Email gets 376 billion sends per day now. Most go unread. The ones that work share something in common: they feel written for one person, not scraped from a template library and blasted to a list.

AI can help you write emails that don’t feel like AI wrote them, but the gap between generic output and something useful is bigger than most people realize. The prompt is everything. Feed the model vague instructions and you get filler text that reads like corporate spam, the kind that makes recipients reach for the unsubscribe link before finishing the first sentence.

These templates are structured to close that gap. They force specificity, context, and constraints into every request because that’s what produces output worth using.

Why Most AI-Written Emails Fail

There’s a cartoon making the rounds that captures the absurdity perfectly. One panel shows AI turning a single bullet point into a long email someone can pretend they wrote. The other shows AI summarizing that long email back to a single bullet point someone can pretend they read.

As one Hacker News commenter put it: “sender: writes summarized prompt / llm: emits excessively lengthy and polite prose / smtp: transports lengthy prose / llm: summarizes lengthy prose to bullet points / recipient: reads summary / what a wonderful waste of energy.”

The problem isn’t the technology. It’s how people use it. Vague prompts produce vague emails. No context about the reader, no specificity about the goal, no constraints on length or tone. The model fills the void with filler: pleasantries, redundant phrases, the linguistic equivalent of empty calories.

According to research from Mailchimp, email open rates hover around 21% on average. That means roughly 4 out of 5 emails go unread. Your subject line has maybe two seconds to earn a click. The body has maybe ten seconds to earn a read. Every word matters.

Subject Lines That Actually Get Opened

Subject lines are where most AI email attempts fall apart. The model defaults to safe, boring options because that’s what statistical patterns predict. “Quick Question” or “Following Up” or “Important Information Inside.” Generic. Forgettable. Deletable.

The fix is forcing variety and constraints into the prompt. Don’t ask for subject lines. Ask for specific types of subject lines with specific character limits and specific psychological hooks.

Generate 10 email subject lines for [describe your email content].

Context:
- Email type: [newsletter, promotion, announcement, transactional]
- Main message: [one sentence describing the key point]
- Who's receiving this: [describe your audience specifically]
- What action you want: [opens, clicks, replies, purchases]
- Voice: [urgent, curious, friendly, professional, playful]

Constraints:
- 3 benefit-focused (what they get)
- 3 curiosity-driven (creates intrigue without clickbait)
- 2 question format (engages directly)
- 2 direct/clear (says exactly what it is)

All subject lines under 50 characters. No spam trigger words. No false urgency. No ALL CAPS.

Why this works: You’re not asking for “good subject lines.” You’re defining what good means for this specific email. The constraints force variety. The character limit forces brevity. The banned elements prevent the model from reaching for cheap tricks.

One Hacker News user captured why this matters: “tl;dr: AI is looking to convey words. A good author is looking to efficiently convey information.” Prompts that prioritize information over word count produce better output.

A/B Test Variations

Once you have a winner, test it. Here’s how to generate structured test variants:

I need A/B test subject lines for this email: [describe content]

Audience: [who they are]
Current best performer: [your current subject line]
Testing for: [opens, clicks, specific metric]

Generate 4 pairs for testing:

Pair 1 - Benefit vs. Curiosity
Pair 2 - Short (under 30 chars) vs. Descriptive (40-50 chars)
Pair 3 - Question vs. Statement
Pair 4 - With personalization token vs. Without

For each pair, explain what the test will reveal.

The personalization question is real. Research compiled by various industry sources shows personalized subject lines can boost open rates by 26%. But personalization without relevance is worse than no personalization at all. A first name in a subject line that’s clearly mass-produced just highlights the automation.

Newsletter Templates That Don’t Read Like Form Letters

Newsletters are relationship builders. They work when readers feel like a person wrote them. They fail when they read like a content dump with a logo attached.

The challenge with AI-generated newsletters is voice. Most outputs sound like corporate communications departments: polished, professional, utterly forgettable. As one Hacker News commenter noted: “That distinct feeling when reading AI is as if someone who wrote it was compelled to write more words.”

Here’s a template that fights that tendency:

Write a newsletter for [company/publication name].

