AI prompts for email marketing — how to actually use ChatGPT well
Most ChatGPT-generated marketing emails are bad. Not because the model can't write — but because the prompts feeding it are vague, contextless, and full of exclamation points the AI loves and your readers hate.
The gap between "write me a marketing email" and a prompt that produces something you'd actually send is huge. And it's mostly fixable. Here's what works, what doesn't, and the exact prompt patterns small businesses and solo creators can copy today.
Why most AI email prompts fail
Adobe reports that 47% of marketers and business owners now use ChatGPT in their workflow. But ask ten of them to share their prompts and most look the same: a one-line request, no audience, no goal, no voice notes. The output reflects the input.
Twilio's analysis of email prompts puts it bluntly — successful prompts deliver results that are "genuinely good quality, not merely good for AI." The difference comes down to four inputs the model needs every single time: a clear deliverable, a customer persona, tone guidelines, and a marketing framework to follow.
Skip any one of those and you're rolling the dice. Skip all four and you'll get exactly what most people get — generic, exclamation-heavy filler.
The 5 W's prompt framework
Mailchimp's email team recommends a structure they call the 5 W's, and it maps cleanly onto how AI parses requests:
- Who — who is the email for, in detail
- What — the specific output you want
- Where — geography, language, or platform context
- Why — the campaign goal and what success looks like
- How — format, tone, length, what to avoid
A prompt missing any of these will hallucinate the missing piece. The model will invent a generic audience, default to a bland tone, or pad word counts to look helpful. Give it all five and the output narrows fast.
Here's the bare-bones template:
You are an email copywriter for [business type].
Write a [email type] for [audience description].
The goal is [specific outcome].
Tone: [3 adjectives].
Length: [word count or paragraph count].
Constraints: no exclamation points, no clichés like "in today's
fast-paced world," no emoji unless I ask.
That alone outperforms 90% of the prompts people actually use.
Prompt template 1 — subject lines that don't sound like spam
Copy.ai's email team and Copyhackers both recommend a two-step approach for subject lines: first generate a list of formula types, then apply each formula to your specific email. One-shot prompts produce the same boring patterns every time.
Try this:
I'm sending an email about [topic]. The body of the email covers:
- [point 1]
- [point 2]
- [point 3]
Generate 12 subject lines under 50 characters using a mix of these
formulas: question, curiosity gap, specific number, direct benefit,
contrarian take, personal/insider framing.
Label each line with the formula used. No exclamation points.
The labelling matters — it lets you spot which formulas land for your audience over time. According to Copyhackers, this systematic approach "consistently produces testable variations for A/B testing." Skip the labels and you can't learn anything from your sends.
What doesn't work
- Asking for "catchy" subject lines without examples — you'll get rhyming filler
- Letting ChatGPT default to its tone — the Twilio team warns it loves exclamation points that "typically don't perform well in marketing messaging"
- Generating one subject line at a time — variety only emerges in batches
Prompt template 2 — the welcome email that earns a reply
Welcome emails get the highest open rates of any sequence — and they're the one place where AI output without editing tends to crash hardest. Generic welcome copy reads like every onboarding email ever written.
The fix is concrete context. Here's the prompt:
Write a welcome email for someone who just joined the [name] newsletter.
About me: [1-2 sentences on who you are and what you write about]
About them: [why someone subscribes — the problem they're trying to solve]
What they'll get: [cadence, topics, format]
End with one open question that invites a reply. The question should
help me understand the new subscriber better.
Tone: warm, direct, slightly informal. Around 150 words. No CTA buttons.
Plain text only.
The reply-prompt at the end is the move most people miss. Snov.io's email research notes that welcome emails with an explicit question generate measurably more replies — and replies are the strongest signal to inbox providers that your emails are wanted.
Prompt template 3 — the sequence, not the email
Single-email prompts force you to re-explain context every time. Sequence prompts make ChatGPT think across emails — pacing, tension, payoff.
Build a 5-email sequence for [audience] who signed up for [lead magnet].
