Email Reply Drafter in Three Variants

When to use this prompt

When you have an inbound email that requires a thoughtful reply and you want options before committing to one. Or when the email is sensitive enough that you don’t want to draft cold and prefer to choose between three pre-thought versions. Or when you simply want to process inbox 5x faster without your replies sounding like the same generic template.

This isn’t a prompt for replying to “thanks!” emails. It’s for the ones where the choice between acknowledgment, action, and escalation actually matters.

The prompt

<role>Executive communications specialist who drafts email replies that match the sender's seniority, the situation's urgency, and your voice.</role>

<task>Draft three reply variants for the inbound email below: an acknowledge-only version, an action-oriented version, and an escalating version. Each must match the supplied voice and stay within the supplied length.</task>

<inputs>
<inbound_email>
[PASTE THE FULL INBOUND EMAIL, INCLUDING SUBJECT, SENDER NAME/ROLE, AND BODY]
</inbound_email>
<your_role>[YOUR ROLE AND SENIORITY, e.g., "Director of Marketing reporting to a CMO"]</your_role>
<voice>
[2-3 sentences on how you write. E.g., "Direct, low-fluff, signs off with first name only. Never uses 'circling back' or 'just following up'. Comfortable being briefly disagreeable."]
</voice>
<context>[OPTIONAL: 1-3 sentences on the relationship and history with this sender that the reply needs to reflect]</context>
<length>[SHORT: under 60 words / MEDIUM: 60-120 words / LONG: 120-200 words. Defaults to MEDIUM if unspecified.]</length>
</inputs>

<instructions>
1. Produce three reply variants:
   - **Acknowledge only**: confirm receipt, no commitment. Used when you need to respond now but can't act yet, or when the sender's request needs more thought before answering.
   - **Action-oriented**: take a clear action or commit to one with a specific timeline. Used when you can move now and want the thread closed.
   - **Escalating**: respectfully push back, raise a concern, or surface a missing piece. Used when the email's premise is wrong, incomplete, or assumes something you don't agree with.
2. Each variant must:
   - Match <voice> in tone, sentence rhythm, and sign-off style.
   - Stay within <length>.
   - Not include "circling back," "just following up," "per my last email," "as discussed," or other corporate filler that erodes signal-to-noise.
   - Address the most important thing in the inbound email first. Do not bury the answer.
3. Each variant must include:
   - A subject line if the inbound subject would be improved (otherwise note "Reply: [original subject]").
   - The full body, ready to paste.
   - One-sentence note on when to pick this variant over the other two.
4. Do not invent commitments, names, or details not implied by <inbound_email> or <context>. If the situation calls for specifics you don't have, mark [INSERT: brief description] in brackets so the user fills in.
</instructions>

<output_format>
### Variant 1: Acknowledge only

**Subject:** [...]

[Body, ready to paste]

**When to pick this:** [one sentence]

---

### Variant 2: Action-oriented

**Subject:** [...]

[Body]

**When to pick this:** [one sentence]

---

### Variant 3: Escalating

**Subject:** [...]

[Body]

**When to pick this:** [one sentence]

---

**Word counts:** V1: XXX | V2: XXX | V3: XXX
**Voice match check:** [Pass / Fail with one-line note]
**[INSERT:] markers:** [List any in any variant]
</output_format>

How it works

The three-variant structure forces a real choice. Most reply-drafting prompts produce one reply, which the user accepts as “the answer” without considering whether a different posture would serve the situation better. Three variants frame the choice as posture, not wording.

The “acknowledge / action / escalate” categorization is the entire point. These are the three real moves in any meaningful email reply, and which one is right depends on context the model can’t see. By producing all three, the user picks the right posture and pastes the right draft in the same step.

The forbidden-phrases list catches the most-used corporate filler. “Circling back” and “just following up” signal that the writer is stalling; eliminating them at the prompt level is faster than editing them out per reply.

The voice-match check is a forcing function. If the model produces three variants that all sound like generic AI prose, the check at the end will surface that. The user can re-run with a tighter input.

The [INSERT:] markers handle the case where the reply needs information you have but didn’t supply (a date, a colleague’s name, a specific number). Better to mark the gap explicitly than to fabricate.

Example output

Variant 1: Acknowledge only

Subject: Reply: Re: Q3 budget review

Got it, Sarah. Let me look at the numbers and come back Friday with a clear take. Want to make sure I’m not reacting before reading the spreadsheet properly.

Arthur

When to pick this: When you need a day to think and the sender accepts that.

---

Variant 2: Action-oriented

Subject: Reply: Re: Q3 budget review — moving the SEO line, holding events

Looked at the spreadsheet. Two changes I’d push:

1. Move 30% of the SEO budget to GEO measurement infrastructure. Current SEO line is overfunded for what it’s producing.
2. Hold events at current spend. The pipeline contribution is real and we’d lose ground reducing.

Happy to walk through on Friday’s call if useful.

Arthur

When to pick this: When you’ve done the work and can commit; closes the thread fastest.

---

Variant 3: Escalating

Subject: Reply: Re: Q3 budget review — flagging an assumption

Sarah, before we cut the SEO line, I want to flag the assumption underneath: we’re treating SEO and GEO as separate budgets, and they’re not. Most of what’s working in GEO is the SEO investment we made 18 months ago.

Cutting SEO now and increasing GEO doesn’t shift the spend; it slows the engine that’s producing the GEO results we like.

Worth a 15-minute call before we lock the numbers.

Arthur

When to pick this: When the email’s premise is wrong and saying so now saves a worse conversation later.

---

Word counts: V1: 32 | V2: 71 | V3: 95 Voice match check: Pass — direct, low-fluff, first-name sign-off, no corporate filler.

Variations

• Customer-facing version: Add a “warmer / cooler” tone dial as input. For replies to customers, the variants need different temperature.
• Negotiation mode: Replace the three categories with “concede / counter / hold” for replies to negotiations specifically.
• Apology mode: When the inbound is a complaint or escalation, replace the three with “apologize-and-fix / explain-and-fix / push-back-respectfully.”