How the summary is written

When someone presses the button on a notebook’s summary page, the entries in that notebook are sent to Claude, an AI model made by Anthropic, with one set of instructions. Claude receives the text of the entries as people wrote them, and the current question if there is one. It does not receive names, even when a contributor added one. It does not receive the name of your school or your district, or anything else about them. Nothing else about anyone is kept, so nothing else can be sent.

Claude is asked to read the entries twice. The first reading is for what people tried, what helped, what didn’t, and what they would do differently. That becomes the overview and the takeaways.

The second reading is for how people and the AI worked together: what was handed to the AI and what was kept; whether what came back was checked, and how; whose name went on work that reached students or families; what people expected; what went into the tool; and where saved time went. That reading produces the questions at the end, when the entries raise any.

Claude is told not to grade anyone, not to describe or quote any person, and not to say that AI improved or harmed anything. When only a few entries speak to a matter, it is told to say nothing rather than guess.

The summary is a starting point for a conversation in your building, not a finding. It stays until the entries change. Then it is retired, and a new one can be written.

The people who develop AI Field Notes read entries the same way, without names, for research, as the privacy page explains.

The way of reading draws on Randy Sparkman’s work on AI literacy and on the human role in AI-assisted work. rsparkman.net