Day 1: deliver the lead magnet, set expectations
Day 2: share the origin story or insight that makes me credible
Day 4: tackle the biggest objection [audience] has about [topic]
Day 6: case study or concrete example
Day 8: soft pitch for [offer], with one clear CTA
For each email give me: subject line, preheader, body (200-300 words),
single CTA. Use the Hook-Value-Proof-Ask framework on every send.
No exclamation points, no urgency manipulation.
The Hook-Value-Proof-Ask structure comes from email copywriting frameworks compiled by Instantly's copywriting team — it forces every email to earn the click rather than just demand it.
Prompt template 4 — re-engagement without the desperation
Most re-engagement emails read like a breakup text. ChatGPT will happily write you one unless you stop it.
Write a re-engagement email for subscribers who haven't opened
in 90 days.
Don't guilt-trip. Don't apologise for emailing. Don't ask "are you
still there?" Instead, lead with one piece of useful, specific value
they can act on right now.
Close with a low-friction option to update preferences or unsubscribe.
Tone: matter-of-fact, respectful. 120 words max.
Mailchimp's re-engagement template guidance emphasises the same point — the messages that win back subscribers are the ones that still feel like the newsletter they signed up for, not a panic plea.
The voice anchor — better than any persona
Most guides tell you to start with "You are an expert email copywriter with 15 years of experience..." It helps a little. What helps more is feeding the model your actual writing.
Here are three emails I've written that performed well. Study the
voice — sentence length, vocabulary, rhythm, what I avoid.
[paste email 1]
[paste email 2]
[paste email 3]
Match this voice for everything we write today. If I ask for
something that would break it, push back.
Foundation Inc's email team calls this the voice anchor. Your three best emails become the reference the model calibrates against for the rest of the session. It does more for output quality than any persona instruction.
Mistakes to stop making
- Asking for "engaging" anything. Specify what should happen — a click, a reply, a forward.
- Defaulting to long emails. Most drafts are 30% too long. Set a word limit and the model edits itself.
- Skipping the constraints list. Listing what to avoid saves 10 minutes of editing per email.
- Treating the first output as the answer. The follow-up prompt — "cut anything that doesn't earn its place" — almost always produces a better draft.
Where AI prompts fit in a Doxiefy workflow
Prompts get you a strong first draft. They don't replace the parts of email marketing that actually convert — list quality, segmentation, send timing, and the relationship behind the inbox.
That's the gap Doxiefy is built to close. Our AI-assisted campaign builder takes the same prompt patterns covered above and folds them into a visual sequence builder — so you're not jumping between ChatGPT, your email tool, and a spreadsheet trying to keep voice consistent across a five-email flow.
If you're a small business or solo creator who's tired of staring at the ChatGPT cursor at 11pm, join the Doxiefy waitlist and we'll show you the workflow that turns prompts into campaigns that actually send.
Frequently asked questions
What's the best ChatGPT prompt for email marketing?
There isn't one — there's a structure. Use the 5 W's (who, what, where, why, how), anchor the voice with three of your existing emails, and always include a constraints list. That structure outperforms any single "best prompt" template.
How long should an AI email prompt be?
Long enough to be specific, short enough to read in 30 seconds. Most effective prompts are 80–150 words. Shorter usually skips audience or tone. Longer tends to contradict itself.
Can ChatGPT write subject lines that get high open rates?
Yes — but only when you give it the email body, ask for a specific number of variations using different formulas, and label each one. Generic "give me catchy subject lines" prompts produce subject lines that look catchy and perform like spam.
Should I edit AI-generated emails before sending?
Always. Treat the output as a 70% draft. Cut filler, add one or two specific details only you'd know, and read it out loud before sending.
Is it okay to use AI prompts for email marketing if I'm a solo creator?
Solo creators benefit the most. Prompts collapse drafting time from hours to minutes — but voice, audience knowledge, and relationship still have to come from you.
Final thoughts
Prompts aren't magic. They're a brief — and like any brief, they get the output you describe, not the output you wished for.
The marketers getting real value from AI in 2026 aren't the ones with the cleverest single prompt. They're the ones who built a small library of patterns, anchored their voice, and treat the model as a fast first-drafter that needs an editor on the other side.
Start with the 5 W's. Add your voice anchor. Keep your constraints list nearby. Send better email faster.