ChatGPT Meeting Minutes Generator: A Client-Call Recap for Event Planners

Type your call notes into a fixed prompt that sorts each line as decided, proposed or open, then asks for a decisions list, an actions table with owner and date, and a draft confirmation email. Check every owner and date against your notes, strip confidential details first, and treat the recap as agreed only when the client replies.

Event-planner minutes differ from board minutes because their job is to stop a dispute three weeks later ("I thought the band was included"). So the useful parts are the line between decided and merely discussed, the deadlines that run backwards from the event date, and every figure that moves the budget. ChatGPT's habits work against all three: it turns a "maybe" into a decision, guesses who owns an action, and invents tidy deadlines. The routine below is built around those habits.

Follow me on Instagram@sagnikteaches

It is a notes-in, minutes-out routine, not a review of recording tools. If you record or transcribe calls, read whether AI meeting note-takers are safe for client calls first. Standalone "meeting minutes generator" sites are separate services with their own terms for whatever you paste into them, so I would keep client notes in a prompt you control.

Connect on LinkedInSagnik Bhattacharya

Stage 1: Mark your notes while you type them

The trick that makes the later stages work happens during the call. Use three marks at the start of each line: a star for anything the client definitely agreed, a dash for something said but not agreed, and a question mark for anything left hanging. A few minutes after the call, add the date, the event date and who attended. That is the whole preparation, and it hands ChatGPT your own judgement of what counted as a decision, which it cannot work out from the words alone. If your notes app turns stars and dashes into bullets, use the letters A, S and Q instead.

Subscribe on YouTube@sagnikteaches

Here are made-up notes from a 40-minute call about a corporate awards dinner for about 90 guests. Notice how messy they are, because real ones are.

Call: Tue 4 Mar, 40 min. Client events lead + me. Their PA joined for last 10 min.
Event: Fri 28 Mar, awards dinner, about 90 guests
* Doors 6.30, dinner 7.30, awards around 9, finish 11
* Mains: option B, plus vegetarian and vegan options
- Dessert not chosen yet, she'll look at sample list
* Guests: 86 RSVPs so far, "90 ish" final. Caterer needs final numbers 10 days before (Tue 18 Mar). Client sends final by Fri 14 Mar
* Dietary counts so far: 7 veg, 3 vegan, 2 GF, 1 nut allergy (counts only, no names)
* Screen + mic needed. I get AV quote. Client sends logo + slide deck by Fri 21 Mar
? Band. "maybe". She'll think. About 1,800 if yes
- Speeches: shorter than last year, 3 min max each? not agreed
* Florals: low centrepieces only, no tall (sightlines)
* Budget ceiling 22,000. She said "fine to go a bit over if we add the band"
- Current total 20,515 = venue 6,500 + catering 90 x 95 = 8,550 + AV 2,400 + florals 1,200 + 10% contingency 1,865
? Seating plan: PA or me? unclear
* Client signs off menu by Fri 7 Mar
? Deposit due 14 days before? check contract
- Arrivals might be outdoors, need rain plan, I ask venue about canopy

Count what is in there: eight starred lines, four dashes, three questions. A few of them are traps on purpose. "Fine to go a bit over" is starred but contains no number. The seating plan has no owner. The deposit date is a guess. Those are the places where a drafting tool is most likely to tidy things up into something the client never said.

Stage 2: Swap out what must not leave your hands

Before the notes go anywhere, strip the details that identify people or carry risk. This takes about three minutes once you have a habit, and it is the step that lets you keep using the tool when a client asks awkward questions about it.

The guest-level rows matter most. A guest's allergy is health information about a named person, and the counts give the caterer and the prompt everything they need. The tutorial on redacting personal data before sharing has a fuller method if your notes are long.

Then the settings. On a personal ChatGPT plan, switch off "Improve the model for everyone" under Data controls; business plans such as ChatGPT Business do not train on business content by default. Temporary Chat stays out of your history, does not create or update memories and is not used to improve models while it stays temporary, though OpenAI says it may keep a copy for up to 30 days for safety. Finally, read your client agreement or any non-disclosure agreement for a clause on third-party tools. If it is silent and the client is large, a one-line email does the job: "I use an AI drafting tool for call recaps, with no names or contact details pasted. Any objection?" Keep the answer on file.

Stage 3: The prompt that turns notes into event-planner minutes

Save this as a ChatGPT Project's instructions or as a text file you paste each time. Square brackets mark what changes per call.

You are helping an event planner write minutes of a client call from the planner's own rough notes. Write in plain English, using the spelling in my notes.

