By Dan Donovan, @askdandonovan
There's one setting that decides whether your conversations are kept for up to five years or for thirty days. You already answered it. Almost nobody remembers which way.
Anthropic has a switch called Help Improve our AI models. It decides one thing: whether your chats and coding sessions get used to help train future versions of Claude.
You've already answered it. If you signed up recently it was part of the signup flow. If you'd been using Claude for a while it appeared as a popup, and from 8 October 2025 you had to answer it before you could keep using Claude at all. Most people were trying to get to the thing they opened Claude for, and clicked whichever button made the box disappear.
That answer is still in effect today. It didn't expire and it hasn't asked again.
If you read nothing else, copy this and work down it. Every line is explained further down the page.
CLAUDE PRIVACY CHECK, about 2 minutes
[ ] Settings > Privacy > "Help Improve our AI models"
[ ] Note which way it is currently set
[ ] Decide: on means up to 5 years kept, off means 30 days
[ ] Change it if it is not what you want
[ ] Go through chat history, delete anything you would not
want used for training
[ ] Remember: deletion is forward only, finished models keep
what they already have
[ ] Settings > Memory, a separate switch, off by default
[ ] Work account (Team or Enterprise)? None of this is yours
to change, ask whoever runs itThis is the part worth knowing, because the two options are a long way apart.
Neither answer is wrong, to be clear. Sharing your chats genuinely does help the models get better, and plenty of people are happy with that. The point is that it should be a decision rather than a reflex.
Open Claude, on desktop or in the app, then:
Settings > Privacy > Help Improve our AI models
That's the whole setting. One switch. Look at which way it's set, because that's the answer you gave, whenever you gave it. You can change it whenever you want, and change it back. Turning it off means new chats and coding sessions won't be used for future training from that point on.
This covers Claude Free, Pro and Max, including Claude Code used from those accounts.
Changing the switch handles what happens next. It does nothing about the conversations already in your history.
So go back through them and delete the ones you'd rather weren't there. Most people have a few. Old work things, something about your health, a business idea, the messy first draft of something personal.
Anthropic's wording is that if you delete a chat from your history, it will not be used to train future models. Deleted conversations leave your chat history immediately and leave their back end storage within thirty days.
Worth thirty seconds, because Memory gets confused with the training question constantly and it's a different feature with a different switch.
Settings > Memory
Memory is Claude remembering things about you between conversations. Your role, what you're working on, how you like things written, your technical preferences. It's off by default, so if you've never turned it on there's nothing in there. If you have, that same screen shows you everything Claude has stored, grouped by category, and you can edit or delete individual entries.
Turning one off does nothing to the other.
Better to know now than to assume you've undone more than you have.
It only works going forward. Anthropic says your data stays included in training runs already in progress, and in models that have already been trained. A model that has finished training doesn't unlearn because you changed a setting afterwards. So the accurate way to think about this is that you're stopping the flow from here, not reaching back into the past.
Everything on this page is from Anthropic's own pages, not from coverage of them. Worth saying, because most of what's written about AI privacy settings is a summary of a summary.