How to Use ChatGPT as an Executive Assistant (From Someone Who Was One)

Most people who try to use ChatGPT as an executive assistant give up within a week. They ask it to plan their day, get back something about "blocking focus time for high-impact work", and conclude it isn't much use. Fair enough. A new assistant who knows nothing about you isn't much use either.

I spent years as an executive assistant and project management analyst running CEO offices. The job was never really diary management. It was holding the whole operation in view: knowing what mattered this week, protecting the time that work needed, chasing the things other people owed, and making sure my boss never walked into a meeting cold. None of that is clever. All of it depends on context.

ChatGPT meets you for the first time every time you open it, so you have to give it what a good EA builds up over months. Once you do, it stops sounding like a productivity blog and starts sounding like someone who works for you.

01 / Write the brief a new EA would need

When a new assistant starts, the useful handover isn't the job description. It's the stuff nobody writes down: who your boss actually listens to, which meetings are political, which decisions are yours and which need someone else's yes, how you like bad news delivered, and what you always let slide.

Put that into one document. Call it your Chief of Staff Brief. Three sections are enough to start:

• Who I am. What you're genuinely accountable for (rarely the same as your title), what you're measured on, what's yours to decide, and what you're quietly bad at.

• My working world. Who you report to and what they care about, who reports to you, the external people who can eat a day of your time, your recurring commitments, and where your work lives.

• How I communicate. Your voice, how it shifts depending on who's reading, and the things you never do.

Don't write it from a blank page, because you'll produce something that reads like a LinkedIn profile. Get interviewed instead. Open a chat, switch to voice, and paste this:

I'm building a permanent context document so you can act as my chief of staff. Interview me one question at a time. Push back when I describe my job in job-description language rather than what actually happens. Cover what I'm accountable for, the people around me (inside and outside work), how I communicate, and where my calendar, email, tasks and documents live. Ask at least twenty questions before you write anything.

Answer out loud. Typing makes people careful, and the honest version (the colleague who never replies, the meeting you resent) is what makes the brief useful. Then upload what already exists: an org chart, your objectives, a screenshot of a normal week and eight to ten messages you've actually sent.

One more thing. Include your life outside work. The school pick-up and the appointment you've moved twice belong in there, because an assistant that only knows your work diary will confidently tell you to protect the exact hour you're collecting someone.

02 / Give it a permanent home in a Project

A brief that lives in one chat is gone the moment that chat ends. In ChatGPT, create a Project for your working life and paste the brief into the project instructions. Those instructions apply to every chat inside the project, so you never explain yourself twice. Your uploads (org chart, objectives, message examples) go into the project files, where every later conversation can reach them.

ChatGPT also has memory, which picks up details across conversations. Treat it as a bonus, not the mechanism. The brief is the mechanism, because you wrote it, you can read it and you can correct it.

Add a few standing rules while you're there:

Lead with decisions, trade-offs and next actions. Separate what you read from what you inferred, and name what's missing rather than filling the gap. Never invent dates or owners. Draft only: never send, spend or change my calendar without my approval.

That middle line matters more than it looks. An inference stated confidently is indistinguishable from a fact until the moment you repeat it to your boss.

Projects, connected apps and scheduling vary by plan and region, and they change often. A paid plan gives you the full set; check what yours includes before you build around a feature. If you already use Claude, the same approach works there: create a Project, put the brief in its instructions or knowledge, and everything below runs the same way.

03 / Start every day with a brief, not a summary

The most visible thing an EA does is the morning handover. Not a read-out of the calendar, which your boss can already see, but: here's the one thing that matters today, here's what's changed, here's what's about to go wrong, here's what you can ignore.

A brief that lists everything is just your calendar with extra steps.

Feed it three things: your calendar for the next seven days (screenshot it, or connect it), the two or three things you'd be judged on if the month ended tomorrow, and what you're waiting on from other people. Then use a structure like this:

Using my Chief of Staff Brief, write my brief for today. THE ONE THING: the single item that matters most, and why. CHANGED: what's moved since yesterday. AT RISK: what's quietly going wrong, including anyone who's gone silent. DROP: what I can skip, decline or stop chasing. No encouragement, no summarising my calendar back to me. Under 250 words. Assume I'm reading this standing up.

