Scheduling looks like a calendar problem. It’s actually a constraints problem. Time zones, travel buffers, focus blocks, prep time, and the exec’s “no morning meetings on Mondays” rule all have to be honoured before a slot is real. Generic AI scheduling prompts ignore most of them, which is why the time the model picks looks fine on paper and then collapses the moment your exec sees it.
The fix is to write the constraints down once. Then every scheduling prompt starts with that paste, and the model is choosing inside a frame instead of guessing.
Write your exec’s scheduling constraints in plain English
Open a doc and answer these. Be specific.
- Time zones the exec works across this quarter.
- Hours the exec considers “meeting hours” (e.g. 9 to 11, 1 to 4).
- Hard no-go windows (school run, gym, weekly 1:1 with their boss, focus block).
- Travel buffer needed between in-person meetings (15 min, 30 min, an hour for cross-town).
- Prep time required before high-stakes meetings (client first-meet, board call).
- Maximum back-to-back meetings before the exec is fried (often three).
- Lunch protocol (locked, flexible, working through is fine).
- Day-of-week patterns (e.g. Monday mornings are strategic, Friday afternoons are wind-down).
This list, two paragraphs at most, becomes your constraints block. Save it where you can grab it in a second.
The booking prompt
Now for the actual scheduling. Open Microsoft Copilot (which can read native calendars in Microsoft 365), Gemini for Google Workspace, or paste a calendar export into ChatGPT and run:
“You are scheduling a meeting for [EXEC NAME, ROLE].
Constraints (always honour these): [PASTE CONSTRAINTS BLOCK]
Meeting details: Topic: [TOPIC] Length: [MINUTES] Attendees and their time zones: [LIST] Importance: [ROUTINE / IMPORTANT / HIGH-STAKES] Prep needed: [YES/NO, AND HOW MUCH] In-person, virtual, or hybrid: [PICK] Earliest the meeting can happen: [DATE] Latest: [DATE]
Look at my exec’s calendar (next 14 days) and propose three options. For each option, give me:
- Date and time in the exec’s local zone, plus each attendee’s local time.
- What’s directly before and after on the exec’s calendar (so I can see prep and recovery space).
- One reason this slot is good and one reason it might not be.
Don’t pick a slot that breaks any constraint. If no slot works in the date range, say so and tell me which constraint is the binding one.”
That last sentence is the one that saves you. Without it, the model will quietly bend a rule (skip the prep buffer, ignore the no-go window) to give you three options. With it, the model tells you the truth: there’s no slot, and here’s why, so you can renegotiate the constraint with your exec instead of with reality on the day.
The scheduling email that doesn’t sound like a robot
Once you have a slot, you still need to send the invitation note. Drop the chosen slot back in:
“Draft the email confirming this meeting. Voice: [PASTE EXEC OR EA VOICE CARD IF YOU HAVE ONE].
Include: meeting topic in one short sentence, time in each attendee’s zone, the video link or address, two lines on what they should bring or read in advance, and a clear next step if they need to reschedule.
Keep it under 90 words. No ‘I hope this finds you well’. No bullet lists unless there’s actually a list to make.”
Two prompts, one calendar invitation, no rewrite cycle.
A worked example with time zones
Your CEO is in London. The other three attendees are in New York, Sydney, and Singapore. The constraints block says no meetings before 8 am or after 7 pm in the exec’s local zone, no exceptions for global meetings. Sydney is 11 hours ahead of London in summer. Singapore is 7. New York is 5 behind.
Run the prompt and the model will tell you: there is no slot in the next two weeks where all four attendees are inside their normal working hours simultaneously. The honest answers are (a) Sydney joins at 6 pm their time, (b) New York joins at 6 am their time, or (c) the meeting becomes a video update where someone is asynchronous. The model presenting these three trade-offs explicitly is what lets you walk into your exec’s office with options instead of an apology.
A weaker prompt would have invented a slot at 5 pm London, which is 4 am in Sydney, and you would have found out only when the Sydney attendee opened the invite.
The dependency check that prevents reschedule disasters
Before you confirm any meeting, especially one that took effort to coordinate, ask the model one more thing:
“What other meetings or commitments on my exec’s calendar this week depend on this one happening at this time? List anything that would also need to move if this meeting moved.”
This is the question that stops you from booking a Wednesday client meeting that, if it slips, also blows up the Thursday prep call and the Friday board update because they were stacked in sequence. The model can see the calendar in a way you can’t hold in your head. Use it.
The failure mode to watch for
The biggest one: trusting AI-suggested “free” slots without checking what’s classified as private or tentative. Many calendars have personal blocks marked private, and a model with read access to the visible-busy layer will treat them as “tentative” and offer them up. Always include in your prompt: “Treat any block marked private, tentative, or out-of-office as fully booked. Don’t offer those slots.”
The second one: letting an automated scheduling tool send invites without you reviewing the time. Tools like Reclaim or Clockwise can send invites directly. Turn that off for any external attendee. Internal team coordination, fine. External, your reputation is on the line, and a model that gets a time zone wrong by accident will do so on your exec’s behalf, in their name.
What calendar work shifts to
The hour you used to spend booking a single complex meeting becomes ten minutes. The time you save isn’t lost, it goes into the upstream work scheduling tools can’t do: deciding which meetings should happen at all, which can be a 15-minute video update, and which need an hour. That decision still has to come from someone who knows the work, the relationships, and the exec’s actual priorities. AI books the slot. You decide whether the slot should exist.