The client-meeting workflow that turns transcripts into next actions

Four client meetings in a week can easily mean twelve minutes of prep before each one and twenty minutes after each one writing follow-ups, summarizing what was decided, and turning vague conclusions into action items. That's two hours at least, most of it at awkward times and right when switching between tasks costs the most.

The expensive part of a meeting is the context reload before it and the synthesis after it. Given the right inputs, AI can do nearly all of both.

The meeting that causes trouble later is often one that went well. Because it went well, nobody took careful notes, and three weeks later it's hard to remember what was decided and what was only discussed. That's the gap this workflow closes.

This is a Category 3 task — structured, repeatable, doesn't really need your voice in the artifacts (here's the broader frame). The default should be: AI handles the meeting paperwork unless you spot a reason it shouldn't.

The three artifacts

For every meaningful client meeting, three artifacts:

1. Pre-brief (5 minutes) — context reload. What's the goal of this meeting, what was decided last time, what's the one decision I need to drive today.
2. Debrief (5 minutes) — the synthesis. What was decided, what was deferred, what's the next action and who owns it.
3. Follow-up email (2 minutes) — the version of the debrief you send to the client, in their language.

That's twelve minutes per meeting, against thirty or more by hand. Done by hand, it's easy to manage only one of the three reliably, and that's how action items disappear between weeks.

The debrief and follow-up depend on you having a transcript or solid notes. If your meetings aren't transcribed, fix that first. Otter, Granola, Fathom, Zoom's native transcription — pick one, install it once, never think about it again. The rest of this article assumes you have one.

Artifact 1: The pre-brief (5 minutes)

A 1-page document you read in the five minutes before the call starts. The job: surface the one decision you need to drive today, not the meeting agenda the client sent.

The prompt:

You are my prep assistant. I have a client meeting in [TIME, e.g.,
"30 minutes"]. Generate a 1-page pre-brief.

Sections:

1. **Goal of this meeting** — 1 sentence. The single decision or
   outcome I need by the end.
2. **Where we left off last time** — 3-4 bullets from the prior
   meeting's debrief.
3. **Open items I owe them** — what they're expecting from me.
4. **Open items they owe me** — what I'm waiting on.
5. **The one question I need answered today** — the question, if
   nothing else gets answered, that I cannot leave the meeting without.
6. **Risk to flag** — anything I noticed in the engagement that the
   client should know about, in case it doesn't come up organically.

Inputs:
- Client / engagement context: [PASTE OR REFERENCE INTERNAL KICKOFF DOC]
- Prior meeting debrief: [PASTE]
- My recent thinking / unresolved questions: [WHATEVER'S IN YOUR HEAD]

The "one question I need answered today" line is the discipline. Most client meetings drift because the consultant didn't decide in advance what they actually needed. Five minutes of pre-brief means the meeting probably arrives at the decision; without the pre-brief, the meeting probably doesn't.

Artifact 2: The debrief (5 minutes)

A 1-page document you produce immediately after the call. This is for you, not the client. It captures decisions, action items, surprises, and your read on the room — all of which fade fast.

The prompt:

Here's the transcript or my notes from a client meeting that just
ended. Produce a 1-page debrief for my files.

Sections:

1. **Decisions made** — what was actually agreed. Each as a one-line
   bullet. If something was discussed but not decided, that's not a
   decision — flag it for next time.
2. **Action items** — each with owner (me / them / specific person)
   and due date if mentioned. If no due date, propose one.
3. **What was deferred** — items raised but pushed to a future meeting,
   with my read on whether that's safe or whether deferring is risk.
4. **Surprises** — anything that came up that wasn't on the agenda.
   Most useful section. Often the real signal.
5. **My read on the room** — how the client seemed (engaged?
   distracted? confused?). Flag if any answer changed mid-meeting —
   that often indicates an internal disagreement they're working out.
6. **What I owe them in the next 48 hours** — 1-3 specific things to
   do *fast* while the meeting is fresh.

Transcript / notes:
[PASTE]

Two sections do most of the work. Surprises — anything unexpected that came up — is information about what the client actually cares about, which is often different from what they say they care about. My read on the room catches a good share of what an experienced consultant would notice. AI is decent at it because transcripts carry tone: commitment rising and falling, mid-sentence hedges, the moment someone went quiet.

