How to avoid the feeling of "this meeting could have been an email" pt3

Parts one and two of this series looked at how to audit where your time is going and how to make the meetings you do run worth attending. This instalment is about something slightly different: the quiet, persistent overhead that surrounds meetings — the preparation before them, the documentation after them, the decisions that should have been captured clearly but weren't — and how AI can take a significant portion of that off your plate.

Because the issue was never really the meeting itself. It was everything around it.

Start with the data

Before you can improve anything, you need to understand the current state of play. Part one of this series covered how to export a calendar into a spreadsheet using Google Apps Script. That approach still holds — but if you'd rather not touch any code, AI has made the audit considerably simpler.

Paste a week or month's worth of calendar events (exported from Google Calendar as a CSV, or simply copied from your screen) into a tool like Claude or ChatGPT and ask it to categorise and analyse the data for you. Ask it to group meetings by type — strategic, operational, administrative, external — and calculate the total time spent in each category. Ask it to identify where your principal is double-booked, where meetings consistently run long, and where a pattern of recurring obligations is consuming time that ought to be used differently.

What you get back isn't just numbers. It's a clear, shareable picture of how the calendar is actually functioning — which is a very different thing from how it's supposed to function. That distinction tends to make for a more focused conversation in a 1:1.

Agendas that actually reflect the meeting

One of the most common agenda problems is that they don't change. The same items appear in the same order, week after week, regardless of what's happening in the business. This is partly habit, and partly because building a genuinely responsive agenda takes time that isn't always available.

AI can help break that loop — but only if you feed it the right material. Before drafting an agenda, share the previous meeting's minutes, any presentations that have been circulated in the intervening period, and a brief note on what has changed or what's currently pressing. Ask it to identify recurring themes from the minutes, flag any actions that remain outstanding, and suggest agenda items based on what the pre-read material indicates needs discussion.

The output won't be perfect, and it shouldn't be submitted unedited. What it will do is surface connections and carry-overs that are easy to miss when you're working from memory and a packed inbox. You review, reorder, and add the things only you would know — the conversation you had with a department head on Thursday, the concern your principal mentioned in passing — and you end up with an agenda that is both efficient to produce and genuinely fit for purpose.

Minutes that are useful beyond the meeting itself

Minutes are one of the most undervalued documents in any organisation, and one of the most poorly executed. Too often they read either as a near-verbatim transcript or as such a high-level summary that they're meaningless to anyone who wasn't in the room. Neither version is useful.

AI transcription and summarisation tools — Otter.ai, Fireflies, and similar — have made it significantly easier to capture meeting content without diverting your attention from the discussion itself. The real value, though, is in what you do with the transcript afterwards. Feeding a clean transcript into an AI and asking it to produce structured minutes — with agenda items as section headers, decisions clearly separated from discussion, and a distinct section for agreed actions with named owners and deadlines — produces a considerably more useful document than most minute-takers produce under time pressure.

It also frees you to be present in the meeting rather than furiously typing. The quality of your contribution to the room is almost certainly more valuable than a verbatim record of what was said.

The gap between what was said and what was decided

This is where the gap between what people think was decided and what was actually agreed tends to live.

Meetings frequently end with a general sense that something has been concluded without anyone having stated it plainly. People leave with different impressions. The follow-up email dances around the decision. Three weeks later, someone asks why nothing has moved forward.

After a meeting, take the transcript or your own notes and ask an AI tool to extract every decision or agreement made during the session, and every action committed to — including who committed to it and by when. Ask it to flag anything that sounds like it might be a decision but wasn't explicitly confirmed. Ask it to identify topics that were raised but not resolved.

Read back through its output critically. You will almost certainly find at least one item where the agreement is less clear than you remembered, or where an action was implied but not owned. Catching that before the minutes are circulated is considerably less complicated than chasing it down after the fact.

The time this gives back

There is a version of meeting management where a senior EA or Chief of Staff spends three to four hours a week on meeting-adjacent administration — preparing agendas, writing up minutes, chasing actions, resending summaries. That time is not negligible, and it is almost entirely transferable to AI-assisted processes without any loss of quality. In most cases, with a marked improvement in it.

The caveat is that AI will not do this well without clear inputs and a critical review of its outputs. Vague prompts produce vague results. A transcript fed in without any context about the meeting's purpose or the organisation's terminology will produce generic minutes that need significant rework. The skill is in knowing what to give it, what to ask of it, and what still requires your judgement — which is, when you think about it, not so different from how the best EAs have always operated.

The meeting itself is only as good as what surrounds it. That part has always been yours to own. AI just makes it faster.