Writing AI Emails That Don't Sound Like AI

Why AI emails are easy to spot

By 2026, decision-makers can identify AI-written cold email in under 3 seconds. The tells aren't the words specifically — they're the rhythm and structure. Common AI fingerprints:

• Three-sentence paragraphs of similar length. Real humans vary. AI defaults to balanced.
• Setup-payoff structure for every paragraph. AI was trained on essays. Real emails are messier.
• Specific buzzwords in clusters. "Innovative solution," "leverage cutting-edge," "synergize."
• The exact phrase "I hope this email finds you well." Now functions as the literal AI signature.
• Generic compliments. "I was really impressed by what you're doing." (What specifically? AI doesn't actually know.)
• Perfect grammar in casual contexts. Real cold emails between professionals use contractions, sentence fragments, and the occasional ellipsis.
• The smooth-arc structure. Open → context → pitch → CTA → sign-off. Always exactly that. Predictable = AI.

The 5-rule framework for human-sounding AI emails

Rule 1: Ban the openers

Block these phrases at generation time. Not "discourage" — block:

• "I hope this email finds you well"
• "I came across your company"
• "I wanted to reach out"
• "I was really impressed by"
• "I'd love to connect"
• "I noticed that you"

Replace with a specific observation. "Saw your relaunch post." "Noticed the new pricing page." "Just read your last blog post on ___." Concrete > generic.

Rule 2: Vary sentence length

Real humans write sentences of wildly different length. A 4-word sentence followed by a 22-word sentence reads human. Three medium sentences in a row reads AI. Force at least one sentence under 8 words and one over 18 in every paragraph.

Rule 3: Use contractions and fragments

"I'm" not "I am." "Don't" not "do not." Occasional sentence fragments — "Worth a quick look?" — are how real humans actually write business email. Perfect-grammar emails read AI.

Rule 4: Include one specific, falsifiable detail

The single biggest signal that an email was written by a human is one piece of information that AI couldn't have hallucinated. "The way you defer the workspace setup until after the first 'aha' moment in your onboarding flow." That's a falsifiable claim. Either it's true (proves research) or it's wrong (instant credibility loss). AI either fakes it or skips it. Real research includes it.

Rule 5: Sign off like a human

First name only, no title, no fancy signature, no banner image. "Sam" beats "Sam Davies, Senior Account Executive | Acme Corp | acme.com | LinkedIn | Calendly." The fancier the signature, the more it screams campaign.

Before vs after: same email, different result

How to actually enforce these rules at scale

You can manually edit every AI-generated email to follow these rules. Most people do this for the first 5 emails, then give up. The right answer is to enforce the rules at generation time, not edit time. Tools that do this:

• Mail Doneky: Built around these rules. 50+ banned phrases hard-blocked at generation, AI-cliché detection, sentence-rhythm variation, shallowness scoring per draft.
• Custom GPT prompt: If you're rolling your own, your system prompt must explicitly ban the openers, demand sentence variation, and require one specific researched detail per email. Most generic "write a cold email" prompts skip all of this.
• Manual edit pass: Acceptable for 5 emails. Doesn't scale to 50.

The shallowness test (do this every time)

Before sending any AI-generated email, ask: "Could I have written this email by reading only the prospect's LinkedIn headline?"

If yes, it's shallow. AI defaulted to surface-level language. The email will read generic.

If no — if the email contains a detail you could only know from actually reading their site, blog, or recent posts — it's depth-tested. That's the bar.

Mail Doneky runs this test automatically as a "shallowness score" on every draft. Drafts below threshold are regenerated with more research.