A chaotic calendar isn’t usually a problem of too many meetings. It’s a problem of meetings landing in the wrong places: high-stakes calls back-to-back, prep time carved out before the wrong meeting, recurring 30-minute syncs that should have been killed two quarters ago. Per-meeting scheduling tools don’t fix any of that, because the chaos is at the week level. The fix is a weekly review that runs every Friday afternoon and reshapes the next week before it hits.
This post is about that review. Not booking individual meetings (that’s a separate workflow). The thing AI is genuinely good at, when given the right inputs, is reading a week-shaped chunk of calendar and telling you where the trouble is.
The weekly review prompt
Friday, 4 pm, before you sign off. Open Microsoft Copilot (best for native Microsoft 365 calendar access), Gemini for Google Workspace, or paste a calendar export into Claude. Run:
“I’m doing a weekly review of [EXEC]‘s calendar for next week.
Exec’s energy pattern: Best-thinking hours: [E.G. 9–11 AM] Drop-off hours: [E.G. 3–5 PM] Hard rules: [E.G. NO MEETINGS BEFORE 9, MAX 3 BACK-TO-BACKS, MONDAY MORNINGS PROTECTED FOR PLANNING, FRIDAY 3 PM ONWARDS NO NEW BOOKINGS]
External commitments next week: [LIST ANY TRAVEL, EVENTS, TIME-OFF]
Look at the calendar I’m pasting/sharing and answer four questions:
Which meetings next week are in the wrong slot for the exec’s energy pattern? (E.g., a strategic decision in a 4 pm slot when best-thinking hours are mornings.) For each, suggest a swap.
Where are there four or more back-to-backs in a row? Flag and propose a buffer.
Which prep blocks are in the wrong place (too far before, too soon after, missing entirely for high-stakes meetings)? List the affected meetings.
Is there at least one 90-minute uninterrupted block for deep work? If not, where’s the candidate slot to clear?
Output as a four-section list, with specific times and meeting titles, not generalities. Don’t make changes, just propose them.”
The four questions are the whole point. Each one is a different lens on the same calendar, and the model can hold all four simultaneously in a way you can’t when you’re scrolling Outlook on a Friday afternoon.
What you do with the output
You read it, decide which proposed changes are real (some won’t be, the model doesn’t know the politics), and run a follow-up:
“I’m accepting these specific suggestions: [LIST].
For each one, draft the email I’d send to the affected attendees explaining the move. Keep each under 80 words, voice: [PASTE EA VOICE CARD], no over-apologising. If a meeting needs to move outside business hours for an attendee in a different time zone, flag that and ask me before drafting.”
Now you have your reshape plan and the comms to make it happen, and you can have it all queued by 5 pm Friday.
The recurring-meeting audit, monthly
This is the one that returns the most time over a year. Every month, run this prompt:
“Here’s a list of every recurring meeting on [EXEC]‘s calendar this month, with title, frequency, attendees, and average duration:
[PASTE LIST OR CALENDAR EXPORT]
For each, give me:
- Plausible original purpose (your inference based on title and attendees).
- Has it produced a recent decision or just a status update?
- Could it be: kept as-is, halved in length, halved in frequency, replaced by an async update, or killed?
- The case for keeping it (one sentence) and the case for cutting it (one sentence).
Sort by ‘most likely killable’ first. Don’t recommend killing the 1:1 with [EXEC]‘s direct boss or board-level recurring meetings; just note them as protected.”
Then walk into your exec’s office Monday morning with the top three killable candidates. Most execs cancel 20 to 30% of their recurring meetings the first time someone presents them like this, because the meetings were on the calendar from 18 months ago and nobody has had time to question them.
The energy-mapping observation
McKinsey’s State of AI 2025 reports that only 39% of organisations can attribute any EBIT impact to AI, a reminder that adoption alone doesn’t move outcomes. The same is true at the calendar level. Adopting an AI scheduling tool doesn’t help if the underlying pattern of the week is broken. Energy mapping is the part that moves the needle: getting your exec’s two hardest meetings into their two strongest hours and pushing routine syncs to the post-lunch slump where they belong.
A model can run this analysis in seconds. A human couldn’t, because holding 25 meetings in your head and running four different evaluations against them simultaneously is the exact kind of thing the model is for.
The failure mode to avoid
Treating the model’s suggestions as orders. The model can see the calendar but not the political context: that the 4 pm Tuesday slot is in the wrong place for energy reasons but is the only time a key board member can do, that the back-to-back you’d want to break is actually a deliberate “get all three updates done together” pattern your exec prefers. Your job in the review is to filter the suggestions through context the model doesn’t have.
The version that goes wrong: an EA accepts every model suggestion, sends a flurry of reschedule emails Friday evening, and Monday morning everyone is upset. The version that works: pick three to five moves with the biggest impact, leave the rest, and never touch a recurring meeting on calendar without checking with the exec first.
A counterintuitive note
The single best calendar move you can make for most executives isn’t adding focus time. It’s adding decompression time. A 30-minute open block after a high-stakes meeting where the exec can write, walk, or stare at a wall is the slot that keeps them sharp for the next thing. The energy-pattern prompt above will surface the missing decompression slots if you ask it to. Most calendars have zero. Two well-placed 30-minute blocks across a week are worth more than three new “focus time” entries on Friday afternoon.
What stays human
The relationships. The model can recommend killing a recurring sync, but you’re the one who knows that the sync is the only weekly contact between two teams that otherwise drift apart, and that killing it has knock-on effects no calendar can show. Read every recommendation, accept the ones that are clearly right, defer the ones that touch sensitive territory until you’ve talked to your exec. The system isn’t a calendar autopilot, it’s a Friday afternoon strategic input that takes 25 minutes and reshapes the entire next week.
The executive assistants who run calmer calendars don’t have fewer meetings. They have meetings that landed in the right slots, with prep where prep was needed, and the killable recurring ones gone. AI made the analysis fast enough that a weekly review actually fits into a real Friday.