What Should I Tell ChatGPT About Myself?

The short answer

Tell ChatGPT the things that change what a good answer looks like: how you want replies written, what you're trying to accomplish, the general shape of your work and home life, and the tasks you find yourself repeating. That's it. You do not need to describe your medical history, your finances, or exactly where you live to get answers that feel written for you.

A useful way to decide, question by question: would this change the advice? "I cook for four people on weeknights and have about 30 minutes" changes the advice. Your bank balance doesn't.

Useful to share

This is the material that does nearly all of the personalization work. It's low-sensitivity and high-value:

• Communication preferences — short or detailed, formal or casual, bullets or prose, whether you want to be asked a clarifying question instead of a guess.
• Your goals — what you're working toward this month or this year, in plain terms.
• Interests — what you read, cook, watch, build, or want to learn.
• General work context — your role, your industry, who you write for. No client names or internal specifics needed.
• Learning preferences — examples first, step-by-step, analogies, or just the conclusion.
• Preferred answer style — length, formatting, how much caveating you can tolerate.
• Recurring tasks — the emails, plans, summaries, and lists you write over and over.
• General lifestyle preferences — dietary preferences, how much time you realistically have, whether you drive, budget level in broad terms.

Optional

Helpful in some situations, unnecessary in others. Include these when they're relevant to how you actually use AI, and leave them out when they aren't:

• General family context — "two kids, primary school age" is usually enough; names and schools are not needed.
• Household information — who you cook for, who shares chores, pets.
• Broad location — a city or region helps with seasons, units, and availability. A precise address adds nothing.
• Shopping preferences — brands you like, stores you use, what you avoid.
• Routines — when your week is busiest, when you plan meals, when you exercise.

Think carefully before sharing

None of this is forbidden, and none of it makes you unsafe by itself. It simply deserves a pause, because it rarely improves the answer as much as it increases what you've disclosed:

• Medical information — diagnoses, medications, test results, mental health detail.
• Detailed financial information — balances, debts, salary, account specifics.
• Precise identifying information — full address, date of birth, phone number, identification numbers.
• Confidential workplace information — unreleased plans, client data, contracts, anything covered by an agreement or an employer AI policy.
• Highly sensitive personal information — anything you'd be uncomfortable seeing quoted back to you later, including information about other people who didn't agree to it.

And a firm line rather than a judgment call: never enter passwords, one-time or authentication codes, Social Security or national identification numbers, banking credentials, or card numbers. A chat assistant has no reason to hold a secret, and there's no version of personalization that requires one. Our guide to what not to tell ChatGPT covers each of these categories and why they matter.

How much does ChatGPT actually need to know about you?

Far less than the "share everything" instinct suggests. Personalization has a steep early curve and a flat tail. Tell an assistant how long you want answers, what you're working on, and the practical constraints of your week, and the improvement is immediate and obvious. Add another twenty personal facts on top and most answers won't change at all.

That's the useful thing to internalize: better personalization is about the right information, not the most information. A tightly written paragraph about how you like to work usually beats three pages of biography.

It's also worth knowing where this information lives. Standing instructions and saved memory are settings you control and can review, edit, or delete — see OpenAI's memory documentation and privacy policy for their current behavior and data controls. We're a setup tool, not the operator of any AI service, so treat those pages as the authority on how your data is handled.

Should I put personal information in custom instructions?

Custom instructions are the right home for stable, low-sensitivity context — your preferred answer style, your general situation, your recurring tasks — because they apply to every conversation without you retyping them. That permanence is exactly why sensitive details don't belong there: a one-off medical or financial question can be asked in a single chat and left behind, rather than attached to every chat you'll ever have. Our guide to personalizing ChatGPT without oversharing walks through how to split context between standing instructions and individual conversations.

Not sure what your AI should know?

Understanding the categories is the easy part. The tedious part is sitting down, working out which of them apply to you, phrasing them clearly, and organizing the result into something an assistant can actually use.

That's what the guided Personal AI Setup is for. You answer simple questions one at a time — mostly by tapping — and each one tells you whether the answer is useful to include, optional, or worth considering carefully. Then a review screen shows you every line before anything is generated, so you can include it, edit it, or remove it. Your Personal AI Profile is built from only what you approve, and you can copy or download it and paste it into the assistant you already use.