What Should You Not Tell ChatGPT?
A calm, practical list. Most of what you'd naturally type is fine — a small number of categories are worth a second thought, and a few should simply never be entered.
This isn't a warning about AI. Typing your preferences, your goals, or what you're cooking on Thursday into an assistant is unremarkable. But a chat window feels private in a way that isn't quite accurate — what you write may be stored, reviewed, or used according to the provider's policies, so a few categories deserve more thought than the rest.
Never enter these
Passwords and authentication information
Passwords, PINs, one-time codes, recovery phrases, security answers, API keys. An assistant can help you think about password hygiene without ever seeing a password, and there is no task that requires one. If a page or a chatbot asks for a code, treat that as a red flag regardless of where it appears.
Banking credentials
Card numbers, CVVs, account and routing numbers, online banking logins. You can ask for help with a budget, a dispute letter, or understanding a fee using rounded figures instead.
Government identification numbers
Social Security, national insurance, passport, driver's licence, tax identification numbers. These are used to verify who you are, which is exactly why they don't belong in a chat log.
Think twice about these
Highly sensitive medical information
Diagnoses, medications, test results, mental health history, anything relating to a child's health. Asking a general health question is normal and often useful. Attaching your full medical history to your permanent settings isn't — if you need that context for one conversation, give it in that conversation and keep it out of your saved profile. And verify anything consequential with a clinician; assistants can be confidently wrong.
Confidential employer information
Unreleased plans, internal documents, client data, contract terms, source code you don't own. Many organizations have an explicit AI policy — check yours first, because this is one of the few categories where sharing can have consequences beyond privacy.
Information belonging to other people
The easiest one to do accidentally: pasting an email thread, a group chat, a résumé, or a colleague's details. They didn't make the choice you made. Strip names and contact details, or summarize instead of pasting.
Precise personal identifying information
Your exact address, date of birth, phone number, employee or student ID, licence plate. "Northeast Ohio" gives an assistant everything it needs for weather, seasons, and units. Your street address gives it nothing extra.
Detailed financial information
Balances, debts, salary, investment positions. Round numbers and ranges work just as well for budgeting or planning help: "roughly £3,000 a month after tax" is enough.
What can I tell ChatGPT?
Most of what makes answers feel personal, as it turns out. How you want replies written, what you're trying to accomplish, your interests, your general work and household context, your recurring tasks, your real constraints on time and budget. None of that is sensitive, and it's where nearly all the improvement comes from — our guide to what to tell ChatGPT about yourself sorts it into useful, optional, and think-carefully so you can see the line clearly.
It's also worth knowing which controls exist. Saved memories can be viewed, edited, and deleted, and training on your content can be turned off in settings — OpenAI's memory documentation, data controls help page, and privacy policy describe the current behavior. We're a setup tool: we can help you decide what to share, but we don't operate any AI service and can't change how those companies store or process information.
You don't have to choose between two extremes
People tend to land in one of two camps: tell AI nothing and get generic, forgettable answers, or tell it everything and quietly hope it's fine. Both are worse than the middle. A short, deliberate profile of preferences and general context gets you the personalized version without the oversharing — you just have to decide what goes in it.
Knowing that is one thing; organizing it is another. The guided Personal AI Setup walks you through the decisions: simple questions one at a time, a plain note on each telling you whether the answer is useful, optional, or worth considering carefully, and a review screen where you include, edit, or remove every single line before anything is generated. Then you copy or download your Personal AI Profile and paste it into the assistant you already use. If you want the reasoning behind the split first, read how to personalize ChatGPT without unnecessarily oversharing.
There's a middle ground, and it takes minutes
The Personal AI Setup asks simple questions, tells you which answers are useful and which deserve care, and builds your profile from only the lines you approve.
Set Up My AI