Voice context:
- Our newsletters sound like: [describe how you actually write]
- We never say things like: [list phrases that aren't you]
- Example of our voice: [paste a sentence or two from a previous email]

This week's content:
- Main story: [what it is and why readers care]
- Secondary items: [list 2-3 other things]
- Primary CTA: [what you most want them to do]
- Any timely hooks: [current events, seasonal, industry news]

Structure:
- Subject line options (5 different angles)
- Opening hook (2-3 sentences that pull readers in)
- Main section (150-200 words maximum)
- Secondary items (50-75 words each)
- Closing CTA
- Optional P.S. line

Total word count under [your target]. Make it scannable. Short paragraphs. Clear breaks between sections.

The voice context section is critical. Without examples of how you actually write, the model will default to how newsletters generally sound. That’s not your voice. That’s everyone’s voice. And everyone’s voice is nobody’s voice.

Content Repurposing

Got a blog post or social content that needs to become newsletter material? Transformation prompts work better than generation prompts because you’re giving the model something concrete to work with.

Turn this content into a newsletter section:

Original:
[paste your blog post excerpt, social content, or whatever you're repurposing]

Newsletter context:
- Audience: [who reads your newsletter]
- Section this becomes: [main feature, quick tip, etc.]
- Word limit: [how long this section should be]
- CTA that should follow: [what action you want]

Rewrite for email:
- Short paragraphs (email readers scan)
- Clear takeaway (one main point)
- Transition to the CTA (natural bridge to the action)

Keep the core value. Change the format.

Drip Sequences That Build Instead of Badger

Welcome sequences are where most email automation goes wrong. Companies set up a series that makes sense from their perspective: introduce the product, show features, ask for the sale. But that’s not how relationships work.

According to email marketing research, properly segmented email campaigns can generate up to 760% more revenue than generic blasts. Segmentation matters. But so does pacing. A sequence that pushes for conversion in every email doesn’t build trust. It erodes it.

Create a [number]-email welcome sequence for new [subscribers/customers/trial users].

Sequence context:
- What triggered signup: [lead magnet, purchase, free trial, etc.]
- What they actually want: [why they signed up, in their words]
- What you want eventually: [purchase, engagement, referral]
- Voice: [describe your tone]

For each email, provide:
- Timing relative to signup
- Subject line
- Goal of this specific email (not the whole sequence)
- Main message in 1-2 sentences
- Content outline
- CTA for this email

Suggested structure:
- Email 1: Deliver what they signed up for + set expectations
- Email 2: Quick win or immediate value (no selling)
- Email 3: Story or social proof (let them see others like them)
- Email 4: Address the questions they haven't asked yet
- Email 5: Primary CTA with clear reason to act now

Each email should stand alone. They should also build toward something.

Re-engagement Sequences

Dead subscribers cost money. List hygiene matters. But the re-engagement sequence most companies run is basically begging: “We miss you! Please come back!”

That doesn’t work. Here’s what to send instead:

Create a re-engagement sequence for subscribers who haven't opened an email in [timeframe].

Context:
- What they originally signed up for: [if known]
- What they've been ignoring: [how many emails, what kind]
- Last thing they engaged with: [if you know]

Goals:
- Primary: Get them active again
- Secondary: Get inactive people to unsubscribe (clean the list)

3-email sequence:

Email 1 (check-in):
- Tone: curious, not desperate
- Ask if they still want to hear from you
- Remind them why they signed up (specific value)

Email 2 (value):
- Offer something genuinely valuable
- Show what they've missed that might interest them
- Clear single action to re-engage

Email 3 (breakup):
- Final notice before removal
- Make staying simple
- Make leaving dignified

Space these [X days] apart. Each email should feel different from the last.

The breakup email matters more than most people think. Someone who explicitly leaves is better than someone who silently ignores everything. Dead weight on a list hurts deliverability, skews metrics, and costs money.

Campaign Emails That Convert Without Screaming

Product launches, promotions, event invitations. These are the emails where companies forget everything they know about good communication and start shouting.

HUGE SALE! ACT NOW! LIMITED TIME! DON’T MISS OUT!

