ChatGPT Prompts for Meeting Summaries That Work

The short version: Good ChatGPT prompts for meeting summaries do three things badly-written ones skip: they force a structure (decisions, actions, owners, deadlines), they name who the summary is for, and they ask the model to flag what wasn't decided. Paste a transcript in without that structure and you get a wall of polite nothing. Below are the prompts I'd use, plus a worked example so you can see the difference.

Why most meeting summary prompts produce mush

Type "summarise this meeting" into ChatGPT and you'll get three paragraphs of calm, even-handed prose that reads like a press release. Nobody argued, nobody was wrong, nothing is urgent. That's not a summary, it's a diplomatic smoothing-over of a conversation where someone probably got annoyed and a deadline probably got missed.

The fix isn't a cleverer single prompt. It's breaking the job into parts: extraction, structure, and tone. Most people skip straight to "make this shorter" and wonder why the output is useless three weeks later when nobody remembers who agreed to send the contract.

The core prompt structure that works

Every decent meeting summary prompt needs four ingredients. Leave one out and the summary degrades.

• Audience - who reads this and what do they need from it (the CEO wants outcomes, the project manager wants actions)
• Format - headings, bullets, a table, whichever matches how your team reads things
• What counts as a decision versus a discussion point - this is the bit everyone forgets
• Explicit instruction to flag gaps - what was raised but never resolved

Prompt 1: the standard team meeting summary

"Here is a transcript of a 45-minute team meeting. Summarise it for someone who wasn't there, using these sections: Key decisions made, Action items with owner and deadline, Open questions still unresolved, One-line overall summary at the top. If no deadline was given for an action, write 'no deadline set' rather than guessing one. Keep it under 300 words excluding the action list."

That last sentence matters more than it looks. Without a word limit, ChatGPT will happily produce 900 words of throat-clearing.

Prompt 2: the client or stakeholder update

"Summarise this call transcript as an update email to a client who was not present. Tone: professional, warm, no jargon. Structure: what we agreed, what happens next and by when, anything we need from them. Do not include internal disagreements or anything that would embarrass either party if forwarded."

This is the one to use if you're consulting or client-facing, which overlaps heavily with the kind of work covered in ChatGPT prompts for consultants. If you're in a regulated profession where phrasing matters legally, the approach in ChatGPT prompts for lawyers is worth reading alongside this, because "don't guess a deadline" and "don't invent a commitment" are the same rule wearing different hats.

Prompt 3: the board or leadership summary

"Summarise this transcript for a board that needs to make a decision next week. Separate: facts presented, risks raised, decisions requested from the board, and anything that was glossed over too quickly given its importance. Be blunt about the glossed-over part even if it's awkward."

That last instruction is the uncomfortable bit. Most people writing meeting summaries soften exactly the thing that most needs flagging, because nobody wants to be the one who writes "this was rushed" in an email the whole leadership team can see. ChatGPT doesn't have that social cost, so use it. Ask it directly to flag what was rushed, dismissed, or talked over, and you'll often get a more honest read of the meeting than the person chairing it would ever write themselves.

Prompt 4: the recurring weekly meeting

"Summarise this transcript in the same format as previous weeks: last week's actions (done/not done), this week's decisions, this week's new actions. If an action from last week wasn't mentioned this week, flag it as 'not followed up' rather than assuming it's done."

This one needs you to paste last week's summary in as context, which is more typing but it's the only way to catch the actions that quietly die from neglect.

A worked example

Say you run a 12-person marketing agency and you've just had a 40-minute call about a client's rebrand timeline. The transcript is messy: two people talk over each other about the logo, someone mentions budget briefly and it's never resolved, and the account lead promises a revised timeline "by Friday" without saying which Friday.

