This 5-minute analysis on his own stats replaced every generic subject line tip he’d ever read.
Welcome to Day 23 of the 30 Days of Growth.
This is a pop-up newsletter put together by Chenell from Growth In Reverse. I’ve pulled 30 creators together to help give one short, actionable way you can either grow or improve your email list.
You can view all past issues of the 30 Days of Growth here.
You probably have a theory about what makes your subscribers open your emails. Maybe it’s a certain topic, a keyword, or a format you’ve noticed doing well.
Dan Cumberland had a theory too. He figured his best-performing subject lines were driven by keywords like AI, Claude, and ChatGPT. Seemed obvious, right? The hot topics get the clicks.
So he exported 6 months of his broadcast data and fed it to AI to confirm what he already believed.
He was wrong.
Keywords weren’t the pattern. It was the structure of the subject lines that stood out. Dan found 5 specific structural patterns that separated his top-performing subject lines (55-58% open rates) from his worst (high 20s).
That’s nearly a 30-point gap, and it had almost nothing to do with the topic.
How Dan Did It
He pulled 6 months of broadcast subject lines along with their open rate data. You can do this with an export from your ESP.
He dropped everything into Claude with a simple prompt:
“Here are my broadcast subject lines with open rates. What structural patterns appear in my top performers? What do my lowest performers share? Give me 3-5 rules I should apply going forward, based only on my data.”
For context, Dan writes about how companies and brands can use AI. I wanted to call that out because otherwise the results below sound really off.
Here were the 5 patterns that kept showing up in his highest-performing subject lines:
- Contrarian claim. A statement that contradicts what people assume. “Everyone’s building AI agents. Most are building the wrong thing.”
- Unexpected comparison. 2 things that don’t belong together, put together. “ChatGPT and OnlyFans: Same business model (seriously).”
- Specific number + parenthetical. The aside does more work than it looks like. “5 words that fix bad AI (seriously).”
- Question that implies risk. The reader has to find out. “Is your ChatGPT memory working against you?”
- First-person confession. Simple, yet specific and honest. “I stopped re-explaining myself to AI.”
The key part is that these came from Dan’s own data, not a blog post about “the 96 best subject line tips.” Generic advice says stuff like “use numbers” or “create urgency.” Dan’s AI analysis told him which specific structures his specific audience responds to.
Why It Works
Every email list has its own personality. The subject lines that crush it for a tech-savvy audience might fall flat for a parenting newsletter. Generic subject line advice ignores this completely.
You’re building a rubric from actual behavior, because you’re running this analysis on your own broadcasts. The AI spots structural patterns you’d probably miss because you’re too close to the data (or because you’re looking for the wrong thing, like Dan was with keywords).
This only takes minutes. You’re not hiring a copywriter or running A/B tests for 3 months. You’re asking a question about data you already have and getting a personalized answer back.
Results
- His best subject lines hit 55-58% open rates, the worst landed in the high 20s (nearly a 30-point spread from the same list)
- 5 repeatable structural patterns identified from 6 months of data
- A personalized subject line rubric built in minutes, not months
- The analysis changed Dan’s assumption (he thought it was keywords, it was structure)
How You Can Implement It
Step 1: Export your broadcast data from your ESP. You want subject lines and open rates. A CSV works, or just copy and paste the data directly if your platform lets you select it. Or use an MCP like Beehiiv and Kit have (spoiler alert, I’ve been playing around with Kit’s MCP in beta and it’s awesome.
Step 2: Drop the data into Claude or ChatGPT with this prompt: “Here are my broadcast subject lines with open rates. What structural patterns appear in my top performers? What do my lowest performers share? Give me 3-5 rules I should apply going forward, based only on my data.”
Step 3: Read what comes back and look for the patterns that surprise you. The obvious ones are fine, but the value is in the ones you didn’t expect (like Dan’s keyword assumption being wrong).
Step 4: Write your next 5 subject lines using the patterns the AI identified. Keep the rubric somewhere you’ll actually reference it.
Step 5: If you want to go deeper, connect your ESP via API to a tool like Claude Code, Cursor, or an MCP. That way you can skip the export step and query your broadcast data in real time.
Tools
- Your ESP’s broadcast export (CSV or copy-paste)
- Claude or ChatGPT for the analysis
- A doc to save your personalized rubric
Final Thought
You’re sitting on 6 months (or more) of subject line data right now.
The patterns are already in there. Dan spent a few minutes asking AI to find them and walked away with a rubric that explains a 30-point open rate gap. That’s a pretty good return on a few minutes of work.
P.S. You can follow Dan on LinkedIn, or check out his work at Cumberland Labs.
See you tomorrow,
Chenell
