Draft a first reply that ends the email thread, not starts one

Count how many of your replies get a "thanks, but what about...?" back. Often it's one of three things. The answer was in paragraph three. You asked for one piece of information, got it, then asked for another. Or you promised something nobody could deliver.

An order the AI has to follow fixes all three.

Draft a reply to the customer message below. Use this order:
1. The answer or outcome, in the first sentence. If we can't give one
   yet, say that and say when we will.
2. What happens next and who does it, with a date or timeframe.
3. Anything we need from them, all in one list, so they only reply once.
4. One line on where to go if it's urgent.
Rules: no "we apologise for any inconvenience". If something went wrong,
say what in one plain sentence. Don't promise anything that isn't in the
facts below. Match their tone: short if they were short.

What we know: [FACTS, POLICY, WHAT WE CAN DO]
Customer message: [PASTE]

The facts block keeps it honest. Without it the model makes up a helpful-sounding promise, a refund timescale or a callback, and the customer holds you to it.

Read the draft once as the customer. Can they tell what happens next without reading it twice? If the answer's "it depends", say what it depends on.

Keep names and account details out of tools your organisation hasn't approved for customer data. A summary of the problem works just as well as the original message.