Anthropic now stores your full chats for up to 5 years when you press thumbs up or down

What you could do with it

• A company using Claude Free, Pro, or Max could let the thumbs-up/down button collect its chats so Anthropic trains future AI models ².
• A business owner on a Team or Enterprise plan could turn off the feedback feature so members' chats are never stored or used ¹.
• Anthropic researchers could study how users behave and measure service effectiveness from the de-linked feedback data ¹.

What exactly gets saved

When you press the feedback button on Claude, Anthropic keeps a copy of your entire chat history ¹. This includes everything you typed, any custom styles or settings you had active, and model configuration details ². The data is stored in a secured backend and can remain there for as long as five years .

However, there are important limits. Raw content pulled from connectors like Google Drive is not included — unless you personally copied and pasted it directly into your chat with Claude ¹. Remote and local MCP servers (tools that let the AI access outside data) are also excluded from feedback storage ².

Your feedback is also de-linked from your user ID before it's used by Anthropic ². This means the company strips away your email address and account information so the data can't be easily traced back to you personally ¹.

How the data can be used

Anthropic is upfront about what this feedback data might be used for. The company says it can be used to analyze how well its services work, conduct research, study user behavior, and train AI models ¹.

For consumer accounts — Claude Free, Pro, and Max — chats can also be used if you actively opt in to the Model Improvement program, or if your conversation is flagged during safety reviews ². Safety-flagged chats may be used to help the Safeguards team improve the system that detects and enforces the company's usage policies .

Incognito chats are an exception: they are never used to improve Claude, even if you have the Model Improvement setting turned on ².

What organizations can do

If you're part of a Team or Enterprise plan, a Primary Owner or account holder can turn off the feedback feature entirely for their members ¹. This is done through the Rate chats setting under Organization settings > Data and Privacy .

For commercial products like Claude for Work and the Anthropic API, Anthropic states that inputs and outputs are not used for model training by default — unless you explicitly report feedback or bugs through the thumbs up/down button or otherwise give permission ¹.

Why this matters for everyday users

The thumbs up and thumbs down buttons were designed as a simple way to help the company improve. But now those same clicks can create a permanent, de-linked record of your conversations that feeds into model training. If you share sensitive information in a chat and then hit thumbs up, that exchange could end up in the data used to build the next version of Claude — even though your personal identity is removed from it.

Key facts

• Full chats are stored for up to 5 years when using the feedback button
• Feedback data is de-linked from your user ID before use
• Data can be used to train AI models and conduct research
• Incognito chats are exempt from being used for model improvement
• Organization owners can disable the feedback feature for team members

Sources

1. Is my data used for model training? | Anthropic Privacy Center
2. Is my data used for model training? | Anthropic Privacy Center