When you type a message into an AI tool, your words travel to the company’s servers and are processed by the model. What happens next depends on which tool you use, which plan you are on, and your privacy settings — and the defaults vary widely.
The short answer
Most free, consumer-facing AI tools can use your conversations to improve their models — unless you opt out. Enterprise and API tiers are different: they almost universally do not train on customer data by default. If you are using an AI tool for sensitive work, the plan tier matters as much as the tool itself.
What the main tools do with your data
ChatGPT (free tier) may use your conversations to train future OpenAI models by default. You can opt out in Settings → Data Controls → turn off “Improve the model for everyone.” Alternatively, Temporary Chats are never used for training. OpenAI’s API, Team, and Enterprise plans do not train on your data by default.
Claude (claude.ai) moved to an opt-in model in 2025: you choose whether to allow Anthropic to use your conversations for model training. If you opt in, data may be retained in de-identified form for up to five years; if you opt out, it is deleted within 30 days. Incognito chats are never used for training regardless. The Claude API and business plans do not train on data by default.
Google Gemini (free) trains on your conversations when the “Keep Activity” setting is on. Turn that setting off, and future conversations will not be used for training. Temporary chats are always excluded. Gemini in Google Workspace does not use your data to train models.
Microsoft Copilot (free/consumer) trains on your conversations by default, though you can opt out at any time. Microsoft 365 Copilot with a corporate account is not used for training and is covered by enterprise data protection commitments.
The pattern across all four: free consumer → trains by default (opt-out available); enterprise/API → does not train by default.
How to opt out in three steps
- Find your account’s privacy or data settings — look for labels like “Data Controls,” “Privacy Settings,” or “Activity.”
- Disable the training toggle: “Improve the model for everyone,” “Model Improvement,” or “Keep Activity.”
- Switch to temporary or incognito mode when you do not want any record kept at all.
If you cannot find these settings, check the tool’s help center. The opt-out process for each major tool is documented in their privacy FAQ.
What not to share with AI tools
Even after opting out of training, data is still sent to the company’s servers and processed there. A few categories to avoid sharing unless you are on a secured enterprise plan:
- Personal identification: ID numbers, passport numbers, tax codes
- Financial details: card numbers, bank account credentials
- Medical information: diagnoses, prescriptions, patient records
- Confidential business data: unreleased financials, M&A plans, client lists
- Legal documents under privilege
No AI privacy policy — not even enterprise ones — is a substitute for professional confidentiality obligations.
Your rights: GDPR and beyond
If you are in the European Union, GDPR gives you the right to access the data a company holds about you and to request its deletion. All major AI providers have processes for GDPR requests — check their privacy center or contact their Data Protection Officer.
The catch: deletion from AI training data is technically complex. Once data has been incorporated into model weights, it cannot be surgically removed without retraining the model. Most companies handle this by committing to exclude your data from future training runs and to delete stored copies; they cannot retroactively erase influence on deployed models.
Outside the EU, rights vary by country. Many US states have passed privacy laws (California’s CCPA, Colorado’s CPA) that offer similar access and deletion rights, though enforcement differs.
FAQ
Does opting out of training mean the company cannot see my conversations?
No. Opting out of training means your conversations are not used to train future models — but the company’s servers still process your messages to generate responses. Even opted-out conversations may be reviewed by safety teams in exceptional circumstances.
Is the paid/Plus tier more private than the free tier?
Not necessarily for the training question — the opt-out mechanism is the same. The real privacy upgrade comes with enterprise or team plans, which exclude data from training by default and often include stricter data isolation.
What if I use an AI through a third-party app?
The underlying AI provider’s policies still apply, plus the third-party app’s own policy. Always check both — the app layer can add additional data collection.
How long do AI companies store my conversations?
It varies: typically 30 days for API calls, longer for consumer products (especially if you have history enabled). Check the specific tool’s privacy page for current retention periods, as these change.