Here is an uncomfortable truth: the default settings of most AI tools are not designed for your privacy — they are designed for the company’s data appetite. Your chats train future models, your voice clips improve speech recognition, and your usage patterns feed advertising profiles. None of this is secret; it is written in the privacy policies almost nobody reads.
The good news? You can fight back in about twenty minutes. As a European, you have the GDPR on your side — the strongest data-protection law in the world. This checklist walks you through the AI privacy settings you should change today, across ChatGPT, Gemini, Claude and beyond, plus how to exercise your legal rights when settings are not enough.
Why the defaults are against you
AI companies improve their models with real user data. That creates a structural incentive to set the defaults to maximum collection: chat history saved forever, conversations used for training, data shared with “service providers” and advertising partners. Changing a default requires the company to give up something valuable, so the privacy-friendly option is usually buried three menus deep — if it exists at all.
There is also a second problem: many people paste genuinely sensitive material into AI chats — CVs, medical questions, business plans, children’s homework, client data. Under the GDPR, much of this counts as personal data, and some of it (health, religion, political views) is special-category data with extra protection. Every setting below reduces the chance that a casual chat today becomes a data problem tomorrow.
The checklist: 10 settings to change today
1. ChatGPT: opt out of model training
Open ChatGPT → your profile → Settings → Data controls → “Improve the model for everyone” and turn it off. This stops OpenAI from using your conversations to train future models. Note the nuance: turning this off does not delete past chats or stop OpenAI storing them — it only stops training use. For genuinely sensitive questions, use a Temporary Chat (click the ChatGPT model name at the top → “Temporary chat”), which is not saved to history.
2. Gemini: stop activity saving (or auto-delete it)
Google’s Gemini saves your prompts to “Gemini Apps activity” by default, and human reviewers may read a sample of them. Go to myactivity.google.com → Gemini Apps activity → Turn off, or set auto-delete to 3 months. Remember that turning activity off does not stop Google keeping some data for security and abuse prevention — read the fine print — but it sharply reduces the retained profile.
3. Claude: check your privacy controls
Anthropic’s Claude does not use your conversations for training by default on consumer plans — one of the better defaults in the industry — but verify it. Go to Settings → Privacy and confirm training use is off. Also review which integrations (Google Drive, Gmail) you have connected; each connection expands what the assistant can see, so disconnect anything you no longer use.
4. Limit ad personalisation and data sharing
Several AI apps — especially free ones — share data with advertising and analytics partners. In each app’s settings, look for “Personalised ads”, “Data sharing” or “Third-party sharing” and disable them. On your phone, also check the OS level: on iOS, Settings → Privacy & Security → Tracking (deny tracking requests); on Android, Settings → Privacy → Ads → Delete advertising ID. These two minutes cut off a surprising amount of cross-app profiling.
5. Review mobile app permissions
AI apps often request microphone, camera, contacts and location access. Ask of each permission: does this app need it to do the job I hired it for? A chatbot does not need your contacts or precise location. On both iOS and Android, go to Settings → Apps → [AI app] → Permissions and revoke anything non-essential. Set location to “Never” unless the feature genuinely requires it.
6. Delete — or auto-delete — your chat history
Old chats are a data breach waiting to happen: they contain the unfiltered version of your questions. Once a month, archive or delete conversations you no longer need. In ChatGPT: Settings → Data controls → Archived chats → Manage, and delete from there. Better still, enable any auto-delete option the tool offers. Treat your AI chat history like your email inbox: a tidy one is a safer one.
7. Manage voice recordings and audio data
Voice mode is convenient — and it creates biometric-adjacent data. Check whether your AI app stores voice recordings separately from text chats, and whether human reviewers can listen to them. In app settings look for “Voice data”, “Audio recordings” or “Speech” and disable storage or enable auto-deletion. If you dictate sensitive content, prefer typing: text is easier to review before sending, and easier to delete afterwards.
8. Be careful on public Wi-Fi — use a VPN
That café Wi-Fi in the train station is a classic interception risk. AI chats can contain passwords, personal details and business secrets — exactly what you do not want sniffed on an open network. Avoid sensitive AI use on public Wi-Fi; if you must, use a reputable VPN and confirm the site uses HTTPS. At home, keep your router firmware updated — your AI conversations travel through it.
9. Check the DPA and data residency before you sign up
Before giving a new AI tool your data — especially for work — check two things: does it offer a data processing agreement (DPA), and where is data stored? EU-based storage (or an EU data boundary from a US provider) plus a signed DPA is the baseline for business use. No DPA and vague “we may store data globally” language is a red flag. Our guide on how to check if an AI tool is GDPR-compliant before you sign up walks through the full vetting process.
10. Use temporary, anonymous or local modes where available
Several tools now offer low-retention modes: ChatGPT’s Temporary Chat, anonymous browsing without an account, and — the strongest option — open-source models running locally on your own machine, where no data leaves your device at all. Match the mode to the sensitivity: everyday brainstorming in a normal chat is fine; a draft redundancy letter belongs in a temporary or local session.
How to exercise your GDPR rights
Settings are the first line of defence; your legal rights are the second. Under the GDPR you can, free of charge:
- Right of access: ask any AI company for a copy of the personal data it holds about you.
- Right to erasure: ask them to delete it — the famous “right to be forgotten”. This is especially powerful for training data and old chat logs.
- Right to object: object to processing based on “legitimate interests”, which is the legal basis many companies use for model training.
- Data portability: get your data in a usable format and take it elsewhere.
How to do it in practice: look for the company’s privacy request form or “data subject request” page (usually linked in the privacy policy footer), or email their Data Protection Officer directly. They must respond within one month. If they refuse or stall, you can complain to your national data protection authority — in Germany the BfDI or state DPAs, in France the CNIL, in the Netherlands the Autoriteit Persoonsgegevens. For a step-by-step walkthrough, read AI tools and the right to be forgotten: how to delete your data.
Your 20-minute action plan
You do not need to do everything today — but do the big three: turn off model training in ChatGPT, disable Gemini activity saving, and delete old chats containing anything sensitive. Then work through the rest of the checklist this week.
Privacy with AI tools is not a one-time fix; companies change settings, merge, and update policies. Put a twice-yearly reminder in your calendar to re-check the key toggles. Twenty minutes today, ten minutes twice a year — a small price for keeping your conversations yours.
2 thoughts on “AI Privacy Settings You Should Change Today: A European’s Checklist”