Writing personalized cold email first lines at scale without sounding like ChatGPT: 90 days of testing what actually moves reply rates
firstline_labs_raya · 2026-09-21 · 0 views
I spent 90 days testing personalized first lines at scale. Here is what the data actually showed.
The problem with AI-generated first lines. Tools like Clay and Apollo have made it trivially easy to generate personalized openers at volume. Which means every prospect on your list is probably getting five AI-written first lines a week that all sound eerily similar. "I noticed your company just raised a Series B, congrats." "Saw you recently expanded into new markets." These phrases are becoming the cold email equivalent of "I hope this email finds you well." They read as automated because they are.
What actually worked in my tests. The highest reply rates came from first lines referencing something genuinely specific and hard to automate. A podcast the person had recently guested on. A specific point from a LinkedIn post they wrote themselves. A job listing that implied a concrete business problem. These require actual research but the reply rate difference was significant. 5.2% vs 2.1% on the same sequence with the same list.
The compromise approach that scales. Segment your ICP tightly enough that one well-written first line feels personal without being individually researched. If you are targeting VP of Sales at Series B SaaS companies with 20 to 50 reps, you can write a first line referencing the specific challenge that person faces at that stage without naming them individually. At that level of specificity, the email reads as researched even when it's not one-to-one.
The rule I apply now. AI handles tier 3 personalization: company name, title, industry. Human judgment handles the positioning and framing. A good first line answers "why this person, why this problem, why now" in one sentence. If your AI output cannot answer all three, rewrite it.
Volume without relevance is noise. You are not sending 600 emails a day hoping the 1 in 600 bites. You are sending 600 because the math of 4% reply rate means 24 qualified conversations. The 4% comes from relevance, not volume.