What we send to Claude

Read Master's AI features use Claude, a model made by Anthropic. Here is what leaves our servers when you use them.

What we send

Only the text a feature needs:

• Pre-reading guides: passages picked from across the book.
• Explanations: the passage you selected and up to 500 words before it, or a picture and the text around it.
• Other AI features: the chapter or passage they work on.

Each request also carries basic details such as which feature it is for. We never send your password or payment details.

Sends you don't trigger yourself

If AI is on, two things are sent without a separate request:

• a token count before each request, to check its size, using the same content;
• a batch of an imported book's pictures and captions, so they can be described for screen readers. It runs once at import, and again for any chapter you open after changing your AI output language.

Each week we also send a small quality sample, so we can check our AI's work:

• a few saved guides, with the book's title and author;
• a few simplified chapters, each with the original chapter. Only public-domain books are sampled.

Nothing of yours is picked while AI is off. We check the setting again right before each item is sent.

How long it is kept

Anthropic keeps each request and answer for up to 30 days, then deletes it. If its safety systems flag a request, Anthropic may keep it for up to two years. We do not have a zero-retention arrangement with Anthropic. Our understanding is that content sent through Anthropic's API is not used to train its models by default, but we cannot guarantee that on Anthropic's behalf.

On our side, guides are kept about 30 days and saved explanations 90 days.

Turning AI off stops everything above, the weekly quality sample included. The privacy policy has the full detail.