How to use ChatGPT, Claude and Gemini safely

The seven rules

**Share the minimum.** Leave out names, addresses, account numbers, health details and anything about other people that they haven't agreed to share.
**Check what matters.** Health, money, legal, safety and anything you will publish or act on: verify it against an official or expert source.
**Ask for sources, then open them.** A citation you haven't read is not a source.
**Set your privacy controls once.** Decide whether your chats can be used for training, and turn on temporary chats for sensitive topics.
**Keep a human in charge of actions.** Don't let an assistant or [agent](/glossary/ai-agent) send, buy, sign or delete anything without your confirmation.
**Be open about AI help.** At school and at work, follow the rules and say where you used AI.
**Never trust a voice or video alone for money.** Scammers clone voices and faces. Hang up and call back on a number you know.

What never to paste into a chatbot

Anything you type may be stored on the company's servers, reviewed by staff for safety, kept for a period after you delete it, and — depending on your settings — used to train future models. Treat a chatbot like a helpful stranger, not a vault.

Information to keep out of AI chats, and safer alternatives
Don't paste	Why	Do this instead
SIN, passport, driver's licence, health card numbers	Identity theft if leaked	Replace with "[ID number]"
Passwords, PINs, banking logins, card numbers	Direct financial risk	Never needed — leave them out
Medical records, test results with your name	Highly sensitive; hard to retract	Describe the question without identifiers; ask your clinician
Other people's private details (clients, students, patients, family)	Not yours to share; may break privacy law or professional rules	Anonymise: "a client in her 40s"
Confidential work documents, contracts, source code	May breach your employer's policy or an NDA	Use the tool your employer approved, under its business terms
Photos of children or of IDs, or screenshots with names visible	Faces and documents are among the most sensitive data	Crop or blur first, or describe in words

Privacy settings to check

Every major assistant lets you control some of what happens to your chats. Names and menus change often, so look for these kinds of controls in Settings (names as of our October 2026 review):

Where to find the main privacy controls
Control	What it does	Usually found as
Training on your chats	Decides whether your conversations can be used to improve future models	ChatGPT: Data controls → "Improve the model for everyone". Claude: Privacy → "Help improve Claude". Gemini: "Gemini Apps Activity" / "Keep Activity". Copilot: Privacy → model training
Temporary or incognito chat	A chat that isn't saved to your history or used for memory or training	A "temporary", "incognito" or ghost icon when starting a new chat
Memory	Lets the assistant remember facts about you across chats	Settings → Personalisation / Memory: view, delete or turn off
Connected apps	Access to your email, calendar, files or other tools	Settings → Connectors / Apps / Extensions: remove what you don't use
Delete and export	Remove old chats or download a copy of your data	Settings → Data / Account

Good to know

Even with training switched off, companies usually keep chats for a limited time to detect abuse, and safety staff may review flagged conversations. Business and education plans generally offer stronger guarantees — that's why employers prefer them.

Check before you trust

Assistants write fluently whether they are right or wrong. [Hallucinations](/glossary/hallucination) are most likely with specific facts: names, numbers, dates, quotes, laws, prices and references.

Ask: "What are your sources? Which parts are you less sure about?"
Open every source it gives and check that it says what was claimed.
For anything official — taxes, benefits, immigration, health — go to the government or professional body's own page.
Read laterally: search the claim in a new tab and see what trusted outlets say.
Ask the assistant to argue the opposite. Weak answers often fall apart.

Medical and mental health

AI can help you prepare questions for a doctor, but it is not a clinician. If you or someone you know is in crisis, call or text 9-8-8 (Suicide Crisis Helpline, Canada) or call 911.

Scams that use AI

**Voice-clone emergencies.** A call in a loved one's [cloned voice](/glossary/voice-cloning) asks for money urgently. Agree on a family code word; hang up and call back.
**Deepfake investment ads.** Videos of celebrities or politicians "endorsing" trading platforms or crypto. Real public figures don't sell investments in social-media ads.
**Polished phishing.** AI removes the spelling mistakes that used to give scams away. Check the sender's address and never log in from a link in a message.
**Fake AI apps.** Look-alike "ChatGPT" or "Gemini" apps that charge subscriptions or steal data. Download only from the company's official site or verified store listing.

If you've been targeted, stop contact, call your bank, and report it to the [Canadian Anti-Fraud Centre](https://antifraudcentre-centreantifraude.ca/) (1-888-495-8501) and your local police.

Children and teens

Check each service's minimum age before a young person signs up: Claude.ai requires users to be 18; ChatGPT requires 13 and a parent's permission under 18; other services vary and change. Talk about AI the way you talk about the internet: what it's good for, what to keep private, and that it can be wrong.

For ages 8 to 17, our [Young Lab](/labs/young) teaches how AI works through games that collect nothing and have no open chatbot. Parents and teachers have their own [guide](/labs/young/grown-ups).

At work

Use the AI tools your employer has approved, signed in with your work account. Business plans usually exclude your data from training.
Never paste client, patient or student information into a personal account.
You remain responsible for what you send. Review every AI draft as if you wrote it.
If there's no policy yet, ask for one — our [small-business guide](/guides/ai-for-small-business#policy) has a one-page template.

Your wellbeing

Chatbots are designed to be agreeable and always available, which can make them feel like a friend. Use them as a tool, keep real people close, and be careful about leaning on an AI for emotional support or major life decisions. If a conversation leaves you feeling worse, step away.

Frequently asked questions

Is ChatGPT safe to use?

For everyday tasks, yes, if you keep personal and confidential information out, check important facts, and review your privacy settings. The same applies to Claude, Gemini, Copilot and other assistants.

Do AI companies read my chats?

Automated systems scan chats for abuse, and trained staff may review a small number, especially flagged ones. Depending on your settings, chats may also be used to train models. Turning off training and using temporary chats reduces this.

Can I delete what I've shared?

You can delete chats from your history, and they are then removed from the company's systems after a retention period. Data already used to train a model can't practically be removed from it, which is why it's best not to share sensitive information in the first place.

Which AI assistant is the most private?

It depends on settings and plan more than brand. Business and education plans with training off are generally the most protective; an open-weight model running on your own device keeps data entirely local.

How do I report an AI scam in Canada?

Contact your bank first if money is involved, then report to the Canadian Anti-Fraud Centre (1-888-495-8501 or its online system) and your local police. Report fake ads to the platform that showed them.

Words to know

hallucination prompt ai memory data retention voice cloning deepfake prompt injection personal information