Dismissing Claude's request for feedback is itself considered feedback

Dismissing the “How is Claude doing this session?” prompt in Claude Code counts as feedback, according to keydiscussions.com's report. On September 29, 2025, a paying power user wrote that tapping “0” to Dismiss made Claude thank them for the feedback and for helping to improve Anthropic’s models. The line vanished after about two seconds, so that reply is a paraphrase. A day earlier, Dismiss was only a theory that might hand the session to Anthropic for model training.

This piece covers how Dismiss became feedback, how Anthropic’s policy stores and uses it, and who is exposed by answering the Claude Code prompt with 0, 1, 2, or 3. It is for users who opted out of helping “improve Anthropic AI models,” for readers comparing that rule with OpenAI’s feedback clause, and for people paying extra to keep chats from training. The author says they are not a lawyer.

Dismissing Claude's request for feedback: what actually changed

A day before September 29, the writer treated “How is Claude doing this session?” as a privacy loophole in Claude Code. Typing 1, 2, or 3, and maybe even 0, might expose the whole session, and readers could only speculate. They wrote “shame on the product people” for not explaining the sacrifice to users who opted out of helping “improve Anthropic AI models.” They called the prompt a slap in the face for people paying hundreds of dollars a month.

Tapping “0” to Dismiss is feedback, a word the writer calls important in privacy policies. Claude thanked them for the feedback and for helping to improve Anthropic’s models. The line lasted about two seconds. That is not what they or others intend by pressing Dismiss. They assume it is not a typo and want clarification from Anthropic. They say classifying the tap as privacy-defeating feedback could run afoul of contract law. They clicked it so others would not have to.

Dismissing Claude's request for feedback: how it works

Under the quoted thumbs up and thumbs down rule, Anthropic stores the entire related conversation, including content, custom styles, or conversation preferences, in its secured back-end for up to five years. Feedback data excludes raw content from connectors such as Google Drive, including remote and local MCP servers, unless that data is copied into the chat. Anthropic may use the feedback to analyze the effectiveness of its Services, conduct research, study user behavior, and train its AI models as permitted under applicable laws.

The same passage says Anthropic does not combine that feedback with the user’s other conversations with Claude. The writer treats the Claude Code prompt as this kind of feedback. Keys 1, 2, and 3 were scores they feared. Key 0, Dismiss, joined them on September 29, when Claude thanked the user for helping to improve Anthropic’s models. The five-year store is written for the thumbs control, while the prompt asks for a number. Copied connector text can enter that store.

Dismissing Claude's request for feedback: why it matters now

Anthropic attracted the writer with stronger privacy guarantees out of the gate. Later changes to the privacy policy made them suspicious, including how the company handled the change. They pay a premium for plans that claim tighter privacy, and they say Claude Code should not show “How is Claude doing this session?” after an opt-out from helping “improve Anthropic AI models.” Dismiss still produced thanks for helping to improve those models. That mismatch is what the September 29 post confirms.

OpenAI’s caveat says that if you choose to provide feedback, the entire conversation associated with that feedback may be used to train its models, for instance by selecting thumbs up or thumbs down on a model response. The wording leaves feedback open. A follow-up that starts “Good job. Now…” is, in their view, feedback. They do not know whether OpenAI would train on that chat. They ask if they can get the boundary in writing or should assume zero privacy.

Dismissing Claude's request for feedback: who is affected

At OpenAI the writer pays $50 a month for a business account with two seats, against $20 a month for an individual plan, because of stronger privacy promises. They do not need the extra seat, so they are paying $30 more each month. They say the extra $30 could be saved under zero privacy. At Anthropic, users paying hundreds of dollars a month still see “How is Claude doing this session?” in Claude Code after opting out of helping “improve Anthropic AI models.”

Keys 1, 2, and 3 were the ratings in the earlier post. Key 0, Dismiss, is in that set because Claude treated the tap as feedback. The writer pressed it so others would not have to and says not to interact. Raw content from Google Drive and from remote or local MCP servers stays out of feedback data unless someone copies it into the chat. The writer says they are not a lawyer, so the contract-law point is their opinion.

Dismissing Claude's request for feedback: what to watch

The writer wants clarification from Anthropic either way and assumes the thanks for the feedback is not a typo. They say a fair case could be made that calling the tap privacy-defeating feedback runs afoul of contract law. Until that clarification, their rule is to ignore “How is Claude doing this session?” and not press 0, 1, 2, or 3. The quoted policy attaches the five-year store, training, and the non-combination promise to the thumbs up and thumbs down button.

OpenAI’s clause names thumbs up and thumbs down and, the writer argues, could cover a later message that begins “Good job. Now…”. They never use those buttons. They still ask whether to get the training boundary in writing or save the extra $30 a month and assume zero privacy. The earlier post expects whatever principles founders had, including privacy, to keep eroding as AI startups burn through an unprecedented amount of cash, and it says to be careful out there.