Three things protect most people: switch off model training and review memory in the AI apps you use, agree a family code word so a cloned voice cannot fool you, and never paste anything into a chatbot that you would not post publicly. This checklist gives the menu paths we verified on the vendors' own help pages on November 1, 2026. Menus change often, so if a label differs on your screen, search the app's settings for the same words.
Key takeaways
- ChatGPT: Settings, Data controls, then turn off "Improve the model for everyone". Temporary chats are not used to improve models.
- Gemini: manage "Keep Activity" in Gemini Apps Activity. With it off, chats are not used to train models but are still kept for 72 hours, per Google.
- Copilot: turn off "Training on conversation activity" and "Training on voice conversations" under Privacy, and manage "Personalization and memory" separately.
- Voice-clone scams work on urgency. A family code word and a call-back rule beat any "detector".
- Training controls and memory controls are separate. Turning one off does not turn off the other. Turning training off does not delete saved chats, and what happens to saved memories differs by vendor, so check each one below.
Verified on November 1, 2026: what to switch off, by assistant
We checked each path against the vendor's help material. Where a vendor page did not give click-by-click steps we say so instead of guessing. Each assistant also behaves differently if you use it through a work or school account, where administrators may set the rules.
ChatGPT (personal Free, Plus or Pro accounts)
According to OpenAI's help center, data sharing for model improvement is on by default for personal accounts, and you can turn it off while keeping your chat history.
- Web, signed in: open your account menu, select Settings, then Data controls, then select "Improve the model for everyone", turn it off, and select Done.
- iOS and Android: open the sidebar, select your profile icon to open Settings, then Data controls, and turn off "Improve the model for everyone".
- OpenAI says turning this off does not delete or hide your saved chats, and that after you opt out, new conversations are not used to improve its models. Exceptions: if you give thumbs-up or thumbs-down feedback on a reply, that conversation may be used for training. Voice mode has its own control for sharing data for training. OpenAI's Privacy Portal lists additional privacy options.
- Temporary chat: OpenAI says temporary chats do not appear in history, do not create or update memories, are not used to improve models, and may be retained for up to 30 days for safety purposes. If you save a temporary chat it becomes a regular chat and follows your normal settings.
- Memory: memory and training are separate controls. OpenAI's privacy center lists Memory under "Personalized chats and ads", and notes that deleting a chat does not automatically delete a memory saved from it. Review saved memories and delete the ones you do not want.
Google Gemini (personal Google account)
- Manage the "Keep Activity" setting in Gemini Apps Activity (myactivity.google.com/product/gemini). Google says you can turn it off or review and delete activity at any time.
- With Keep Activity off, Google says future chats are not saved to your Activity or used to train models unless you send feedback. They are still kept for 72 hours, and some connected apps become unavailable.
- Auto-delete: the default is 18 months, and you can change it to 3 months, 36 months, or indefinite. Chats already reviewed by human reviewers are not deleted with your activity, and Google says those are kept for up to three years.
- Temporary chats are not used to train Google's models, and Google says they are still stored with your account for 72 hours.
- Memory: Google says you can view, edit, delete or switch off the memory summary at gemini.google.com/personalization-settings, and manage saved info at gemini.google.com/saved-info.
Microsoft Copilot (personal Microsoft account)
- On copilot.com: the same training settings may also appear under your profile and Privacy. Microsoft's page lists them for the apps below, so check Settings if you do not see them.
- Windows or macOS app: select your profile icon, open Settings, then Privacy, and turn off the same two settings.
- Mobile app: open the menu, select your profile icon, then Account, then Privacy, and turn off the same two settings.
- Memory: on copilot.com or the Windows and macOS apps, select your profile icon, then your profile name, then Memory, and turn off "Personalization and memory". In the Edge sidebar it is under the sidebar menu, Settings, Memory. Microsoft says that turning off Personalization and memory makes Copilot forget its memories of your conversations; your past conversations stay visible until you delete them.
- Microsoft notes that opting out of training does not exclude conversations from product improvement, advertising, safety, security or compliance uses, and that the personalization setting does not control personalized ads. Its page for deleting history did not give a menu path in the page we read, so check the linked article inside the app.
Not sure what is wrong? Get a diagnosis first.
Our certified tech can connect remotely, check your software, drivers and backups, and tell you honestly whether the problem needs a hands-on repair.
Training, memory and history are three different things
People often switch one setting and think they are covered. They are different controls with different effects.
| Setting | What it controls | What it does not do |
|---|---|---|
| Model training (improve the model, Keep Activity, training on conversation activity) | Whether your new chats can be used to improve the vendor's models | Does not delete saved chats, and may not cover safety, legal or support uses |
| Memory or personalization | Whether the assistant remembers facts about you across chats | Varies by vendor. Microsoft says Copilot forgets its memories when you turn personalization and memory off, so check what each vendor says |
| Chat history and auto-delete | How long your chats are kept in your account | Does not change what a vendor retains for safety review |
| Temporary or private chat | One conversation that skips history and memory | Still may be kept for a short period for safety, per vendors |
A good default for a personal account: training off, memory reviewed or off, auto-delete set to a short period, and temporary chat for anything sensitive. That balances privacy and convenience without giving up the tool.