Fed through Prompt 1 above, a usable output looks like this:

• One-line summary: Rebrand timeline pushed back two weeks; logo direction agreed; budget still unresolved.
• Key decisions: New logo concept B approved. Launch date moved from 14 March to 28 March.
• Action items: Sarah to send revised timeline, no deadline set (said "by Friday" without specifying which). Tom to confirm printer costs by next Tuesday.
• Open questions: Budget overspend of roughly 8 percent mentioned but not addressed. Who approves the final sign-off was not clarified.

That "no deadline set" line is the whole point. A lazier prompt, or a tired human note-taker, would have just written "Sarah to send timeline by Friday" and quietly invented a date nobody agreed. That's how projects slip.

Where ChatGPT prompts for meeting summaries fall down

Here's the uncomfortable truth nobody selling "AI meeting notes" wants to say loudly: ChatGPT cannot tell the difference between a firm commitment and someone being polite. If your sales lead says "yeah, I'll probably look at that next week" to get the client off the phone, ChatGPT will often list it as an action item with a deadline, because linguistically it looks like one. You have to read the summary and sanity-check anything that sounds like a commitment against what you remember people meaning.

The same goes for tone. If two people were visibly annoyed with each other on a call, a transcript-only summary won't know that unless someone's words were sharp enough to show it in text. Video and audio carry information text transcripts throw away, and no prompt fixes that.

Formatting prompts for different outputs

Once you've got the content right, the output format matters more than people think, because a summary nobody reads is worthless regardless of accuracy.

• For Slack or Teams: "Format as a Slack message with bold headers and no more than two bullets per section."
• For email: "Format as a plain email, no markdown symbols, suitable for pasting directly into Outlook."
• For a project management tool: "Format action items as a table with columns Task, Owner, Deadline, Status."
• For a quick voice note recap: "Write this as three spoken sentences I could say out loud in a standup."

If you're building out a wider note-taking system rather than doing this one meeting at a time, it's worth looking at ChatGPT prompts for note taking and the broader ChatGPT prompts for summarizing reports and emails, because meeting summaries are really just one branch of a document-processing habit worth building.

A quick process that saves time

Step by step, for a one-hour meeting:

• Record and auto-transcribe the call (most video tools do this now, including free tiers)
• Paste the raw transcript into ChatGPT with Prompt 1 or 3 depending on audience
• Read the output and correct any invented deadlines or misread commitments, this takes about 3 minutes
• Send it within the hour, while people still remember the context

That whole process takes under 10 minutes for a 45-minute meeting, compared with the 20 to 30 minutes it usually takes someone to write decent notes from memory two days later, by which point half the detail is gone anyway.

If you're using this across a team

If you're rolling meeting summary prompts out across a whole business rather than just for yourself, the hard part isn't the prompts, it's getting everyone to read the output and flag errors instead of forwarding it unread. That's a training and habit problem, not a prompt problem, and it's one reason some businesses bring in outside help to set the system up rather than letting ten people each invent their own version. If that's where you are, it's worth reading about what an AI implementation coach does before assuming you need one.

Frequently asked questions

What is the best ChatGPT prompt for summarising a meeting transcript?

Ask for a one-line summary, key decisions, action items with named owners and deadlines, and a separate list of open questions, and tell ChatGPT to write "no deadline set" rather than guessing one when a date wasn't agreed.

Can ChatGPT summarise a meeting from audio directly?

Not reliably on its own. You'll generally need a transcription tool or a meeting platform's built-in transcript first, then paste that text into ChatGPT for the summary itself.

How do I stop ChatGPT inventing deadlines or commitments that weren't agreed?

Tell it explicitly in the prompt not to assume a deadline if one wasn't stated, and always read the output against your own memory of the call before sending it anywhere, because vague language gets turned into firm-sounding action items more often than you'd expect.

Is it safe to paste a client or internal meeting transcript into ChatGPT?

Check your organisation's data policy and the chat tool's settings first, since some meetings contain confidential or personal information that shouldn't go into any AI tool without the right settings or a business-tier account that restricts data use.