SDRcold-email · intermediate
Cold email personalization at scale (without the AI-template smell)
Last reviewed: 2026-05-23 · saves ~8 min/email/run
Our take
The hard part isn't generating personalized emails; it's making them not smell like AI. The 2026 reality: prospects can tell when a first line is auto-generated within 3 words. The fix is constraint-heavy prompting that forces specificity and bans the obvious AI signatures.
Tool stack
Steps
- Use Apollo or Clay to build the prospect list (max 50 per send batch).
- Run the research-brief playbook first; that becomes the input to the personalization prompt.
- Generate first-line + tied-back-to-value-prop second sentence using the prompt below.
- Lavender scores the draft. Anything < 75 → manual rewrite.
- Send via Apollo or Instantly. Track reply rates by personalization quality.
Prompts
Generate cold email first line + value tie-back · Claude Sonnet 4.6
You are an SDR writing cold email openers that don't sound like AI wrote them. Inputs: prospect name, role, recent LinkedIn post or comment (last 14 days), my product's value prop. Output: 2-sentence opener. Sentence 1: reference their recent post/comment specifically, in a way that ties to their job-to-be-done. Use natural phrasing — like a human writing to another human, not a template. Sentence 2: bridge to my value prop in their language, not mine. BANNED phrases (auto-fail if used): - "I noticed", "I came across", "Came across your" - "Hope this finds you well" - "Congrats on", "Great post on" - "Love what you're doing" - "Quick question" - Anything starting with "Just" CONSTRAINTS: - Under 50 words total. - First sentence must reference something < 30 days old. - Don't use exclamation points. - Don't compliment them on being smart, savvy, or thought-leadery.
Pitfalls
- Banned phrases: 'I noticed', 'congrats on', 'love what you're doing', 'I came across'. AI defaults to these.
- Don't reference podcast appearances older than 60 days — they sound out of touch.
- First line tied to LinkedIn activity > company news 9 times out of 10.