How to Run Better Meetings with AI

Why meetings break

Most meetings don't fail in the room. They fail at the bookends.

A meeting fails before it starts when no one has named the decision it exists to make. People arrive with the topic ("Q3 roadmap") but not the question ("Which two initiatives do we cut to protect the launch date?"). Without a named decision, discussion has no terminal condition. It expands to fill the calendar block, and everyone leaves with a different idea of what was concluded.

A meeting fails after it ends when no one owns what was agreed. Decisions get made, but they evaporate because they were spoken, not recorded, and assigned to "the team" instead of a person with a date. By the next session, half the action items are forgotten and the other half are re-litigated.

The middle of the meeting — the part everyone tries to fix with better facilitation, stricter time-boxing, or a louder timer — is rarely the real problem. The real problem is the setup and the teardown. AI is unusually good at exactly these two jobs: structuring a decision before the meeting and converting a messy transcript into owned actions after it. It does not make you a better facilitator. It removes the work that stops you from being one.

This is a three-part workflow: Prep → Capture → Follow-up. Each part is a single repeatable prompt. None of it requires special tooling — a chat model and your calendar are enough to start.

The workflow at a glance

• Prep — Before the meeting, generate a one-page brief that names the decision, the inputs needed, and the agenda. This forces the question to exist before anyone walks in.
• Capture — During the meeting, structure the live notes into four buckets: Decisions, Actions (with owner and date), Open questions, and Parking lot.
• Follow-up — Within 15 minutes of ending, turn the notes or transcript into a recap that anyone can act on without having attended.

The discipline is in keeping the structure identical every time. When your team sees the same four buckets and the same recap shape week after week, the format becomes invisible and the content becomes the only thing they read.

Prep — the brief

The single highest-leverage move is writing the brief, because it's where you name the decision. Here is a copy-pasteable prompt. Fill the placeholders and run it the day before.

You are helping me prepare a one-page brief for a meeting I'm running.

Meeting: [meeting name]
Attendees and roles: [names — role, e.g. "Priya — eng lead, Marcus — design"]
The decision we must leave with: [the single decision, phrased as a question]
Why now: [the deadline or trigger forcing this decision]
Context the room needs: [2–4 bullets of background, links, prior decisions]
Constraints: [budget, headcount, dates, anything fixed]

Produce:
1. DECISION TO MAKE — one sentence, phrased as a question.
2. WHY THIS MEETING EXISTS — two sentences max.
3. INPUTS NEEDED — what each attendee should bring or have read.
4. AGENDA — 3–5 items, each with a time box and the sub-question it answers.
5. WHAT "DONE" LOOKS LIKE — the artifact we leave with (a choice, a number, an owner).

Keep it under one page. No filler. If the decision is unclear or
unanswerable as written, say so and ask me one clarifying question.

The last line matters. A good brief prompt should push back when the decision is vague. If the model can't tell what you're deciding, neither can your attendees.

Before — the typical calendar invite:

Q3 Roadmap Sync Let's align on priorities for next quarter. Bring your thoughts.

This names a topic, not a decision. "Align" is not a verb anyone can complete. There's nothing to prepare and no way to know when the meeting is over.

After — the generated brief:

Q3 Roadmap Sync — 45 min Decision to make: Which two of our five candidate initiatives ship in Q3, given we can staff exactly two? Why this meeting exists: Eng capacity is fixed at two initiatives and the board update is in three weeks. We decide here or we slip. Inputs needed: Priya — effort estimate per initiative. Marcus — design readiness. Me — revenue impact ranking. Agenda: (1) Confirm capacity constraint — 5 min. (2) Walk the five candidates against impact and effort — 20 min. (3) Cut to two — 15 min. (4) Name owners and first milestone — 5 min. What "done" looks like: Two initiatives chosen, each with a named owner and a first milestone date.

The difference is not polish. It's that the second version cannot be attended passively. Everyone knows what they owe and what the room is for.

Variants. For a 1:1, the decision is rarely a single choice, so swap the decision line for "What does [report] need from me to stay unblocked this week, and what do I need to flag?" and replace the agenda with three buckets: their topics, your topics, growth. For a status or information-sharing meeting, be honest — if there's no decision, ask whether it should be a meeting at all, or a written update. The brief prompt is most useful precisely because it exposes meetings that don't need to exist.

Capture — during the meeting

The job during the meeting is to keep one running document with a fixed structure, so that nothing said gets lost and nothing decided stays unowned. Four buckets:

• Decisions — what was actually concluded, in past tense.
• Actions — each with an owner (a person, never a team) and a date.
• Open questions — things raised that need an answer but weren't resolved.
• Parking lot — relevant but off-topic; captured so you can move on without losing it.

