Preparing a meeting
~ min read
30-second summary
- A poorly prepared meeting multiplies its cost by the number of people in the room. Prepping it well with AI takes a few minutes, and the return is high especially at the two right moments: before and after.
- Before: an agenda with objective, owner, time, expected decision; a brief under 150 words; a pass asking “what objections might come up?” with AI as devil’s advocate.
- During: AI isn’t at the table. Your notes, or a recording with explicit consent.
- After: ask for an actionable extract (decisions, action items with owner and deadline, open questions), not a generic summary. From there come the follow-up and the update for those who weren’t on the call.
Thirty minutes of a poorly prepared meeting, multiplied by eight people on the call, is four hours of company work burned. The cost of a meeting multiplies by the number of people in it: that’s why prepping it well gives a high return, far bigger than the few minutes it takes.
The point of this lesson is simple. During the meeting AI isn’t there, and that’s how it should be: the meeting is the human moment. But before and after it works well, and those are the two moments where time usually gets lost. Before: agenda, brief for participants, likely questions. After: extract of decisions, action items, bridge to those who weren’t on the call. If the meeting runs an hour, ten minutes of prep and ten of wrap-up save you twice that, multiplied across everyone on the call.
Before the meeting
Section titled “Before the meeting”Structured agenda
Section titled “Structured agenda”A useful agenda has four fields for each item: objective, who owns it, time, expected decision. Miss one, and the item drifts or ends without reaching a conclusion.
AI handles this step well because the format is rigid. A prompt that works: “organize these items into a table with objective, owner, time in minutes, expected decision. Total cap 50 minutes.” The total cap forces it to cut rather than stretch. You then revise where its time estimates are off (you know the room, it doesn’t) and adjust.
Preparation brief
Section titled “Preparation brief”A meeting where everyone shows up without having read anything starts with ten minutes of catch-up. Ten minutes across eight people is eighty minutes, more than twice what it takes to write a brief.
A good brief answers three questions: what to read beforehand (one or two links max, not fifteen), what to bring (data, an opinion, a decision already made), what to expect at the end. The AI produces it in two minutes if you hand it the agenda and a line or two of context. Rule of thumb: under 150 words, or nobody reads it.
Anticipating objections
Section titled “Anticipating objections”Before an important presentation, ask it to do the opposite of your job. The prompt only works if you give it context: not “I’m presenting X to team Y” but “I’m presenting the new roadmap to the sales team, they want short-term results, we’re proposing six months of technical investment; last quarter there was friction on a similar decision. What questions are they likely to ask me? What objections are likely to come up?”. Three ingredients usually do it: participants’ roles, what’s at stake in the decision, relevant recent history. The AI plays devil’s advocate (challenges your ideas to surface the weak spots), lists a dozen questions, you spot three you hadn’t thought of, and you prepare the answer.
It isn’t a fortune teller: some objections you anticipated never come up, others do that the AI hadn’t foreseen. But on average it gets 60-70% of the likely questions right, and prep time shrinks. A second round in the style of Iterate the conversation often pays off: “of these, which are the three most likely and why?”. The second list beats the first.
During
Section titled “During”During the meeting, the AI isn’t at the table. If you record, do it yourself, and only with consent: stated at the start for routine internal meetings, in writing for formal cases or with external parties. Otherwise don’t record. Handwritten notes, a transcription app on your laptop, pen-and-paper minutes: those are all options, AI as an active participant isn’t.
The reason is simple: knowing an AI is listening changes how people speak. If the meeting touches a conflict, a negotiation, delicate feedback, recording is decided with the participants, not assumed.
If you record, keep the file on your device and don’t upload it to a third-party service before clearing it with whoever was on the call: the transcript contains data from people who didn’t consent to an AI listening in.
After the meeting
Section titled “After the meeting”Actionable extract, not a generic summary
Section titled “Actionable extract, not a generic summary”This is where most people get it wrong. “Summarize the meeting” produces a paragraph nobody reads: fine for the archive, useless for getting things done.
The prompt that works has a rigid structure: *“from this transcript (or these notes) extract, in three separate sections:
- Decisions made, 2. Action items with owner and deadline, 3. Open questions. No narrative.”* The difference is huge: you get an operational list you can already turn into tickets in your system.
It also works on messy notes you jotted down yourself, not just clean transcripts: the quality of the output follows the quality of the input. And a quick completeness check: run down the agenda you had prepared and make sure every item actually discussed shows up at least once, across decisions, action items, or open questions. If a point is missing from the extract, it’s usually the AI that skipped it, not that nobody raised it.
The follow-up email
Section titled “The follow-up email”Once the core is extracted, the follow-up email is just formatting. You ask “write the follow-up to the team based on this, 150 words max, closing with the next steps” and you have the draft to tweak.
The draft isn’t perfect: you usually adjust the tone (often too formal), add the detail you know (the client moved the next call up, the budget needs revisiting), cut the action item that’s actually already in motion. The five minutes you spend are much less than the thirty you’d spend from a blank page.
Update for those not on the call
Section titled “Update for those not on the call”People who weren’t on the call need less detail but a clearer frame. Prompt: “for people who weren’t in the meeting, 4-5 bullets on what changed and what they need to know for their work”. It isn’t the summary, it’s a selective extract, already tailored to the recipient. Five extra words of context in the prompt (“the recipient is the legal team, only risks matter”) change a lot here: the 4-5 bullets that come out are the right ones, not generic ones.
A concrete example
Section titled “A concrete example”You’ve just wrapped a weekly status meeting for the product team (six people, 45 minutes), and you have a transcript from the call tool. You hand it over like this:
In two minutes you have an output that’s already the backbone of the follow-up, and the two topics the meeting didn’t close are right there in the open questions. The same extraction by hand would be twenty minutes of rereading.
What NOT to do
Section titled “What NOT to do”Don’t delegate the call on priorities. The AI extracts the action items, but the order they go in is yours. It depends on who’s in more of a hurry, what the client expects, internal constraints it doesn’t know about. A generically ordered list is often ordered badly for your case.
Don’t share sensitive transcripts without thinking. The rule gets set before the meeting, not after. If the call touches topics where your company has NDAs or personal data, decide in advance what a public AI can see and what it can’t. The broader company case is covered later in the module, in the lesson on company data and privacy.
Don’t skip reviewing the transcript. Automatic transcripts contain errors: names wrong, technical terms mangled, sentences cut off. A quick pass to fix at least the names and the most egregious bits before handing it over, otherwise the extract comes out wrong in a plausible way, which is worse than obviously wrong.
What comes next
Section titled “What comes next”A meeting often produces slides, reports, or both. The next two lessons shift the module to those formats: Slides and presentations on slide content, Working with data and tables on numbers and spreadsheets.