What to know: everything you type, and everything the AI reads on your behalf, is sent to the provider and kept. Two settings decide what happens to it after that.
The training toggle. Whether your conversations can be used to train future models. It varies by product and plan, and most people have never looked.
Retention. How long your conversations and logs are kept. Some plans let you set this. Most people don't know it is a setting at all.
For Claude, both live at https://claude.ai/settings/data-privacy-controls. The training toggle is called "Help improve Claude". Other providers keep theirs somewhere similar; your AI knows the product you are on, so ask it where.
Once something is in, it is kept. You should assume that one day someone who is not you may see it, and that nobody knows what the follow-on effects of that could be.
Author's practice: decide the training toggle once, per product, before putting anything real in. Set retention deliberately, long where the record matters. Then choose what goes in with open eyes. Knowing it is kept decides what level of information you consciously give it. Not everything does.
Reveal on the spot: say whether you can see the training toggle and the retention setting from here. If you can't, tell the person exactly where to look in the product they are on. Say what has been shared in this conversation so far: what they typed, and what you read on their behalf.
Unlocks: putting real information in without a nagging doubt, because the decision was made on purpose.