Skip to content
← All posts

AI Agents

Prompts for cold email personalization (with examples)

Copy-ready AI prompt templates for cold email personalization: research, first lines, angle selection and quality checks that prevent generic output.

Victor Mendonça
Victor Mendonça

Founder, OretonGTM · ·1 min read

These templates are starting points. Replace the bracketed parts with your own offer and inputs. They work in Clay AI columns, n8n or directly with an LLM API.

Prompt 1: research summary

You prepare account research for a B2B outbound team at [Company], which [one-sentence offer].
Inputs:
- Website text: {website_text}
- Job posts: {job_posts}
- Firmographics: {industry}, {headcount}, {location}
Return JSON with: what_they_sell, who_they_sell_to, relevant_signals (list), fit_reason.
Rules: use only the inputs. If something isn't in the inputs, write "unknown".

Prompt 2: angle selection

Given this research: {research_json}
And these angles:
1. Hiring SDRs: ramp new reps faster
2. Recent funding: pipeline targets increased
3. New sales leader: rebuilding the outbound process
Pick the single most relevant angle, or "none". Return JSON: angle, evidence (quote from research).

Prompt 3: first line

Write the first line of a cold email to a {persona} at {company}.
Angle: {angle}. Evidence: {evidence}.
Rules:
- One sentence, under 25 words
- Must reference the evidence specifically
- Must connect to the problem: [problem you solve]
- No compliments, no exclamation marks, no "I hope this finds you well"
- If the evidence is weak, return NONE

Prompt 4: quality check

Review this opening line: {line}
Research: {research_json}
Return JSON: pass (true/false), reason.
Fail if: it contains facts not in the research, it is flattery, it is generic, or it doesn't relate to [problem].

My take: The quality-check prompt pays for itself. It catches the strange lines that would otherwise go to hundreds of prospects.

Tips

  • Keep examples of great lines in the prompt for tone
  • Review a sample per segment before launch
  • When many lines start the same way, add variety rules

For the strategy behind this, read AI personalization in cold email and how to stop AI hallucinations.

Frequently asked questions

What makes a good AI prompt for cold email personalization?

Clear role and offer context, only verified inputs about the prospect, strict rules on length and tone, structured output and a fallback when no relevant insight exists.

Should one prompt write the whole email?

Usually not. Splitting research, angle selection and first-line writing into separate steps gives more control and better quality.

How do I stop AI personalization from sounding generic?

Ban compliments and vague phrases, require a reference to a specific input, require a connection to the problem you solve and allow the model to return nothing.

Ready to stop doing outbound by hand?

Book a strategy call. We’ll map your ICP, review what’s working (and what isn’t) and show you exactly what we’d automate first.

Book a strategy call

Keep reading