Because a summary is organised the way the meeting happened rather than the way the follow-up work needs to happen. It informs you and then leaves you exactly where you were, because there is no action in it. The fix is to stop asking for a summary and instead ask the same two or three fixed questions of every transcript, aimed at the next action rather than the recap.
I ran meeting summaries for months before I noticed I was not doing anything with them.
What a summary actually gives you
It gives you the meeting again, shorter.
That feels valuable, and there is a specific moment of relief when it lands in your inbox, and then you read it once and file it. Nothing in it tells you what to do, because it was never asked to. It was asked to represent the conversation faithfully, and it did.
The structural problem is ordering. A conversation is ordered by when things came up, which is nearly random. It depends on who spoke first, what tangent someone went down, and whether the person with the important objection raised it at minute four or minute forty. Follow-up work has a completely different order: what is urgent, what is blocked, who owes whom.
So you are handed a document sorted by one logic and you need it sorted by another, and the translation work is left to you, and that work is most of the job.
Ask fixed questions instead
The change is small and it is the whole thing. Instead of asking what happened, ask the same short set of questions of every transcript.
Three that work:
What did this person say, in their own words, about the problem they have? Their words, not paraphrased. This is the input that later feeds your copy, your positioning and your product decisions, and paraphrasing sands the useful edges off it.
What did I commit to, and by when? Commitments made out loud in conversation are the most reliably forgotten thing in any business. You said you would send the thing. You meant it. It is now Thursday.
What decisions came out of this, and who owns each one? Decisions without owners are not decisions. This question surfaces the ones that were made ambiguously, which is most of them.
Same three questions, every transcript, every time. The consistency is what makes the output usable, because you learn to read it fast and you notice immediately when a field comes back empty.
The Meeting to Deliverable Playbook
This is the second of six playbooks we teach, and it is second for a specific reason: it produces finished work out of something you were already doing.
That matters more than it sounds. Systems built for aspirational work fail, reliably, because nobody maintains a system for a task they were not doing anyway. You were already having the meetings. You were already recording some of them. The playbook costs you no new behaviour, which is why it survives.
The build.
- Pick one recurring meeting type. Sales calls, or team check-ins, or discovery conversations. One type, not all meetings.
- Get the transcript into your context brain. If you have not built one yet, start with the context playbook, because the output quality depends on it more than on anything in this post.
- Write your fixed questions. Two or three. Same ones every time.
- Run it on your last recorded call and read the output against your memory of the conversation.
- Adjust the questions, not the transcript. If the output is wrong, the question was wrong.
- Once three consecutive calls produce useful output, put it on a schedule.
The mechanics of the capture side are covered separately in what Otter is and how it works with Claude and how to automate meeting follow-ups.
The trap
Asking for a summary anyway, because it is what the tool offers by default and because the button is right there.
Almost every transcription product now has a summarise function and it is genuinely good at summarising. The quality of the summary is not the issue. The issue is that a summary is the wrong artifact, and a very good version of the wrong artifact is still the wrong artifact.
The second trap is asking too many questions. Six or seven fields feels thorough and produces a document nobody reads, which puts you back where you started with extra steps.
What you end up holding
Two or three finished deliverables from every recorded conversation, drafted and waiting for you to approve. The follow-up email. The brief for whoever was not in the room. The decisions that need owners.
Not a record of the meeting. The work the meeting was supposed to produce.
Frequently asked
Why are AI meeting summaries not useful? Because a summary is ordered by when things were said, and follow-up work is ordered by urgency and ownership. You are handed a document sorted by the wrong logic and left to translate it yourself.
What should I ask AI about a meeting transcript? Fixed questions aimed at the next action. What the person said in their own words about their problem, what you committed to and by when, and what decisions were made and who owns each.
How do I turn meeting recordings into finished work? Connect the transcript to a context brain holding your positioning and voice, then run the same fixed questions on every transcript rather than requesting a summary.
What is the best AI meeting notes tool? The tool matters far less than what you ask it. Any transcription that hands a clean transcript to a workspace holding your context will work.
How long does this take to set up? About forty-five minutes for the first version, and it runs on a schedule afterwards.