What not to paste into a chatbot
Settings reduce risk, but the safest data is data you never share. A simple rule: if you would not put it on a public forum, do not paste it unless your organization has approved the tool for that data.
- Passwords, recovery codes, private keys and API keys. Never. Rotate any that you pasted by mistake.
- Government ID and financial details: SIN or SSN numbers, passport and licence numbers, full card or bank numbers.
- Health records with your name, and other people's medical details.
- Client or employer data: contracts, customer lists, unreleased financials, source code under NDA. Our guide on shadow AI explains why this is the business risk to watch, and the AI usage policy guide shows how to set rules.
- Other people's private details (a friend's problems, a colleague's complaint) that they did not agree to share.
- Anything sent to a browser extension or agent that can read pages. See AI browser agents and security risks.
If you need help with a document that contains sensitive details, remove or replace names, numbers and addresses first. Use placeholders such as "Client A" and "[amount]". Also remember that anything you share may appear in an account that someone else can open, such as a shared family computer.
AI voice-clone scams and the family code word
Scammers can now clone a voice from a short audio clip, such as a social media video, and use it in a phone call: "Mom, I'm in trouble, I need money now." The caller ID can be spoofed too. The giveaway is almost never the voice quality. It is the pattern: urgency, secrecy, and a payment method that cannot be reversed. Our article on AI voice cloning scams covers the business side.
Set up a family code word in five minutes
- Pick a word or short phrase that is not guessable and not on social media: not a pet's name, not a street.
- Agree in person or by a trusted channel who uses it and when: any request for money or urgent help by phone.
- Agree on a second rule: hang up and call the person back on the number you already have, then ask for the code word.
- Share it with older relatives first. They are often the targets.
- Change it if it was ever said aloud on a call you did not trust.
- For a business, agree a verification step for payments: a call back to a known number and a second approver.
What to do during a suspicious call
- Do not stay on the line. Hang up and call back on a saved number.
- Ask for the code word, or a question only the real person could answer.
- Never pay by gift card, crypto or wire at someone's urgent request.
- Tell your bank at once if you already sent money, and report the call through your country's fraud-reporting service (for Canada, the Canadian Anti-Fraud Centre at reportcyberandfraud.canada.ca).
Deepfake checks that actually help
No free tool reliably detects every deepfake, and people who rely on one get a false sense of safety. These habits work better.
- Check the source, not the pixels. Find the same claim on the outlet's or person's own verified channel.
- Look for context mismatch: an unusual request, a new account, no original upload, or comments turned off.
- Slow down for emotional triggers. Fear, urgency and "limited time" are the scammer's tools.
- Use a second channel. Video call plus text on a known number; for a boss's request, an in-person or known-line confirmation.
- Treat live video with care. Ask the person to do something unscripted, though a skilled fake may still pass. A verified call-back is stronger.
- Be careful what you post. Public videos of your voice and face can feed a clone. You do not have to disappear, but consider limiting who sees family clips.
Seven-minute AI privacy tune-up
- Open ChatGPT settings and turn off "Improve the model for everyone", then review memory.
- Open Gemini Apps Activity, decide on Keep Activity, and set auto-delete to a period you accept.
- Open Copilot's Privacy settings and turn off the two training settings, then review memory.
- Remove old chats that contain personal details you should not have shared.
- Check which browser extensions and apps have access to your AI accounts and delete the ones you do not use.
- Turn on two-step verification on every AI account, since a stolen account exposes your whole chat history.
- Agree a family code word.
Want the other side, which assistants are worth using? See our best free AI tools and the ChatGPT, Claude, Gemini and Copilot comparison.
When to call a professional
Call if you pasted a password or client data into a chatbot and are unsure what to rotate, if an AI account was taken over, if you clicked a link in a suspected AI-generated phishing message (see how to spot AI-generated phishing), or if you want your family's or office's devices and accounts reviewed. Our remote repair team can review settings with you step by step.
FAQ
How do I stop ChatGPT from training on my chats?
Open Settings, then Data controls, and turn off "Improve the model for everyone". OpenAI says this keeps your history and stops new conversations being used to improve its models. Exceptions: thumbs-up or thumbs-down feedback on a reply may let that conversation be used for training, and voice mode has its own training control.
Does turning off training delete my past chats?
No. Vendors say these are separate controls. Delete chats and review memory separately if you want them gone.
Does Gemini keep chats if Keep Activity is off?
Google says chats are not saved to your Activity or used to train models, but are still kept for 72 hours.
How do I opt out of training in Copilot?
Under Privacy, turn off "Training on conversation activity" and "Training on voice conversations". The exact path differs for copilot.com, the Windows or macOS app, and the mobile app.
What is a family code word for?
It lets you check that a caller asking for money or urgent help is who they say they are, even if the voice sounds exactly right.
Can AI really clone a voice from a short clip?
Voice-cloning tools exist and scammers have used them, so do not trust a voice alone. Verify by calling back on a number you already have.
Is it safe to paste work documents into a chatbot?
Only if your organization has approved the tool for that data. Otherwise remove names, numbers and identifying details first.