telling an actor "be funny" and handing him a character sheet.
"Be funny" gets you a nervous improv — the actor guesses, hedges, and defaults to his own voice within a minute. But hand him: "You're a night-shift diner cook, forty years in, you've seen everything, you answer in short sentences and you're never impressed" — and he knows instantly how to stand, how long to pause, what to leave out. He can stay in it for an hour, because he's not guessing at a mood, he's playing a person.
Custom instructions are the character sheet. The model has read millions of pages written by skeptical editors, patient tutors, and blunt engineers. Name the person and it has something to imitate. Name a vibe and it has nothing.
How to do it
- Open ChatGPT on web or desktop and click your profile icon, then Settings.
- Select Personalization and make sure the customization toggle is turned on.
- Under base style and tone, pick a preset personality — as of 2026-07 the options include Default, Friendly, Efficient, Professional, Candid, Quirky, Cynical, and Nerdy. Pick the one closest to what you want; you'll fine-tune next.
- Open Custom Instructions. In the field about how ChatGPT should respond, describe a specific role rather than an adjective — for example: "You're a blunt senior engineer reviewing my work. Lead with the biggest problem. Three sentences max unless I ask for detail. No preamble, no summary of what I just said."
- In the field about you, add the context that changes the answer, not your life story: your job, your expertise level, what you're usually trying to do, what you already know so it stops explaining it.
- Save. It applies immediately to all new conversations — old chats keep their old behavior.
- Test it with a question you've asked before, then edit. Treat the first version as a draft; the fastest improvements come from deleting the lines that did nothing.
- For a persona you only want sometimes, skip global settings and build a Project or a custom GPT instead, so it doesn't leak into every chat.
Key facts
- Custom Instructions fields have a 1,500-character limit each, as of 2026-07.
- Custom Instructions are available on all ChatGPT plans — including free — on Web, Desktop, iOS, and Android.
- Preset personalities as of 2026-07 include Default, Friendly (formerly Listener), Efficient (formerly Robot), Professional, Candid, and Quirky, with Cynical and Nerdy also selectable under Personalization.
- Presets and Custom Instructions stack: choose the preset closest to your target, then use instructions to fine-tune specific aspects.
- Changes apply immediately to all new chats and persist until you change, disable, or remove them; existing conversations are unaffected.
- The setting path is Settings → Personalization → Custom Instructions on web; on iOS and Android it is under Personalization → base style and tone.
OpenAI's conversational assistant — the most-used AI chatbot in the world.
▶ The 60-second explainer (script)
How do you give ChatGPT a personality? Two settings, thirty seconds. Settings, then Personalization. Pick a preset — as of July 2026 you've got Default, Friendly, Efficient, Professional, Candid, Quirky, Cynical, and Nerdy. Then open Custom Instructions and describe what you want. Fifteen hundred characters, free on every plan, applies to every new chat. But here's the part that actually decides whether it works. Your instructions don't get obeyed like a command. They get added to the model's context, and the model continues in whatever style fits. So you're not flipping a switch — you're casting a role. Which means: describe a person, not a mood. Be interesting does nothing, because text that follows the words be interesting doesn't look like anything in particular. You are a skeptical editor who finds the one real hole in my argument — that works, because the world is full of writing by exactly that person. The model has something to imitate. Second rule: say what you want, not what you don't. Don't be verbose leaves a thousand options open. Answer in three sentences, then stop, names the target. And if the personality fades after forty messages, that's normal — your instructions are competing with a long conversation. Start a fresh chat. That's the fix.
What authoritative sources say
People also ask
Why does the personality fade halfway through a long chat?
Your instructions sit at the start of the context window, and by turn forty they're outweighed by thousands of tokens of ordinary conversation. Start a new chat rather than arguing with the drift.
Do custom instructions work on the free plan?
Yes. As of 2026-07, custom instructions and preset personalities are available on all plans, on web, desktop, iOS, and Android.
Can I give it different personalities for different tasks?
Yes — use Projects or a custom GPT for role-specific personas. Global custom instructions apply to every new chat, which you usually don't want for a niche persona.
Why does 'be less wordy' never stick?
It's a negative with no target. There are countless ways to be less wordy. Give a measurable rule instead: "three sentences, then stop."
Can a personality prompt remove its restrictions?
No. Personas sit underneath OpenAI's system prompt and safety training. You can change tone, length, and role — not the guardrails.