Meeting Summaries and Action Items: AI Prompt Workflows

A good meeting summary is short, accurate, and ends with a list of who-does-what-by-when. A bad one is a 600-word essay nobody reads. This tutorial shows you how to prompt AI to turn raw meeting notes or transcripts into the first kind, consistently, across any team.

1. Introduction

Meetings produce three useful artefacts: decisions made, action items assigned, and open questions still to resolve. Everything else is noise. Yet most meeting summaries bury those three things under paragraphs of context, side discussions, and pleasantries. The result is that nobody re-reads the summary, action items slip through the cracks, and the same topics resurface a week later.

AI is genuinely excellent at this job — provided you brief it correctly. We will look at the structure of a strong summary prompt, the most common failure modes, and a reusable template you can copy into every meeting workflow.

2. The Concept Explained

A meeting summary prompt has two halves. The first half describes the meeting and the audience: who attended, what was the purpose, who will read the summary later. The second half describes the output shape: which sections to include, what to put in each, what to leave out. Both halves matter — drop one and the AI either invents context or invents structure.

The cleanest mental model is a small pipeline. Raw transcript or notes go in, a clear extract comes out. The prompt is the filter that decides what survives.

What good summaries actually contain

A reusable summary has roughly five parts: a one-paragraph context, the key decisions, the assigned action items (with owner and deadline), the open questions, and the date of the next check-in. That is it. If your prompt asks for those five things explicitly, you will rarely need a second pass.

3. The Problem Without This Technique

Most people paste a transcript and type something like "summarise this meeting". The AI obliges with a long, novel-style recap that mentions everyone who spoke, every tangent that was raised, and no clear next steps.

summarise this meeting

4. The Solution

Act as an experienced executive assistant.

Below is the raw transcript from a weekly product sync
between the engineering lead (Aarav), the design lead
(Mei-Lin), and the product manager (Diego) at a B2B SaaS
company called Cohort Labs.

Audience for this summary: the wider product org (about
30 people) who did not attend the meeting.

Produce the summary in this exact structure:

1. **Context** — one short paragraph (max 60 words).
2. **Decisions made** — bulleted list. One line each.
3. **Action items** — table with columns: Owner | Action |
   Deadline. Only include actions explicitly agreed in
   the meeting; do not invent owners.
4. **Open questions** — bulleted list of things raised
   but not resolved.
5. **Next check-in** — date and purpose.

Rules:
- Skip pleasantries, side jokes, and tangents.
- Use neutral, factual language. No corporate filler.
- If something is ambiguous, say "unclear from transcript"
  rather than guessing.

Transcript:
"""
... paste transcript here ...
"""

5. Step-by-Step Breakdown

1. Capture the input. Use a meeting recorder, a transcription tool, or even your own typed notes. The rawer the input, the more value the AI adds.
2. Name the attendees and roles. The AI uses names to attribute action items correctly. Generic "the team said" is far less useful than "Mei-Lin agreed to…".
3. State the audience. A summary for the CEO looks different from one for the engineering team. Specify who will read it.
4. Lock the structure. Spell out the exact sections: Context, Decisions, Action items, Open questions, Next check-in.
5. Add anti-hallucination guardrails. Tell the AI to mark anything unclear, and to never invent owners or deadlines.
6. Review before sending. Always read the action items once. The AI is right most of the time, but you are the accountable human.

Tip: Save this prompt as a personal template. Next week, all you change is the transcript and the attendee list. A reusable template is worth ten clever one-off prompts.

6. Practice Exercises

7. Key Takeaways