Be honest about how this gets populated. There are two modes, and they trade off effort against control.

Manual. One person — ideally not the facilitator — keeps the four-bucket doc live during the meeting. This is the most reliable method because a human applies judgment in real time about what counts as a decision. The cost is that someone is half-out of the conversation.

Automated. A transcription tool (your video platform's built-in transcript, or a dedicated notetaker) records the meeting, and you run the prompt below on the transcript afterward. This frees everyone to participate, at the cost of accuracy — models will occasionally invent an owner or miss an implicit decision, so the output always needs a human pass before it goes out.

Either way, the same prompt does the structuring. Paste raw notes or a transcript:

Below are raw notes/transcript from a meeting I ran. Organize them into
exactly four sections. Do not add anything that isn't supported by the text.

DECISIONS — what was concluded. Past tense. One line each.
ACTIONS — each line: [Owner name] — [action] — [due date].
  If an owner or date is missing, write "OWNER?" or "DATE?" so I can fill it.
  Never assign an action to a team or "we" — flag it instead.
OPEN QUESTIONS — raised but not resolved.
PARKING LOT — relevant but off-topic, captured for later.

Raw notes/transcript:
[paste here]

The OWNER? and DATE? flags are the point. They surface the exact failure mode from the top of this article — the unowned action item — instead of papering over it with vague language. An action with OWNER? next to it is a question you must answer before the recap goes out.

Follow-up — the 15-minute recap

The half-life of a meeting's energy is short. If the recap lands within 15 minutes, people act on it while the context is fresh; a day later it reads like archaeology. Run this on your captured notes:

Turn the four-section notes below into a recap email that someone who
missed the meeting could act on. Structure:

- One-line summary: what we decided, in plain language.
- Decisions: bulleted, past tense.
- Action items: a table — Owner | Action | Due date. No blanks; if I left
  an OWNER? or DATE?, flag it at the top under "Needs an owner/date."
- Open questions: who will resolve each and by when.
- Next checkpoint: when we confirm these are done.

Tone: direct, no filler, no thanking-everyone preamble. Under 200 words.

Notes:
[paste your four-section notes]

A sample of what comes back:

Subject: Q3 Roadmap — decided

Summary: We're shipping Onboarding Revamp and Billing v2 in Q3. The other three candidates are deferred to Q4 review.

Decisions:

• Capacity is fixed at two initiatives for Q3; confirmed, not revisited.
• Onboarding Revamp and Billing v2 selected on impact-to-effort.
• Reporting Dashboard, SSO, and Mobile deferred to the Q4 planning cycle.

Action items:

Owner	Action	Due
Priya	Staffing plan for both initiatives	Jun 20
Marcus	Design-readiness check on Billing v2	Jun 18
Me	Send board the chosen two with rationale	Jun 16

Needs a date: Onboarding Revamp first milestone — Priya to propose by Jun 20.

Open questions: Does Billing v2 need legal review? — Marcus to confirm by Jun 18.

Next checkpoint: Standing roadmap sync, Jun 23. We confirm all three actions closed.

This recap is usable by someone who wasn't in the room, which is the real test. It names owners, names dates, and names the moment those commitments get checked.

Rollout and measurement

Do not roll this out across every meeting at once. The structure has to become habit, and habit forms on one repetition at a time.

Pick one recurring meeting — a weekly team sync or a standing 1:1 — and run all three parts for four weeks. A recurring meeting is the right choice because it gives you a built-in checkpoint: last week's actions are this week's first agenda item, which is how you measure whether follow-up is actually working.

Track three numbers, because "it feels better" is not evidence:

• Prep time per meeting. Time-box yourself before and after adopting the brief prompt. The brief should cost less than ten minutes and replace the scramble in the five minutes before the call.
• Percentage of action items with a named owner and date. Count them in the recap. The target is 100 percent — every OWNER? and DATE? resolved before sending. This is the single metric most tied to whether work actually happens.
• Percentage of action items closed by the next session. This is the outcome metric. If owned, dated actions still aren't closing, the problem has moved past meetings and into capacity or priorities — which is itself worth knowing.

Once one meeting holds the pattern for a month, the format will travel on its own. People who receive a clean recap start expecting one, and the discipline spreads by demand rather than mandate. That is the point of fixing the bookends: not a better meeting, but a meeting whose decisions survive contact with the week that follows.