Nobody reads that. As one Hacker News commenter observed: “When I see something written by AI I don’t read it. Its a waste of time.” The same applies to anything that reads like a template, which includes most promotional email.

Write a product launch email for [product name].

Product:
- What it is: [one sentence]
- Key features: [list 3-5]
- Main problem it solves: [be specific]
- Price/offer: [pricing, any launch special]
- Availability: [when, where, any limits]

Audience:
- Who's receiving this: [describe specifically]
- Relationship: [existing customers, warm leads, cold list]
- What they already know: [have they heard about this before]

Write:
- 5 subject lines (different angles, not variations on one angle)
- Preview text for each
- Email body:
  - Opening hook (why they should care, not what it is)
  - Product intro (what it does for them)
  - 3-4 benefits (customer-focused, not feature-focused)
  - Social proof if available
  - Offer details
  - CTA (prominent and single)
  - P.S. line (urgency or secondary hook)

Tone: [describe how you actually talk to customers]
Length: Under [X] words.

The instruction about subject line angles matters. Most prompts produce five variations of the same idea. “New Product Alert!” vs “Introducing Our New Product” vs “Our New Product Is Here.” Those aren’t different angles. Those are different words for the same angle.

Different angles look like: benefit-focused, curiosity-driven, problem-aware, social proof, direct announcement. Five emails, five reasons to open.

Transactional Emails Nobody Thinks About

Order confirmations, shipping notifications, password resets. These get opened more than any marketing email you’ll ever send because people actually need the information in them.

Most companies waste this attention completely. Here’s the confirmation for your order. Here’s your tracking number. Goodbye.

Improve this order confirmation email:

Current:
[paste what you're sending now, or describe it]

Purchase context:
- What they bought: [product/service]
- Customer type: [first-time, repeat, VIP]
- Post-purchase opportunity: [upsell, referral, review, community]

Rewrite to include:
- Essential transaction info (order number, items, expected delivery)
- Next steps clearly stated
- Expected timeline
- Support info (present but not prominent)
- One value-add that isn't pushy:
  - Getting started tips
  - Care instructions
  - Community invitation
  - Related content

Keep transaction info clear. Add warmth. Don't turn it into marketing.

The constraint in that last line is important. Transactional emails that become marketing emails lose the trust that makes transactional emails effective. People open order confirmations because they need to. If those emails start selling, people start ignoring them. Then they miss actual order problems.

The Meta-Prompt for Email Generation

All these templates share a structure. Context first. Constraints second. Format last.

Most people write email prompts backwards. They start with “Write an email about X” and wonder why the output is generic. The model has nothing to work with. It fills the void with averages.

Here’s the pattern:

Context: Who is this for, what do they already know, what do they want, where are they in the relationship with you?

Goal: What should this email accomplish? Not "engage subscribers." Something specific. Get them to click one link. Reply with feedback. Complete a purchase.

Constraints: What should this email NOT do? What words shouldn't appear? What length is acceptable? What tone is wrong?

Format: What sections should exist? What's the structure? How should it be organized?

Examples: What does good look like for you? What have you written before that worked?

Feed the model this structure and you get output worth editing. Skip the structure and you get output worth deleting.

What This Won’t Do

AI-generated emails work for first drafts, variations, and getting unstuck. They don’t work for replacing judgment about what to send, when to send it, and who should receive it.

The templates here produce raw material. You make it good. You know your audience. You know what’s worked before. You know what your voice sounds like. The model doesn’t know any of that unless you tell it explicitly, and even then, it’s working from your description of those things, not the things themselves.

As one person put it in a Hacker News thread about AI-written emails: “If you’re sending me an email and you’re expecting that I’ll read it, you’re asking me to invest my time (presumably for your benefit). But, you’re unwilling to make the same investment with your time by using a tool to simulate a human connection.”

The investment is editing. It’s checking that the output matches your voice. It’s reading it as your recipient would read it and asking: does this feel written for me, or does this feel generated at me?

That question matters more than any prompt template.

What patterns have you noticed in AI-generated emails that immediately signal “this wasn’t written by a human”? And what prompts have you found that consistently produce output worth using?