Context
- Call date: [Tuesday 4 March]
- Event date: [Friday 28 March]
- Event: [awards dinner, about 90 guests]
- Attendees by role: [client events lead, planner, client PA for the last 10 minutes]

How to read the notes
- A line starting with * was agreed by the client: treat it as DECIDED.
- A line starting with - was said but not agreed: treat it as PROPOSED or an IDEA that needs the client's reply.
- A line starting with ? is unresolved: treat it as OPEN.
- Never upgrade a status. An unmarked line is PROPOSED.

Rules
- Do not invent any number, date, name or owner. Where the notes do not say, write [OWNER?] or [DATE?].
- Turn relative dates into a weekday and date using the call date, written like "Fri 7 Mar".
- Keep approximate figures approximate ("about 90"). Never round up to a firm number.
- For every action, quote the note line it came from.
- Use only what is in the notes.

Output, in this order
1. Snapshot: event, date, guest numbers and their status.
2. DECIDED: numbered D1, D2...
3. PROPOSED (needs the client's reply): P1, P2...
4. OPEN questions: who must answer, and by when.
5. Money: every figure in the notes, marked agreed or estimate. Show any sum, and say whether the total is above the stated ceiling.
6. Actions: a table with ID, action, owner, due date, source line, and "stated" or "inferred".
7. Risks and plan B.
8. Lines not used: any note line you left out, and why.
9. A recap email under 180 words: what was agreed, what I need from the client, what I will do, and a request to reply "Confirmed" or send changes by [reply-by date].

Notes:
[paste the redacted notes here]

Four lines carry the weight. "Never upgrade a status" stops the maybe-to-decision drift. "Where the notes do not say, write [OWNER?] or [DATE?]" gives the model a legitimate way to admit a gap, which it will otherwise fill. Quoting the source line makes every action checkable in seconds. And "Lines not used" tells you what got dropped, the failure you cannot see by reading the output alone.

Stage 4: What the first draft returned, and where it slipped

An illustrative first draft, trimmed to the parts that matter. The decisions section came back like this.

• D1 Doors 6:30 pm, dinner 7:30 pm, awards 9:00 pm, finish 11:00 pm.
• D2 Mains: option B, with vegetarian and vegan options.
• D3 90 guests; 86 RSVPs so far; final number from the client by Fri 14 Mar.
• D4 Low centrepieces only, to protect sightlines.
• D5 Screen and microphone required.
• D6 Live band, about $1,800.
• D7 Budget ceiling $22,000; the client approved going over to add the band.
• D8 Menu sign-off by Fri 7 Mar.

Proposed: choose a dessert, and ask the venue about a canopy for outdoor arrivals. Open: who drafts the seating plan, and when the deposit is due. The money section lists venue $6,500, catering 90 x $95 = $8,550, AV $2,400, florals $1,200 and a 10% contingency of $1,865, a total of $20,515, and then says that with the band the total is $22,315 and "within the client's approval". "Lines not used" lists one item, the speeches line, because it was "not agreed". The actions table looked like this.

It reads well, and that is the danger. A client would nod along with almost every line. The "Basis" column is what gives it away: three actions are marked inferred, and each of those is something the notes left open.

Stage 5: Check every line against your notes

Take the notes in one window and the draft in another and work through it in this order. Tick each note line against the output: every line should land in DECIDED, PROPOSED, OPEN, Money, Actions or "Lines not used", and a line that appears nowhere has been dropped. Then check the status of each decision against your marks, the owner and date of each action, and every number. For the awards dinner the check finds nine problems.

Two quick sums finish the check. The arithmetic in the money section was right: $6,500 + $8,550 + $2,400 + $1,200 = $18,650, and 10% of that is $1,865, giving $20,515; adding the $1,800 band makes $22,315, which is $315 over the $22,000 ceiling. But the draft concluded "within the client's approval" because of D7, which was wrong. A correct sum can carry a false conclusion, so check the claim as well as the number. Each guest above 90 also adds $95 to catering before contingency, so six more would add $570, which is worth saying to the client now.

Then the dates. Counting back from Fri 28 Mar, ten days is Tue 18 Mar, the caterer's cut-off, and fourteen days is Fri 14 Mar. The client has promised final numbers for Fri 14 Mar, so there are four days of cushion before the caterer needs them, and a deposit due "14 days before" would land on the same Friday if the contract says so. Neither fact was in the draft, and both belong in the email, because a four-day cushion is what the client will want to hear when RSVPs trickle in.