The bug to watch for: AI sometimes promotes things that were discussed into things that were decided. Always scan the decisions list against your own memory of the meeting. If you're not sure whether something was decided, ask the client to confirm — better to look careful than to misremember.

Artifact 3: The follow-up email (2 minutes)

The version of the debrief you send to the client within two hours of the meeting. Follow-up emails tend to over-summarize, but clients don't need the recap. They were in the meeting. They need the decisions, the next actions, and the deadline.

The prompt:

Based on the debrief above, draft the follow-up email I'll send to
the client within the next 2 hours.

Email rules:
- Subject line that names the one most important decision/next step
- 5-8 sentences total
- Lead with the decisions made, not the discussion
- Include my action items + their action items, each with a
  proposed date
- Close with the next meeting time + the one thing I need from them
  before then
- Do NOT include the full meeting summary. They were there.
- No "Thank you for the time"

Recipients: [LIST WHO]
Reply-all or just to primary contact? [INDICATE]
Anything to omit because it's internal-only: [LIST]

The "do NOT include the full summary" constraint is what makes this prompt produce the right output. Without it, AI writes a meeting-minutes email that's far too long and reads like it's covering itself.

A worked example

Two-week-in check-in with Marcus and Sarah at a fictional B2B food-distribution startup. The ops playbook documentation is mostly done. Sarah surfaced midway through that her real concern isn't the playbook — it's whether the new Boise hires will follow it once written. Marcus thinks they will. There's a disagreement neither of them named directly.

The debrief output would surface this:

Decisions made.

• Final playbook draft delivered Friday, May 8.
• Two training sessions scheduled: May 13 (Seattle ops team), May 15 (Boise ops team).

Surprises.

• Sarah doesn't fully trust that the Boise hires will follow a written SOP — she said "hiring a new ops manager would be cheaper than writing this" with a laugh, but Marcus didn't laugh. There's a hiring-vs-documenting tension between them. Worth surfacing in the next 1:1 with Marcus.

My read on the room. Marcus engaged but a little defensive when Sarah pushed on Boise. The real engagement-risk isn't the deliverable — it's whether they're aligned on what success looks like.

What I owe in 48 hours.

• Send Sarah and Marcus a 2-paragraph reframe of the engagement: documentation is necessary but not sufficient.
• Schedule a separate 30-min with Marcus, no Sarah, to surface the hiring-vs-documenting tension.

That's the kind of debrief you can't reliably produce by hand 90 minutes after the meeting when your mind has moved on. Five minutes with the transcript and you have it.

Pitfalls

• Trusting the "decisions made" list without checking. AI sometimes promotes discussion into decisions. Always scan against your own memory.
• Sending the debrief output as the follow-up email. It's too long, too internal, too analytical. Run the follow-up prompt to translate it for client consumption.
• Skipping the pre-brief because the meeting "is just a check-in." Check-ins are where engagements quietly drift. The 5-minute pre-brief is your hedge.
• Over-trusting "my read on the room." It's a hypothesis from a transcript. Update on the next meeting.

What to do this week

Before your next client meeting, run the pre-brief prompt thirty minutes ahead. Run the debrief right after, while the transcript is fresh, and send the follow-up within two hours.

Try it for three meetings. By then you'll know whether it saves you the two hours a week it saves on paper.

Going deeper

This article distills the meeting workflow from Chapter 5 of The Solo Consultant's AI Playbook, which adds a worked example showing the pre-brief informing the debrief informing the next pre-brief — the chain that makes a multi-meeting engagement feel coherent rather than like seven disconnected calls.

A prompt that worked once still has to be rebuilt next Tuesday; a brief can just be run again. The Claude Workspace Kit ships six briefs, each naming the inputs it expects and pairing them with a written standard for the finished document, which is what makes the output checkable rather than merely plausible.

The framing for which tasks to give AI at all is in the first article. Meeting paperwork is squarely Category 3, which is why AI can do so much of it.

All of our procedures for the work between the work are on one page: Run Your Practice.

— Digital Kreative