The first attempt will probably be a tidy list of your week. Accurate and useless. The fix is telling it what you consider important: what makes a day good, who can't be left waiting, what "at risk" means in your world. Those answers are your attention rules, and they belong in the brief so you only say them once.

ChatGPT can connect to services such as Gmail and Google Calendar through plugins (Settings, then Plugins, available in the UK), so once your calendar is connected you can just say "brief me for today" and it reads the week itself, including what moved since you last looked. If your organisation won't allow that, a screenshot works almost as well.

04 / Prepare for meetings the way an EA would

Meetings are where most of your week goes and where most follow-through dies. I never sent a boss into a meeting with just the agenda. They went in knowing what it was really about, what they'd promised last time and hadn't delivered, and the awkward question they'd rather not be asked. ChatGPT can do that prep in a couple of minutes if you give it the invite and any pre-read:

Prepare me for this meeting. WHAT IT'S REALLY ABOUT, in one line. WHAT I OWE FROM LAST TIME. WHAT I WANT OUT OF IT, stated as a decision. WHO'S IN THE ROOM and what each needs. THE AWKWARD BIT, and my honest answer. Under 200 words.

Afterwards, record a forty-second voice note on the way back to your desk: what was decided, what you're on the hook for, what you think is really going on. Drop it in with your scrappy notes and ask for decisions, actions with owners, unresolved points and anything you're now waiting on. Add one instruction: where it's unclear who owns something, say so rather than guessing. An invented owner is worse than an admitted gap.

Then save that output back into the project, so next time's prep can tell you what you promised a fortnight ago.

05 / Keep the list nobody else keeps

Everyone keeps some version of what they owe. Almost nobody keeps what they're owed: the supplier who said they'd confirm by Friday, the approval that has been sitting with someone for nine days. Chasing those was half my job, and it's where the value quietly leaks out of most people's weeks.

You don't need to maintain it by hand. Dump in a week of debris (meeting notes, screenshots of threads, a voice note from the car) and ask it to sort it:

Pull out MINE (what I owe, to whom, by when), THEIRS (what others owe me), WAITING ON ME (anything I'm blocking), and AMBIGUOUS (where you can't tell who owns it; don't guess). Ignore discussion that didn't produce a commitment.

Keep the waiting-on items in a running document with a date against each one. The actions unit in AI Powered Chief of Staff goes further, with escalation ladders and a sweep of your sent mail for promises you forgot you made. The date is the whole point: it turns a vague feeling that you're owed something into "this has been sitting for eleven days". Once a week, ask what's overdue by your own chase rule and have it draft each chase in your voice, firmer the longer it's been. The drafts go into your mail; you send them.

06 / Automate two things, not seven

Everything so far still depends on you remembering to ask. ChatGPT's scheduled tasks can run a saved prompt at a set time, so the morning brief is waiting at 7.30 rather than sitting on your to-do list.

Two cautions. Scheduled tasks created inside a project can't currently read that project's files, so anything relying on an uploaded list stays manual for now. And a scheduled output you don't read trains you to ignore the thing you built. Start with the daily brief and a Friday review. Add a third only after a month of actually reading the first two.

Before anything runs on its own, write down what it can produce without asking, what it must show you before sending, and what it must never do. Most people draw that last line at anything going to a client, anything that commits money and anything going to their boss. That's the same discretion every EA job advert lists first, and it applies to what you feed it too: check your employer's AI policy before connecting work email, and keep confidential HR, board and client material out.

How to use ChatGPT as an executive assistant, starting this week

The test of a good assistant was always simple: does your boss walk into the day knowing something they wouldn't otherwise have known? Ask ChatGPT the same question before and after the brief and you'll see it. The tool doesn't get cleverer. It just knows who's asking.

Start with the brief and the daily briefing, and use them for a week before adding anything else.