Prompts for professional emails: follow-ups and delicate refusals

"I hope this email finds you well." When you read that sentence in an email, you can guess where it came from, and your recipient will guess too. AI-written text gives itself away through its uniform politeness and the fact that it could be addressed to anyone.

The problem comes from the request. Writing a difficult email is mostly about deciding what you want to achieve and what you're willing to say. If you leave those two decisions to the AI, it will pick the smoothest reply. Tell it what they are, and all that's left for it is the wording, where it's effective.

A word on privacy: an email thread contains names, amounts, and sometimes internal information. Remove what identifies people and what must not leave your company, and check which tool your organization allows. Our article on what not to paste into a chatbot helps you sort it out.

The skeleton of an email prompt that works

Compare these two requests, for a site manager whose supplier is delivering late.

Write an email to my supplier telling them this delay is not acceptable.

Write an email to [SUPPLIER], whom I've worked with for [DURATION] and want to keep.
Facts: the window frames due on [DATE] haven't arrived, it's the second delay this quarter, and the installation crew is sitting idle on [DATE].
What I want to get: a firm delivery date before [TIME, DAY].
What I'm willing to say: that I'll have to rethink how I split my orders if this continues.
Tone: firm, no sarcasm.
Length: 90 words maximum.
Start directly with the problem, with no introductory pleasantries. End with the specific request and its deadline.

In the second version, you've set the goal, the limit, and the balance of power. The AI no longer has to guess whether you want to get angry or preserve the relationship. You can reuse this structure for most of your messages, and adjust it with the advice in our guide to output format.

Replying to an unhappy customer without grovelling

Your first reply to an aggressive message shouldn't go out in the heat of the moment. First have the AI sort out what calls for a response.

Here is a customer's message, personal data removed:
[MESSAGE]

What actually happened: [FACTS]
What I can offer: [ACTION]
What I can't accept: [LIMIT]

1. In their message, distinguish the facts, the requests, and the emotion.
2. Write a reply of 110 words maximum that acknowledges the concrete problem, takes responsibility for what falls to us, with no apology for the rest, then states what happens next and a timeframe.
3. Reread your reply and delete any sentence that sounds like a template.

Formal tone. No "we understand your frustration." Don't admit any fault I haven't mentioned.

The last instruction matters if the matter could turn into a dispute. In that case, have the right person review your reply before it goes out.

Following up without coming across as a nag

An effective follow-up restores the context and reduces the person's effort: they can answer in ten seconds. Take an executive assistant who has been waiting eight days for an approval.

I'm following up with [PERSON'S ROLE], who hasn't replied for [DURATION].
Subject: [SUBJECT]
What I'm waiting for: [SPECIFIC ANSWER]
Consequence if nothing arrives before [DATE]: [REAL CONSEQUENCE]

Write a follow-up of 60 words maximum that can be read on its own without having to find the first message. No "just checking in," no "sorry to bother you." Phrase the request so they can answer with yes, no, or a single choice. Suggest an explicit subject line, 8 words maximum.

The subject line is underrated. "Follow-up" tells you nothing, whereas "Dupont quote approval: reply before Thursday" lets someone decide without opening the message. If you often send this kind of follow-up to clients as a freelancer, also look at the templates in our article for freelancers, which covers invoices.

Refusing a request while leaving the door open

Saying no by email often pushes people to write three paragraphs of justification. A good reply refuses clearly, and a single reason is enough.

I'm being asked: [REQUEST, BY WHOM]
I'm declining, for this reason: [REAL REASON, EVEN IF BLUNT]
What I can offer instead: [ALTERNATIVE, OR "NOTHING"]
Relationship with the person: [BOSS, COLLEAGUE, PARTNER]

Write a refusal of 70 words maximum. The "no" must appear in the first two sentences, without "unfortunately." One reason only, without excessive apologies. If an alternative exists, present it last, without making it a promise. The tone should be that of someone who respects the person they're writing to and their own time.

An association president turning down a partnership proposal can thus stay clear, without giving the impression that a follow-up will be enough to change her mind.

Announcing bad news, and summarizing an endless thread

For bad news, the usual mistake is to bury it in the middle of the second paragraph. Ask the AI for the reverse order.

I have to announce to [RECIPIENTS]: [NEWS].
Reason I can give: [REASON]
What changes for them in concrete terms: [IMPACT]
What's planned next: [NEXT STEPS]

Write an email of 120 words maximum. State the news in the first two lines, without presenting it as an opportunity. Then explain what changes, then the next step, and who to contact. Add no promise or information absent from these elements.

For a 40-message thread where everyone is copied, the AI saves you time, provided you demand decisions rather than a narrative.

Here is an email thread, from oldest to most recent, anonymized:
[THREAD]

Write a summary for someone who hasn't followed any of it:
- the current situation in 2 sentences
- the decisions made, and who made them
- the questions still open
- the actions expected, with the person responsible and the date if they're mentioned

If two people contradict each other, point it out without settling it. If a piece of information doesn't appear in the thread, don't mention it.

A summary can distort a nuance or attribute a decision to the wrong person: before forwarding it to your management, reread the important passages. If your generated emails still don't sound like you, feed the AI a few of your messages, as explained in getting the AI to write in your style.

Open the message you've been unable to answer for days, write in one line what you want to get and what you refuse to say, then run the matching prompt.