How much personal data should you share with AI? Enough to help, never enough to hurt.

By Complete OS · Published October 24, 2026 · Updated October 28, 2026

Share the details that make the answer better, like your schedule, your goals and what’s in the fridge, and never share anything that could hurt you if it leaked: passwords, full card or account numbers, Social Security or passport numbers, and other people’s private information. Before you share anything, check two settings: whether the service uses your chats to train its models, and whether you can delete your history. On consumer ChatGPT, training is on by default and you can turn it off.

The one test: would it hurt if it leaked?

Every AI service stores what you send it, at least for a while. Most are careful, but you should still assume that anything you type could one day be read by someone you didn’t intend, through a breach, a shared device or a mistake. That gives you a simple test for each piece of information: does it make the answer better, and would it hurt if it got out?

Your meal plan, your workout routine and your bedtime pass easily: useful, and low-harm. Your bank password fails both ways: it doesn’t improve any answer, and it would be a disaster if leaked.

Never paste these into an AI chat

None of these makes an answer better, and all of them cause real harm if exposed.

1. Passwords, PINs, one-time codes and security answers.
2. Full credit card, bank account or routing numbers.
3. Social Security, passport, driver’s license or tax ID numbers.
4. Other people’s private details (a friend’s diagnosis, a coworker’s address) without their OK.
5. Your employer’s confidential documents, unless your workplace has approved the tool.

Usually fine, and worth sharing

The things that make AI useful for daily life are mostly low-risk: your schedule, your habits, what you like to eat, your workout history, your goals for the week. Health details are in between. They make answers about sleep or food much better, but they are sensitive, so share them only with a service whose privacy terms you’ve read, and leave out anything you wouldn’t tell a stranger at a pharmacy counter.

Budgets are similar. “I spent $412 on groceries last month and want to get under $350” is useful and harmless. Your account numbers are not needed for that and should never be part of it.

Check these two settings first

1. Training. ChatGPT (Free, Plus and Pro) uses your conversations to improve OpenAI’s models by default; turn off “Improve the model for everyone” under Settings → Data controls. Claude lets you choose whether your chats are used for training and change it in its privacy settings; if you turn it off, new chats aren’t used.

2. History and deletion. See whether you can delete past chats, and whether a deleted chat is excluded from training. Both OpenAI and Anthropic explain this in their help centers. When you use an app that calls an AI model through the developer API (as most apps do), the rules are usually different from the chat product’s, so read that app’s own privacy policy.

Exactly what Complete OS sends to its AI

We think you should hold us to the same test, so here it is plainly. Complete OS’s assistant, Echo, asks for your permission the first time you use it. When you send Echo a message, the request includes data from your Complete OS account: habits and daily check-offs, workout history, calendar events if connected, meals and kitchen data, notes, budget and bill information, and up to 40 recent things Echo has remembered. It also includes health data such as sleep, steps, weight and vitals, including data you imported from Apple Health or Health Connect. That context is why Echo can answer “what should I focus on right now?” about your actual day instead of giving generic advice.

It goes to Anthropic (Claude), or to OpenAI if Anthropic is temporarily unavailable. Neither uses data sent through their APIs to train their models. Echo never receives your passwords, payment details, or the GPS route of walks you track. A photo you choose to scan (a nutrition label, a card in the Vault, a work schedule) is sent so the details can be read for you to review. You can withdraw consent at any time in Settings. The full list is in section 3 of our privacy policy.

Frequently asked questions

Is it safe to share personal information with AI?

Low-risk details like your schedule, goals and meal plans are generally fine with a reputable service. Never share passwords, full card or account numbers, government ID numbers or other people’s private information.

What should you not share with ChatGPT?

Passwords and one-time codes, full card and bank numbers, Social Security or passport numbers, other people’s private details, and confidential work documents your employer hasn’t approved.

Does ChatGPT train on my conversations?

On ChatGPT Free, Plus and Pro, yes by default. You can turn it off under Settings → Data controls → Improve the model for everyone; new conversations then aren’t used for training.

Can I stop Claude from training on my chats?

Yes. Claude’s privacy settings let you choose whether your chats are used for model training. If you turn it off, new chats aren’t used for future training.

What data does Complete OS send to its AI?

With your permission: habits, workouts, calendar, meals, notes, budget and bills, up to 40 Echo memories, and health data such as sleep, steps and vitals. It goes to Anthropic, or OpenAI as a fallback, and isn’t used for training. Passwords, payment details and walk GPS routes are never sent.