Using AI assistants without oversharing
AI assistants are now part of ordinary digital life: drafting emails, summarising documents, planning trips. They are useful, and they are also a new place where personal information ends up.
Reviewed November 2026 · 5 minute read
Treat the chat box like an email to a stranger
What you type is sent to the provider’s servers. Depending on the service and your settings, it may be stored, reviewed for safety and quality, or used to improve future models. The simplest rule: do not paste anything you would not put in an email to a company you do not know well.
Keep these out
- Passwords, one-time codes, recovery phrases and security answers
- Full card, bank or government identity numbers
- Other people’s private details, including medical or legal matters
- Confidential work documents, unless your employer has approved that specific tool
- Photos of ID documents
You can usually get the same help by removing names and numbers first. “Rewrite this complaint letter” works just as well with the account number replaced by XXXX.
Three settings worth two minutes
- Training and data use. Look in privacy or data controls for the option that governs whether your conversations are used to improve models, and choose deliberately.
- Chat history and memory. Decide whether you want past conversations kept, and delete those containing anything sensitive. Many assistants offer a temporary mode that is not saved to history.
- Connected apps. If you link email, calendar or files, review exactly what access you granted and remove what you do not use.
Check the answer, not the confidence
Language models generate plausible text. They can state wrong facts, invent sources and make arithmetic slips in a perfectly assured tone. Match your checking to the stakes:
- Low stakes (a recipe swap, a draft message): read it over and go.
- Medium stakes (a product comparison, a how-to): ask for sources and open at least one.
- High stakes (health, legal, tax, money): use the answer to prepare better questions for a qualified person, not as the decision.
AI on the other side
The same technology makes scams more convincing: fluent messages, cloned voices and fake video. Verification habits matter more than detection skills. Agree a family safe word, and confirm any urgent request for money by calling back on a number you already have. See five signals that expose a scam message.
Children and AI
Check the minimum age for the service, use supervised or family accounts where available, and talk about two things early: assistants can be wrong, and they are not a private diary.