How to start the day with decisions instead of email
By the end of this you will have one scheduled prompt that reads your mailbox each morning and hands back a numbered list of decisions you owe. Setup is one sitting at your desk. The work that matters comes after, in the week you spend reading that list and noting what it left out.
What can ChatGPT actually do for me?
A prompt can read the last 100 emails in your mailbox, filter out the noise, and rank what is left by whether you owe a reply, a decision, or an approval. The output is a capped numbered list, delivered at 08:00 every day.
The useful word is decision. A summary of your inbox is worth almost nothing to a founder. You already know what came in. What you do not have at 08:00 is a ranked list of the things that stop moving until you say something.
Decide what counts as a decision
Write your three buckets down on paper first. Reply means someone is waiting on words from you. Decision means someone is waiting on a choice only you can make, a price point, a hire, a supplier. Approval means the work is done and needs your name on it.
Then write the exclusions, because that is where the value sits. Newsletters out. Invoices that accounts already handles, out. Anything where a partner has already answered in the thread, out. Cap the list, and say so in the brief. An uncapped list is another inbox.
Done looks like a short note in plain words you could hand to someone else. If you cannot tell a reply from a decision in your own words, the output will not either.
Connect the mailbox and set the run time
Connect Gmail in ChatGPT settings once, then confirm your own account has the connection before you go further. Availability differs by plan and by country, so check the account you will actually use rather than trusting this paragraph.
Then make it recurring. ChatGPT runs scheduled tasks that repeat daily at a set time, and all of them sit on one page you can pause or edit. Set yours for the time you actually sit down, not the time you wish you did. If you start at 08:00, schedule for 08:00.
Done looks like a task on that page with your brief in it and tomorrow's run time showing.
How do I get better at reading the output?
By checking it against the mailbox, by hand, on three mornings. Not every morning, and not never. Open the list, then open the mailbox, and look only for the gap between them.
Three checks catch most of it. First, what did the list miss that you would have caught, and why, was it a sender it ranked low or a subject line that read as routine. Second, what is on the list that should not be, and what phrase in your brief let it through. Third, is anything ranked as a decision when it is really a reply. That last one is the common failure, and it matters, because a list of nine replies dressed as decisions teaches you to stop reading the list.
When the output is wrong, you fix the brief, not the output. Add the sender to the exclusions. Name the thread type. Say what a decision is in one more sentence. The brief is the thing you are building here, and it gets better only where you tell it what it got wrong.
Run it for a week before you change a word
Leave the brief alone for five working days. The temptation after one bad morning is to rewrite everything, and then you have no idea which change helped. One week of the same brief gives you a pattern. Then change one thing.
What you end up with is small and yours. A prompt, a connection, a run time, and a short record of where the output was weak. That record is worth more than the prompt. It is also what separates somebody who uses AI from somebody who has integrated it into a morning that already existed.
A mailbox brief like this one sits in the prompt library at A.D.A.P.T. AI Coaching.
What else comes up in the first week?
What if the brief misses something urgent?
Assume it will, in the first week, and keep opening the mailbox as well. The brief earns trust by being checked, not by being believed. Run both for five days, note each miss, and fix the brief. After that you will know which senders it cannot be trusted to rank.
Should I let it draft the replies too?
Not in week one. Drafting is a separate brief with separate checks, and mixing the two means you cannot tell which one is failing. Get the ranking right first. Once the list is reliable for a month, add drafting for one low-stakes category and read every draft before it goes.
Does this work with a shared or team mailbox?
It works, but the brief needs one more rule, which is whose decisions you want. A shared mailbox without that rule returns everyone's pending items ranked together. Name the addresses that count as you, and exclude threads where a partner has already replied, or the list will double.