Claude model training data: what the help article says

For Claude model training data, the help article draws one line. By default, Anthropic does not use inputs or outputs from its commercial products to train models, but feedback you submit can be used. Consumer plans, including Claude Code on those plans, are covered by a separate article, so the first check is which plan an account is on.

The screenshot shows Hacker News thread: Is my data used for model training?.

What the article says about Claude model training data

The article is short. Its scope is commercial products such as Claude for Work and the API, and it points consumer plans elsewhere. The training default is “no”, with an exception for things you choose to send.

The mechanism is simple: the feedback button is the opt-in. Press it and the related conversation is stored, including settings and custom styles, and may then be used for research and training.

What the Hacker News thread pushed back on

The thread, titled “Is my data used for model training?”, centred on Claude Code. One commenter argued that even just dismissing the feedback may be considered “feedback”. Another called the prompt a dark pattern.

Replies cut both ways. A commenter pointed to the Claude Code data usage page, which says the session quality survey, even when dismissed, records a rating and not transcripts. The original poster answered Great to know there’s clarity on this, while noting the page had not existed when the first post was written.

None of this is verified beyond the two pages. The commenters are users, not Anthropic, so the thread is a map of worries and not evidence of practice.

Settings a Claude Code user should check

Start with the account type, since the commercial article does not cover Pro or Max. Then look at how feedback is handled. Hooks can also keep sessions tidy, as covered in Claude Code hooks, though they do not change what the vendor stores.

[ ] Confirm the plan: commercial (Work, API) or consumer
[ ] Read the consumer article if the account is Free, Pro or Max
[ ] Avoid thumbs up/down on chats holding private code
[ ] Treat the session survey as a rating, per the data usage page
[ ] Team or Enterprise owner: review the Rate chats setting
[ ] Keep secrets out of pasted text; connectors are excluded, pastes are not

Survey feedback: whether dismissing the Claude Code survey counts as feedback is answered on the data usage page cited in the thread, not in the training article.

Owners on Team or Enterprise plans can disable the ability to submit feedback with the Rate chats setting under Organization settings, then Data and Privacy. Budget-minded readers may also want Claude Code usage limits, since privacy and quota questions often arrive together.

Verdict

On commercial plans, the Claude model training data default is no training, and the exceptions are feedback and explicit consent. Consumer plans follow a different article, so check that one if the account is Pro or Max. The Claude Code survey question is answered by a separate docs page, not this article.

Check your own plan: read the help article