AI Apps Worth Knowing · Module 13 of 22

The Prompter AI: Prove a Better Prompt Works on Your Own Meeting Notes

The Prompter AIBeginner~8 min practice17 steps

Why this matters

After every weekly meeting someone writes the recap. A quick “summarize these notes” gives one tidy paragraph, but the team needs a list: who does what, by when. So the list gets rebuilt by hand, every week.

On The Prompter AI forum, a new member asks whether a better prompt really makes that much difference. You can answer with evidence: the same notes, asked two ways, checked against the notes.

Practice

Do every step yourself in a practice copy of the real tool. It checks your work, and “Do it for me” finishes any step you get stuck on.

The Prompter AIThe Prompter AI: Prove a Better Prompt Works on Your Own Meeting Notes17 steps · ~8 min · nothing touches your real accounts

What you’ll build

You recap the made-up weekly meeting notes of Harbor Insurance Agency in ChatGPT twice: once with a quick one-line ask, then with a four-part prompt (Role, Context, Task, Format).

You check both answers against the notes, then write up the before and after for The Prompter AI’s weekly challenge. The write-up stays on your computer unless you choose to post it.

What you need

  • Nothing to start the practiceThe Prompter AI and ChatGPT are practice copies inside the lesson. Harbor Insurance Agency and its notes are made up.
  • The meeting notes and promptsDownload them below: the notes, both prompts, a check-before-you-send list and a before-and-after template.
  • A ChatGPT account, to do it for realchatgpt.com. Any plan works for this.
  • Your own meeting notesAny notes from a real meeting, with names changed if you are sharing the result.

Concepts

Run the same job twice

The only way to know if a prompt is better is to give two prompts the same notes. In our test the quick ask gave a readable paragraph with the right facts. The four-part ask gave a table with owners and due dates.

It is one test on one set of notes. It shows the pattern, so try it on your own.

Tell it what not to do

The Task said: only include what was actually said, and do not invent dates or owners. Where the notes named no owner, ChatGPT wrote “Not stated” instead of guessing. A visible blank is easier to fix than an invented name.

Ask for the shape you need

The Format asked for a two-sentence summary, a table with Action, Owner and Due date, and a Still open list. Asking for the exact shape means you edit a draft instead of rebuilding one.

Do it for real

Now repeat it in your own account. Tick each step as you go.

Downloads

Quick check

3 questions to check you’ve got it.

1. The quick ask gave a paragraph with the right facts. Why was it not enough?

  • It had wrong facts
  • It did not say who does what by when
  • It was too long
Show the answer

It did not say who does what by when. The facts were fine. The team needs a list of actions, owners and dates, and the quick ask never asked for one.

2. The notes never name who makes the renewal calls. What should a good recap show for the owner?

  • A name that sounds right
  • Not stated, so you can decide
  • Leave the row out
Show the answer

Not stated, so you can decide. The Task said not to invent owners. A visible Not stated is easy to fix. An invented name can slip through.

3. What do you do before you send the recap?

  • Send it, the table looks right
  • Check each row against the notes and fix what is off
  • Run it again until it is perfect
Show the answer

Check each row against the notes and fix what is off. Even the good answer needed two small fixes by hand. Checking against the notes takes minutes.

Up next · Module 14

OLVA: a private AI assistant on your client calls

You tested a prompt and checked it. Next, CoHear turns recorded sessions into notes and action items, so the notes you start from write themselves.

Start module 14 →