Stage 6: Send the recap and make the client confirm it

The corrected recap goes out the same day, from your own email, in the thread you already have with the client. Put the swapped-back names and venue in, keep every "to confirm" visible, and end with a reply-by date. Here is the awards-dinner email after the fixes.

Subject: Recap of Tuesday's call: awards dinner on 28 March

Hi [client first name],

Thanks for the time on Tuesday. Below is what we agreed and what I still need from you. Please reply "Confirmed" or send changes by Friday 7 March.

Agreed
- Doors 6.30 pm, dinner 7.30 pm, awards around 9 pm, finish 11 pm.
- Mains: option B, with vegetarian and vegan options.
- Dietary counts so far: 7 vegetarian, 3 vegan, 2 gluten-free, 1 nut allergy.
- Low centrepieces only, to protect sightlines. Screen and microphone needed.
- Guests: about 90 (86 replies so far). Your final number by Friday 14 March; the caterer needs it by Tuesday 18 March.

Still to decide
- The live band (about $1,800). With it the total is $22,315 against your $22,000 ceiling, so please tell me the most you are happy to approve.
- Menu sign-off, including dessert, by Friday 7 March.
- Speeches: are you happy with a 3-minute limit each?
- Who drafts the seating plan, you or me?

What I will do
- Get the AV quote and send it by [date I propose].
- Check the deposit date in the contract and confirm it.
- Ask the venue about a canopy for outdoor arrivals.

What I need from you: your logo and slide deck by Friday 21 March.

[your name]

Two habits make the email work. First, decide in advance what counts as confirmation: a reply saying "Confirmed", or one listing changes, in the same thread. A read receipt does not count. Second, chase on the reply-by date. If nothing has come back by Friday 7 March, phone and make a note of the call, because silence is not agreement. If the client changes something, issue the minutes again as version 2 and keep both versions in the event file.

In the illustrative first draft, the email also needed trimming. It ended with "we will make sure everything runs perfectly on the night", a promise nobody made on the call. Cut anything in the email that reads as a commitment you did not discuss.

Seven ways these recaps go wrong, and the tell for each

• A maybe becomes a decision. The tell is a DECIDED list that is longer than your starred lines, or an item in it you remember marking with a question mark.
• An owner is guessed. The tell is "inferred" in the Basis column. If the notes do not name an owner, the output should say [OWNER?].
• A date is invented or resolved wrongly. "By Friday" turns into the wrong Friday, or a gap becomes a tidy deadline. Always give the call date, then count back from the event date as in Stage 5.
• An approximate figure becomes firm. "90 ish" turns into "90". Compare every number with the notes, and keep "about" where you wrote it.
• A line vanishes. Dietary counts disappeared in the example. The tick-off of note lines against output sections is the only reliable catch, and the "Lines not used" section helps only when the model is honest about what it left out.
• Two items merge. An AV quote and a lighting quote become "get quotes". Keep one action per supplier so each has its own date.
• The email promises more than you said. Read it as the client would: what could they later point to and say you committed to?

The tutorial on where ChatGPT gets event-planning facts wrong covers the wider habit of confident invention, and it applies to minutes as much as to venue research.

When two people on the client side give different answers

Most event calls have more than one voice on the client side, and the minutes have to survive that. Suppose in the awards-dinner call the events lead said speeches should stay under three minutes, and the PA, joining for the last ten minutes, said five would be fine. Nobody settled it. If your notes just say "speeches 3 or 5 min", the model may quietly pick one, and you will not see it happen.

Add two things. In your notes, tag who said what with a role in brackets: "- speeches max 3 min (lead)" and "- speeches 5 min ok (PA)". In the prompt's rules block, add this line: "If attendees disagree, list both positions under OPEN with their roles and do not choose between them." The prompt's context block can also carry an "Approver" line, for example "Approver for spend and scope: client events lead", so the output flags any budget change that someone other than the approver agreed to.

The email then asks a real question instead of announcing a decision nobody made: "Your events lead mentioned a 3-minute limit for speeches and your PA mentioned 5. Which should we plan the run of show around?" That sentence costs you nothing to send and avoids planning the evening around the wrong answer. It also protects the PA, who might otherwise find a decision attributed to them in writing that they never meant as final.

