Email Marketing with AI: More Clicks, Less Cringe

I’ve received enough AI-generated emails to know exactly what they look like. The subject line has unnecessary urgency. The first line is “I hope this finds you well” or some variation of it. By the third sentence it’s pitching something. By the end I’ve forgotten who sent it.

This is the default output when you give AI a vague prompt and hope for the best. It’s what happens when people think AI is a replacement for thinking rather than a tool that amplifies thinking.

Here’s what actually works.

The Open Rate Problem Is a Subject Line Problem

Email open rates live and die by subject lines. Not the copy. Not the offer. The subject line and the sender name.

AI is genuinely good at subject line generation if you prompt it correctly. The mistake most people make is asking for “a catchy subject line for this email.” That produces generic output. Here’s what I do instead:

I write the email first. Then I ask Claude: “Read this email. The core emotional hook is [X]. The reader’s likely objection is [Y]. Write 10 subject lines that create curiosity without being clickbait, that feel personal not broadcast, and that are under 45 characters. Avoid words that trigger spam filters: free, guarantee, limited time, urgent.”

Then I pick the best two and A/B test them. My open rates went from 28% average to 38% average using this method over six months. That’s not a small difference.

The First Line Is Everything

Most emails lose the reader in the first line. They either restate the subject line, introduce themselves unnecessarily, or lead with the company’s name.

Here’s the rule I live by: the first line of an email must either make the reader feel understood or make them curious. It cannot do both jobs mediocrely. It has to nail one.

For a nurture email to my list: “You signed up a while ago and probably haven’t thought about me since.” That’s understanding. It’s honest. It works.

For a cold outreach email: “I looked at [specific thing about their company] and had a reaction.” That’s curiosity. What reaction? They want to know.

I prompt Claude with: “Write 5 first-line options for this email. Each must either create strong curiosity or signal deep empathy with the reader’s situation. No generic openings. No company names. No ‘I hope you’re doing well.’”

The Framework I Use for Every Email

I don’t write email copy from scratch anymore. I write a framework, then use AI to draft from it. The framework:

Who: Exactly who is receiving this. Not “marketers.” “Marketing directors at Series B SaaS companies who’ve tried content marketing and been disappointed by the results.”

Where they are: What they’re thinking, feeling, and believing right before they open this email. What problem are they sitting with?

Where I want them to go: One action. Not “learn more” and also “check out our blog” and also “reply if interested.” One action.

The honest pitch: What’s the real value? Not features. The specific change in their situation if they do the thing I’m asking.

I drop that framework into Claude and say: “Draft an email from this framework. Under 200 words. One short paragraph per section. Conversational, not formal. Sound like a smart person emailing a smart person.”

200 words is the target for any email that’s asking for action. Longer is fine for newsletters where people opted in for long content. For anything with a CTA, shorter wins.

Segmentation and Personalization at Scale

Here’s where AI gets genuinely interesting for email. Basic personalization - first name, company name - is table stakes. Real personalization is different content for different segments.

I tag my list by behavior: what they clicked, what they downloaded, how far they got in an onboarding sequence, what they bought. Then I use Claude to write segment-specific variations of my emails.

The same campaign might have 3 versions:

Version A for people who’ve engaged with content about automation
Version B for people who’ve engaged with content about strategy
Version C for cold leads who haven’t clicked anything in 60 days

The underlying offer might be the same. The framing, the examples, the hook - those are different because the reader’s context is different.

I prompt this by giving Claude the full email and saying: “Rewrite this email for a reader who [segment description]. Keep the core offer the same. Change the hook, the example in the middle, and the CTA phrasing to match this reader’s specific situation.”

Cold Email Is a Different Animal

Cold email and list email are not the same skill. Cold email requires ruthless brevity and hyper-specificity. List email can have warmth and depth because the reader opted in.

For cold email, I use AI differently. I don’t start with a prompt to write an email. I start with research.

I give Claude a prospect’s LinkedIn bio, company about page, and any recent news about them. Then I ask: “What is this person’s likely biggest professional priority right now? What’s probably keeping them up at night? What would a stranger have to say to make them stop and respond?”

That output shapes the hook. Then I write a 5-sentence email: one line that shows I know something real about them, one line that identifies the problem, one line that states my specific angle on solving it, one line that asks for a small commitment (15-minute call, a specific question, a yes or no).

AI-written cold emails that are generic get deleted. AI-informed cold emails that feel like they came from a human who did their homework get replies.

What I Never Automate

The follow-up reply. When someone responds to an email - cold or warm - the next message needs to be human-written. Every time. AI can help me draft it, but I read and edit it personally before it sends.

The reason: the reply is where the relationship actually starts. Fumbling it with generic AI output at that moment is worse than fumbling the first email.

Also: email sequences where someone just had a bad experience. If someone complained, churned, or expressed frustration - no AI drafts that response. That’s a human conversation.