AI Prompts to Summarize Meetings: How to Write and Use Them
Typing “summarize this meeting” into ChatGPT and hitting enter almost never solves the problem. What comes out is a generic paragraph with no decisions, no deadlines, and no idea who owns which task. The good news is that this is almost always the prompt’s fault — not the AI’s.
The way you write the prompt decides whether the meeting becomes a useless summary or a document with the decisions and next steps you will actually go back to.
What a good summarization prompt needs
Techniques validated by OpenAI and Anthropic themselves come down to five elements.
Role. Tell the AI who it is. “You are an assistant for a product team” changes the tone and what it considers important.
Context. Explain why that meeting matters: what the goal was, who is taking part, who the summary is for. The model generalizes from the explanation.
Output format. Be explicit: “answer in 5 topics”, “build a decisions table”, “use bullets”. Without this, the result changes format every time.
Example. Giving a ready-made example is the most reliable way to control tone and structure. Three to five well-crafted examples work even better.
Grounding in the source. Ask the AI to quote the passages from the transcript before summarizing. This greatly reduces hallucinations — especially the worst one: inventing a decision or task that never existed.
Ready-to-copy prompt examples
If you are using a general-purpose AI with the transcript pasted in, start here.
Structured summary:
You are my meeting assistant. Below is the transcript of a marketing planning meeting. Summarize it in 5 titled topics, quote the points whenever you can, and end with a “Decisions” section of 3 bullets. I transcribed the meeting to text in Sintesy, so do not edit proper nouns.
Extract decisions and tasks:
List all the decisions made in this meeting and build a table with Task, Owner and Deadline columns. Do not invent an owner or deadline that is not explicit in the transcript. If something is not clear, write “not defined”.
Meeting minutes:
Write formal minutes with: attendees, agenda, discussion (2 paragraphs), decisions, and next steps. Use objective language, no adjectives. Format: bold headings and lists.
Mind map (textual):
Turn the transcript into a text-based mind map with a central node, 3 main branches, and sub-items with one sentence each.
Mistakes almost everyone makes
Vague prompt. “Summarize this” produces a generic summary with no hierarchy and no next steps.
No context. Without knowing the type of meeting and the goal, the AI does not know what to prioritize.
Blowing past the context window. Long transcripts (20k tokens or more) exceed the limit, and the model loses key points or starts hallucinating.
Asking for only the summary. The operational value is in the tasks. If you do not explicitly ask for “who does what, by when”, they get left out.
Not grounding in the source. Without quotes, the model “fills in the gaps” and can invent decisions.
Giving negative instructions. Prefer “answer in bullets” to “do not use markdown”. The AI responds better to what to do than to what not to do.
The ceiling of the manual workflow
The “transcribe with a tool + copy and paste into ChatGPT + adjust the prompt” workflow works, but it has a ceiling. Every meeting repeats the ritual. You pay for two tools, waste time fine-tuning the prompt, and the result stays scattered: summaries in ChatGPT, audio in the other tool, tasks spread across a document.
Without a central place to store and search those summaries, the information gets lost again.
When you do not need to write a prompt at all
This is where AI transcription changes the game. Sintesy turns the meeting audio into text and already delivers the summary, the mind map, the action checklist, and the tasks automatically — grounded in the source, in Portuguese, in a few seconds. Not because it has a “hidden” prompt, but because the extraction prompt is already structured and calibrated for you.
In other words: writing good prompts pays off when you work with a general-purpose AI and a raw transcript. When the tool already does that work for you, all that is left is reviewing. Upload a meeting audio to Sintesy and compare it with the summary you could write by hand — the time difference is huge.
Frequently asked questions
What is the best prompt to summarize a meeting? There is no single one. What works is combining role, context, explicit format, and grounding in the source. The examples above cover summary, decisions, minutes, and mind map.
Do I need to transcribe before summarizing? If the audio does not have text yet, yes. You transcribe and then apply the prompt. Tools like Sintesy do both at once.
How do I stop the AI from inventing tasks? Ask it to quote the passages from the transcript and write “do not invent an owner or deadline that is not explicit”. With a faithful transcript in hand, the risk drops a lot.
