Write a brief for the transcript summary and describe the structure of the protocol. Return JSON {"summary_brief":"...","structure":"...","open_questions":"..."}: model instructions on what to extract, protocol structure (decisions, tasks with responsible person and deadline), and how to highlight unclear/dangling places.
The meeting summary is not a retelling, but an extract. You need to pull three different things out of the transcript: decisions (what was finally decided), tasks (who, what, by when) and open questions (what hung). Mechanics: the model reads the entire text and groups the meaning into these categories - but only if you name them. The general request “make a summary” gives a mess; the request “separate decisions from discussions, for each task indicate the person responsible and the deadline” gives the protocol. Pro trick: clearly demand a responsible person and a deadline for each task, and if they are not in the entry, let the model put a dash or a question, and not make it up. This turns the summary into a working document that can be questioned. Second trick: ask at the end for a list of “incomprehensible places” - phrases where the meaning is double; this way you quickly find what to clarify with the participants. Insider: a short half-page protocol is read, a long retelling is not; ask for a volume limit. A common mistake is to take a smooth summary for granted; always check solutions and numbers with the original recording.
Unlock access to submit solutions for instant AI review.