How to Use AI for Cold Email That Actually Gets Replies
Use AI to research prospects and personalize the opening line, not to write the whole email. A formula that works: personalized first line (AI-researched), one specific value proposition relevant to them, a low-friction call to action. Tools: Clay for research at scale, Lavender for scoring, Claude for individual personalization.
Why Most Cold Email Fails
The average cold email gets a 1-3% reply rate. The emails that work get 10-30%. The difference is almost never the CTA, the subject line, or the timing — it’s the opening two sentences. If the recipient reads the first two sentences and thinks ‘this is not about me, this is a template’, the email is deleted.
AI solves both sides of this: the research (finding something specific and relevant to the recipient) and the writing (making the email sound like it was written for one person, not a list of 500).
The AI Cold Email Workflow
Step 1 — Research the Prospect
Before writing anything, you need one specific, relevant hook — something about the recipient that’s neither obvious nor creepy. Good hooks: a recent article they wrote, a product launch, a hiring trend visible on LinkedIn, a podcast appearance, a company milestone.
For individual emails, Claude or ChatGPT can research a person from their LinkedIn URL, company website, and recent content. For campaigns at scale, Clay automates this research across thousands of leads using AI enrichment.
Paid
AI-powered lead research and enrichment. Pulls data from LinkedIn, company websites, news, and dozens of other sources to build prospect profiles automatically. Used for personalization at scale.
Step 2 — Write a Personalized Opening Line
The opening line should reference something specific to the person. Not ‘I saw you’re the CMO at Acme’ (too generic) but ‘Your post about reducing churn with personalized onboarding last month was the best take I’ve seen on it — we’ve been grappling with exactly that problem with our enterprise clients.’
Give Claude the research you found and ask it to write three variations of a personalized opening line — each taking a different angle on the same information.
Step 3 — One Specific Value Proposition
The middle of the email should answer one question: why should they care? The mistake most cold email makes here is listing features or making generic claims. Instead, make one specific, relevant connection between their situation and what you do.
Use the prospect’s own language where possible. If their LinkedIn says they’re focused on ‘reducing time to value for enterprise customers’, use that phrase when describing what you do.
Step 4 — A Low-Friction CTA
‘Book a 30-minute demo’ is too much commitment for a cold email. ‘Would a quick 10-minute call make sense to see if this is relevant?’ is better. ‘Does this resonate with what you’re working on?’ is even lower friction — it invites a reply without requiring a calendar commitment.
Ending with a single, answerable yes-or-no or one-word question gets higher reply rates than any calendar link. ‘Is improving X something on your radar for this quarter?’ takes 3 seconds to reply to.
AI Tools for Cold Email
Free plan
Real-time AI email coach. Scores your email and tells you what’s hurting reply rates — too long, weak subject line, poor opening. Works as a Gmail extension.
Paid
AI prospect research and enrichment at scale. Personalizes opening lines across large contact lists automatically.
Paid
Cold email sequencing with AI-generated variations. A/B test subject lines and email copy automatically. Warm-up feature for new sending domains.
What to Never Use AI For in Cold Email
Don’t use AI to write the whole email and send it unedited. AI cold email at scale is recognizable as AI cold email — the structure is too clean, the opening too warm, the transition to the pitch too smooth. Recipients have seen thousands of these emails. The best cold email sounds slightly imperfect, like a person wrote it quickly between meetings.
Increasing cold email volume to compensate for low reply rates is a race to the bottom. A 10-email campaign with real personalization will consistently outperform a 1,000-email blast. AI makes better personalization possible at reasonable scale — but there’s no substitute for actually caring about the person you’re emailing.
- ✅ Research prospects: Clay (scale) or Claude (individual)
- ✅ Write opening line: AI-generated personalized hook based on research
- ✅ Middle: one specific, relevant value proposition in their language
- ✅ CTA: low-friction question rather than calendar link
- ✅ Score before sending: Lavender for feedback on each email
Install the Gmail extension. Start scoring your emails before you send them — most users see their reply rate improve in the first week.