Company: B2B payments startup
Test: 1000 cold emails - half written by SDRs, half generated via a custom GPT trained on old successful replies.
Result:
AI emails had 20% higher open rates
But human-written emails got 1.8× more positive replies
Why? The AI nailed curiosity and subject lines. But humans knew how to thread relevance into nuance.
Key Insight: Use AI for the hook. Use humans for the subtext.
Automated Personalization Isn’t Personal Enough (Yet)
Founder story: Tried using AI to personalize emails by scraping job titles and recent LinkedIn posts.
Problem: It sounded smart but hollow. “Saw your recent post” without a specific comment killed trust.
Open rate: 42%
Reply rate: 2.1%
When the same founder wrote 20 tailored emails manually, they got a 30% reply rate.
Takeaway: AI can fake familiarity. It can’t fake intent. Personalized ≠ personal.
The Hybrid Play That Beat Both
Startup: LegalTech SaaS doing mid-market outbound
Approach: AI wrote 80% of the email. A human spent 2 minutes tweaking it, changing tone, relevance, and adding one line of insight.
Result:
14% reply rate
4 booked meetings per 100 emails
30% of meetings turned to paid pilots
What worked:
AI reduced the time per email from 8 mins to 2
Humans layered in social context or founder POV
Framework they now use:
AI outlines and drafts
Human adds one line of insight
AI re-optimizes structure and tone
Send with tracking
What's AI’s Real Superpower?
Sales Manager Quote:
“We write the first email. GPT handles the next five.”
Why it works:
AI writes follow-ups that stay on-brand and respectful
Humans often forget or get repetitive
Reply rates went up 23% after email #3
Lesson: Don’t just use AI for the first email. Use it to maintain persistence without sounding like a bot.
When AI Backfires
We reviewed 200 AI-generated cold emails that got ignored, deleted, or publicly posted as bad examples.
Top patterns that tanked replies:
Starting with “Hope this email finds you well”
Referencing “synergies” or “value-driven alignment”
Trying to be overly clever without substance
Founders shared:
“AI writes to impress. Buyers want realness.”
Fix: Instead of “Let’s unlock growth potential together,” say:
“We help ops teams shave 12 hours/week off reporting. Worth a look?”
What’s Changing in 2025
Buyers know when AI is talking. And it’s not necessarily a turn-off unless it feels lazy.
AI-first email tools (e.g. Regie.ai, Smartwriter, Lavender) now offer GPT-4o + CRM insights baked in.
Voice cloning for email tone is on the rise: founders train models to write like them across touchpoints.
“I don’t need to write every email. But I want it to feel like I did.”
— Series A founder, B2B SaaS
It’s Not a Fight, Rather A Formula.
If you’re still asking whether AI or humans are better at writing sales emails, you’re asking the wrong question.
The best teams use:
AI for volume, velocity, and variation
Humans for judgment, timing, and edge
The reply doesn’t come from the writer. It comes from the reader saying:
“This actually feels relevant.”
And that, in 2025, takes both machine and mind.