The AI Prompt for the Email You Keep Avoiding

There is a specific kind of email that does not get written. It is not the one you forgot — it is the one you have opened, closed, and put off a dozen times. The late reply you owe to someone important. The difficult conversation with a client who deserves honesty. The ask you are terrified will get a no. The apology you have rewritten three times and deleted. It sits in your drafts folder as a low-grade guilt, and the longer it sits, the harder it gets to start.

An AI prompt can break that loop, but not by writing a perfect email for you. That is the wrong goal. What a good prompt does for the email you are dreading is produce a draft you can react to — imperfect, maybe a little blunter than you want, maybe softer — and reacting to a draft is a fraction of the cognitive cost of starting from nothing. This guide gives you a prompt sequence built for the avoided message, with the steps, a platform comparison, and what to do when the first attempt misses.

The short answer: ask for three versions at different directness levels#

Paste the following into any AI assistant, swap the brackets for your real situation, and send it as one message. Role: You are helping me write an email I have been putting off. Write in my voice: plain, honest, and professional. Situation: [one sentence on the relationship — who this is to and why it matters]. What I need to say: [the core message, as bluntly as you can state it to yourself]. Why I have been avoiding it: [what makes this hard — fear of the reaction, not knowing how direct to be, guilt about the timing]. Task: Write three versions of this email — one gentle and indirect, one direct and clear, one firm and unapologetic. Keep each under 150 words. Do not invent facts or promises I have not provided.

The three versions are not because you need three emails. They are a temperature scale. One will feel obviously too soft, one obviously too blunt, and that contrast makes it immediately clear which register you actually want. You then edit the version closest to right. That is the whole method.

Before you start: name what makes the email hard#

The most important step before reaching for a prompt is to say — out loud or in writing — what specifically makes this email difficult. Not that it is awkward or complicated, but the actual thing. You owe someone an answer you do not have. You need to decline something and you are worried about the relationship. You missed a commitment and feel guilty about the gap. You need to raise something the other person will not want to hear.

The reason this matters is mechanical: a model given vague context produces a vague draft. Write an email to my client about a project issue gives the model almost nothing. Write an email to my client telling them the delivery will be three weeks late, I have not told them yet, and I am worried they will pull the contract is a real situation the model can write toward. Specificity is what separates a draft you can send from one you close and delete.

If you cannot name what makes the email hard, write it to yourself first — a rough private note with everything you actually think about the situation. You will not use most of it in the email, but writing it clarifies what the message needs to say and what it should leave out. The AI draft will be better for it.

The step-by-step prompt sequence#

1. 1

   State the situation in one sentence

   Before opening any AI tool, write a single sentence that names the relationship, the message, and the timing: I need to tell my client Sarah that the report will be two weeks late and I have not told her yet. This sentence goes directly into the prompt as context. If you cannot write it in one sentence, you are not yet clear enough on what the email needs to say — spend a minute here before moving on.

2. 2

   Paste the prompt and ask for three versions

   Use the template from the short answer above and fill in all four fields with real information. The Role line sets voice and tone. The Situation field names the person and why the relationship matters. The What I need to say field should be the bluntest version of the message you can write — even if it is too blunt for the final email, it orients the model toward the real content. The Why I have been avoiding it field is the unusual one: telling the model what makes the email hard lets it calibrate how much care and softening is actually needed.

3. 3

   Pick the version you can react to — not the best one

   Read all three versions. You are not looking for the best one. You are looking for the version closest to what you actually want to say, even if it is off in tone or detail. That version is your starting point. Delete or change the sentences that feel wrong. Editing a draft that is almost right is a fraction of the cognitive cost of starting from a blank page, and that contrast is the point of the three-version structure.

4. 4

   Tighten with one focused follow-up prompt

   Once you have a version you are mostly happy with, paste it back into the AI and ask one specific question: Make this 30 words shorter without losing the key asks. Or: The second paragraph is too apologetic — rewrite it to be firm but kind. Or: Add a concrete next step in the last line. One targeted change at a time produces a better result than make this better, which gives the model no direction to work toward.

How the prompt works across different AI platforms#

The prompt above works in any current AI assistant. The mechanics vary slightly by platform, and knowing the differences saves setup friction when you are already reluctant to engage with the email.

What to do when the AI draft still does not work#

Sometimes the first round of three versions misses in a specific, identifiable way — too formal, too flat, or too generic. Right structure, wrong sound. That is almost always a context problem. The model filled in what you did not give it, and guesses produce generic output.

• The draft is too formal: Add to the role line — conversational, short sentences, no corporate language — then ask for a rewrite. Describing the mismatch is faster than re-prompting from scratch.
• It sounds too apologetic: Add to the task — Do not over-apologise. One clean acknowledgment of the difficulty, then the message. No grovelling. That one instruction removes the most common tonal failure mode for avoided emails.
• It does not sound like me: Paste two or three sentences from a recent email you wrote and liked, and say Match this voice exactly. Showing the model your writing is more reliable than describing it.
• It invented a detail or made a promise I did not give it: Add to the task — Do not invent any facts, offers, promises, or timelines I have not stated. If a detail is missing, leave a placeholder in brackets. Use this instruction on every draft of a sensitive email.
• It is still too long: Ask specifically — Cut this to under [word count]. Keep the core message and the concrete next step. Remove everything else. A word target gives the model a measurable goal.

A faster way: draft the difficult email inside your inbox#

The prompt sequence above works, and a single hard email is worth the tab-switching. If you write emails like this regularly — the delayed client update, the hard conversation with a vendor, the ask you have been putting off — the switching cost compounds: open a chat tab, paste in the thread, re-describe your voice, copy the draft back, repeat.

AI Emaily drafts the avoided email inside your actual mailbox. The thread is already there. The draft is written in your voice from your user-set Context brain and client profiles — you are not re-describing your communication style each session. Nothing leaves your outbox without your approval; Copilot mode means a draft that missed the tone never becomes a sent mistake. We build AI Emaily. Start a 7-day free trial at aiemaily.com — see what is included at aiemaily.com/pricing.