After a meeting you transcribe the recording, toss it to the AI and say "summarize," and it gives you a meandering log nobody reads. Good meeting minutes don't rely on "summarizing"-they rely on "structure": filter out the chatter and extract decisions, action items, owners, and deadlines. This prompt pack goes three levels deep.
Beginner: Structured Minutes
First, get the structure.
You are a meeting-notes assistant. Based on the following meeting transcript, output structured minutes:
1. Meeting topic (one sentence)
2. Key discussion points (3-5, one sentence each)
3. Decisions (what was explicitly agreed)
4. Action items (who, does what, by when-if mentioned)
Filter out pleasantries, repetition, and off-topic tangents. Stay faithful-don't fabricate unmentioned content.
Transcript: {{transcript}}Intermediate: Action-Item Focus
A meeting's value is its action items.
You are a project coordinator. Extract all action items from the meeting transcript into a table:
| Action item | Owner | Due date | Dependencies | Status |
Rules:
1. No explicit owner -> mark "unassigned"; no explicit date -> mark "TBD"
2. Flag items with upstream dependencies (e.g. "can only start after A is done")
3. Identify "phantom tasks" that were mentioned but have no owner-list them separately as a reminder
4. Action items must be executable ("improve the product" won't do; "deliver login page v2 design by Wednesday" will)
Transcript: {{transcript}}Expert: Follow-up Tracking + Risk Warning
Go beyond recording-drive execution.
You are a PMO. Based on these minutes + the action-item list from the last meeting, output:
1. Status of last meeting's action items (mark each: done / in progress / overdue)
2. Reasons for overdue items (if mentioned) + new due dates
3. New action items from this meeting (same table)
4. Risk warning: identify 2-3 execution risks (e.g. owner overloaded, unresolved dependency, unrealistic deadline) with a one-line recommendation
This meeting's minutes: {{minutes}}
Previous action items: {{prev_actions}}How to Use
Beginner for minutes, intermediate for action items, expert for follow-up. Everyday meetings are fine at intermediate-the action-item table is the only part of a meeting that produces output. Expert suits recurring project meetings, chaining sessions into a trackable execution chain. The key is feeding the AI the previous meeting's action items beforehand-otherwise it can only look at this one meeting in isolation and the tracking breaks.