Every conversation you have with AI is data. Some of it trains the next model. Some of it sits in logs. All of it deserves more attention than you're giving it.
What gets collected
By default, all three major providers store your prompts, responses, and metadata (device, location, timing). Some use the content to improve future models unless you opt out.
ChatGPT
- Free and Plus: prompts may be used for training unless you turn off "Improve the model for everyone" in settings.
- Team and Enterprise: never used for training by default.
- Retention: 30 days minimum, longer if flagged.
Claude
- Default: Anthropic does not train on consumer chats.
- Retention: 30 days, 90 days for flagged content.
- API: same — no training by default.
"Treat every AI prompt like an email to a stranger. Because it kind of is."
Gemini
- Free: Google may use prompts to improve services and models.
- Workspace: not used for training, kept within your domain.
- Activity is stored in your Google Account by default — review at myactivity.google.com.
What to change today
- Turn off training in ChatGPT: Settings → Data Controls → Improve the model.
- Review Gemini Activity: myactivity.google.com → Gemini Apps Activity → Auto-delete after 3 months.
- Use Claude for sensitive drafts: lowest training risk by default.
What to never paste
- Passwords, API keys, full account or card numbers.
- Other people's personal data without consent.
- Confidential work documents (use your employer's enterprise tier instead).
- Medical records or anything you'd want HIPAA-protected.
Local AI: the privacy answer
Open models like Llama 3, Mistral, and Yi-Lightning 2 now run on consumer laptops via Ollama or LM Studio. For sensitive work, this is the privacy upgrade most people are sleeping on.
The bottom line
You can't outsource judgement on privacy. Adjust the settings, learn what each provider does by default, and move sensitive work to local models. Five minutes today saves a lot of regret later.