Time per call, and what the routine costs

Illustrative figures, to be replaced with your own after a few calls. Marking notes during the call costs nothing extra, and the rest comes to roughly 23 minutes: about 3 minutes after the call to add the date and attendees, 3 to swap out confidential details, 2 to run the prompt, 10 to check the draft against your notes and 5 to fix the email and send it. Say a planner used to spend 45 minutes turning a notebook page into a recap. Halving that across six client calls a month saves a little over two hours, and the bigger gain is that the recap goes out the same day.

Cost is small. The routine works on ChatGPT's free plan for light use, though usage limits there change, so check them. A personal paid plan such as Plus ($20 a month, list price in USD) gives more headroom, and ChatGPT Business, which needs at least two seats at $25 a seat monthly or $20 billed annually, does not train on business content by default. For one planner the individual plan with the training switch off is usually enough.

The check in Stage 5 is the expensive part, and it shrinks when calls are better prepared. If you spend the first ten minutes of calls collecting basics such as date, guest count and venue, move them into an intake form so the call can focus on decisions. And once the actions table is right, it needs to leave the minutes and enter your task list; turning meeting notes into tasks automatically shows how.

If you record calls instead of typing notes

Some planners prefer to record and transcribe. That is a different consent and storage problem. Ask every participant for permission before recording, because rules on recording differ by place and you should check yours. Google's own help page says Meet informs all participants when its "Take notes for me" feature is running, and that the feature needs an eligible Google Workspace edition or Google AI plan. Teams and Zoom have their own recap features on eligible paid plans. Whatever produces the transcript, the Stage 2 swap table still applies before it goes into ChatGPT, and your marking system still applies afterwards: you will need to go through the transcript and mark what the client agreed, because a transcript records every word with equal weight.

After the event: turn the misses into rules

Keep a short list of what Stage 5 caught on each call. After five or six events, the repeats become rules. In the example, the missing dietary counts suggest adding "Any starred line giving guest numbers or dietary counts must appear under DECIDED". The wrongly assumed owners suggest "If an owner is not named, write [OWNER?] and add it to OPEN". The invented dates suggest "Never write a due date unless the notes contain one". Each rule you add shortens the next check.

The other use of minutes is the debrief. When the event is over, read the confirmed recap against what happened: which decisions held, which changed after confirmation, and where the client's memory differed from yours. The tutorial on AI for event budgets and run sheets picks up the money side, and the notes you confirmed become the evidence for what was in scope when you quote the next job.

Questions about ChatGPT minutes for client calls

Is it safe to paste client call notes into ChatGPT?

It depends on the plan and on what you paste. On a personal plan, switch off 'Improve the model for everyone' in Data controls; business plans do not train on business content by default. Keep guest names, contact details, payment details, per-person dietary information and anything the client marked confidential out, and check your client agreement for a clause on third-party tools.

Can ChatGPT listen to the call and write the minutes?

This routine starts from notes you type yourself, which keeps you in control of what leaves your hands. If you record or transcribe calls with a separate note-taker, get every participant's agreement first, because rules on recording differ by place, and treat the transcript as confidential client data too.

How soon should the recap go out after the call?

The same day, while both sides remember the call. Ask for a reply within two working days and say what happens if you hear nothing, such as a quick phone call. Silence is not agreement, and a recap nobody confirmed will not protect you in a later dispute.

What if the client disagrees with the minutes?

Change them, issue the corrected version as a second version and keep both in the event file. A disagreement caught in week one is the recap doing its job, and much cheaper than finding it at the final-numbers deadline. Do not argue about wording; agree what was decided.

Further reads

• Which Event Planning Tasks Should You Hand to AI First? — Which other planning tasks are worth handing to AI first.
• Event Planning Checklist: Build One Per Event Type With AI — Build a checklist per event type to feed the actions list.
• Event Planning Using AI After the Day: Feedback, Sponsor Reports, Debrief — Use the same discipline after the day: feedback, sponsor reports, debrief.
• How to Use Take Notes for Me in Google Meet for Client Calls — How Meet's own note feature behaves on client calls.
• Copilot in Teams Meetings: Recaps, Actions and Privacy Settings — Recaps and privacy settings if your calls run on Teams.
• AI Software for Event Planning: What HoneyBook AI Covers — A job-by-job map of what HoneyBook AI handles for event planners, what ChatGPT does better and what stays with a person.
• AI for Meeting Planners: Venue Shortlists and RFP Replies — From client brief to venue shortlist to proposal reply: prompts, a scoring grid and a figure-check table for independent planners and small agencies.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: OpenAI help pages on Temporary Chat and Projects; Google Meet help page on Take notes for me; ChatGPT plan details from OpenAI's help pages (October 2026). All example figures and dates are invented for illustration.