Giving chat feedback gives Anthropic your data

For Anthropic’s commercial products, including Claude for Work, the API, and Claude Gov, prompts and responses are not used to train models by default. That changes if users submit feedback or bug reports, such as through the thumbs-up or thumbs-down button, or otherwise permit data use: Anthropic may then use related chats and coding sessions for service analysis, research, user-behavior studies, and model training, subject to applicable law. The guidance is specifically about commercial offerings; consumer plans such as Claude Free, Pro, and Max are covered separately.

Submitting feedback stores the entire associated conversation in a secured backend for up to five years, including its content, custom styles, preferences, and model settings. Raw content from connected services, such as Google Drive or remote and local MCP servers, is excluded unless it was copied directly into the chat. Before feedback is used, it is separated from user and customer IDs, and it is not combined with the person’s other Claude conversations. Team and Enterprise Primary Owners or Owners can prevent organization members from submitting feedback through the thumbs buttons by turning off the “RATE CHATS” setting in Organization settings under Data and Privacy.