Search for ChatGPT prompts for meeting notes and nearly every result is published by a recording or note-taking tool. The prompts are fine for an internal meeting. On a sales call they stop at the wrong place: a summary. A summary tells you what happened. After a sales call you need to know what changed in the deal, what you still do not know, and what to send in the next hour.
This page treats the transcript as deal input. Every prompt below expects you to paste a transcript or rough notes.
Why generic meeting-notes prompts fail on a sales call
A meeting-notes prompt is built to compress. It drops the buyer's exact wording, rounds hesitation into agreement, and turns "we would need to check with finance" into "next steps: pricing discussion." Those are the details a deal turns on. The fix is to ask for extraction instead of compression, and to make the model say unknown when something was not said.
The one prompt that does 80 percent of the work
Call to deal update: "Here is a sales call transcript: [paste]. Return five sections. One: the problem as the buyer described it, quoted. Two: what the buyer committed to, with any date mentioned. Three: what I committed to. Four: what is still unknown that the deal needs, such as budget, decision process or other stakeholders. Five: the single most important next step, with an owner. If a section has nothing, write none stated. Do not infer."
If you only save one prompt from this page, save that one. The rest are variations for specific jobs.
Prompts that extract what the buyer actually said
Verbatim pull: "List every sentence where the buyer describes a problem, a goal or a constraint. Quote them exactly. Do not include anything I said."
Hedge detector: "Find every place the buyer softened, deferred or qualified an answer, such as probably, we will see, or I would need to check. For each, tell me what they may be avoiding."
The hedge detector catches the difference between a yes and a polite not yet, which is the most common thing a tired rep misses in their own notes.
Prompts that produce a MEDDPICC or MEDDIC gap list
Gap list: "Map this transcript to MEDDPICC. For each element, quote the line that supports it. If there is no supporting line, mark it as a gap. Then rank the gaps by how much they put the deal at risk, and give me one question to close each."
Champion test: "Based only on what was said, is [name] a champion, a coach or a contact? Quote the evidence. What would they need to do on the next call for me to be sure?"
The rule that makes these work is quote or mark as gap. Without it, the model fills blanks with plausible guesses and your forecast inherits them. More structure lives in deal strategy and stakeholder management.
Prompts that find what you failed to ask
Missed questions: "I am selling [product] to [role]. Read this discovery call and list the five most important questions I did not ask, in the order I should ask them next time. For each, say what the answer would change."
Talk ratio check: "Estimate how much of this call I spent talking versus the buyer. Find the three longest stretches where I talked, and suggest a question that could have replaced each."
The discovery prompts that prevent these gaps in the first place are in ChatGPT prompts for discovery calls and meeting prep and discovery.
Prompts that write the recap email the buyer will forward
Forwardable recap: "Write a recap email of this call for a reader who was not on it, because the buyer will forward it to their boss. Under 150 words. State the problem in the buyer's own words, what we agreed, and the next step with a date. No product pitch and no adjectives about the meeting."
Internal-sell version: "Now write the two sentences the buyer could paste into a message to their CFO to explain why this is worth thirty more minutes."
A recap written for the absent reader is the one most likely to travel inside the account. See also ChatGPT prompts for sales follow-up emails.
Prompts that produce the CRM note your manager can read
CRM note: "Turn this call into a CRM note under 80 words using these headings: stage change, new information, risks, next step and date. No narrative. If the stage did not change, say so."
Forecast honesty check: "Based on this call alone, argue that this deal will not close this quarter. Then tell me which of your arguments I can disprove with evidence from the transcript."
Managers read far more notes than they write. A fixed format is worth more than a well-written one. Related prompts: sales operations, CRM and productivity.
Prompts for a multi-stakeholder call
Who said what: "This call had several people from the buyer's side. For each person, list what they cared about, what they objected to, and whether they seemed to have decision power. Quote the evidence. Flag anyone who said nothing."
Committee map: "From this call, draw the buying committee as a list: role, stated priority, stance toward us, and who they seemed to defer to. Mark every inference as an inference."
The person who said nothing on a group call is often the one who decides. More on this in account research and buyer intelligence.
Doing this on Claude or Gemini instead
Every prompt on this page is plain text and runs unchanged on Claude or Gemini. Claude holds very long transcripts and several calls at once better, which helps when you want patterns across a whole deal; see Claude prompts for sales. Gemini is convenient if your recordings already live in Google Meet and Drive; see Gemini prompts for sales teams.
What not to paste in, and the consent question
Check your company's AI policy before pasting a customer transcript into any consumer chat tool, and use the business or enterprise tier if one is approved. Strip personal data you do not need, such as phone numbers and anything about health or family. Make sure the call was recorded with the consent your region and your buyer's region require. The prompts work just as well on your own rough notes as on a verbatim transcript.
Where to keep these prompts
Call-note prompts get reused after every call, so they are the worst ones to rebuild from memory. Promptifi keeps them organized by sales stage and task, including post-call synthesis and recap emails, and every prompt runs on ChatGPT, Claude or Gemini. You can see the whole structure on the full taxonomy page or open the library.
Try one before you decide anything: Browse the library →
Frequently asked questions
What is the best ChatGPT prompt for meeting notes?
For a sales call, the best prompt asks for decisions and gaps rather than a summary: what the buyer committed to, what they said in their own words about the problem, what is still unknown, and the single next step with an owner and a date. A general summary prompt misses all four.
How do I turn a sales call transcript into next steps?
Paste the transcript and ask for next steps in three columns: what the buyer agreed to, what you agreed to, and what nobody agreed to but the deal needs. Require an owner and a date for each, and tell the model to write unknown rather than guess.
Can ChatGPT fill out MEDDPICC from a call?
It can fill in what was said and, more usefully, list what was not. Ask it to quote the line behind each element and to mark any element with no supporting quote as a gap. Never let it infer an economic buyer or decision criteria that nobody stated.
How do I write a recap email from a call transcript?
Ask for a recap written for someone who was not on the call, because the buyer will forward it. It should state the problem in the buyer's words, what was agreed and the next step, in under 150 words, with no product pitch.
Is Claude better than ChatGPT for long call transcripts?
For very long transcripts or several calls at once, Claude's larger context window helps it keep details straight. For a single call under an hour, ChatGPT, Claude and Gemini all handle the prompts on this page well. The prompt matters more than the model.
Pro is $12 per month, $89 per year, or $149 for lifetime access. See pricing