On Claude's consumer plans, pressing thumbs up or thumbs down stores the full related conversation in Anthropic's back end for up to five years. Anthropic removes the link to your user ID before using it and says it may be used to train models where legally permitted.
What Anthropic says
From the privacy center, for Free, Pro, Max and Claude Code accounts on those plans:
- Using the thumbs up or down button stores the full related conversation for up to five years.
- Anthropic removes the link to your user ID before using the feedback.
- It may be used to train AI models where legally permitted.
- It is not combined with your other conversations.
A second article states that data tied to feedback, such as thumbs and bug reports, is kept for 5 years. A third lists feedback among the cases where chats can be used for training even when you have not opted in generally.
Why people overlook it
A thumb feels like a tiny signal, like a rating on a recipe. The data it creates is the whole conversation, not a score. If the chat holds your medical question, a draft of a hard email or an address, the thumb sends all of it.
How to give feedback safely
- Look at the conversation first. Scroll up for names, numbers and pasted documents.
- If it is sensitive, skip the button. Describe the problem to the vendor in your own words instead.
- If you must report a bug, reproduce it in a fresh chat with made up details, then press the button there.
- Use incognito for tests, though check how that interacts with feedback in the product before relying on it. The pages we read do not say.
A family rule
Tell everyone who shares your account that the feedback buttons share the entire chat. A teenager pressing a thumb on a homework chat is low risk. A parent pressing it on a chat about a child's health is not.
What this does not mean
Using thumbs is not a mistake in general. Feedback is how the product improves, and Anthropic says it removes the user ID link first. The point is to choose it knowingly.
Quick check
Does a thumbs up send only that one message?
No. Anthropic says the full related conversation is stored.
Does turning off model improvement stop this?
Anthropic lists feedback as a separate path. Do not assume the toggle covers it.
Sources
- Is my data used for model training? (Anthropic Privacy Center)
- How long do you store my data? (Anthropic Privacy Center)
- How Do You Use Personal Data in Model Training? (Anthropic Privacy Center)
- How do I change my model improvement privacy settings? (Anthropic Privacy Center)
Last reviewed: 19 October 2026.
