Blog · AI & Outreach
April 2, 2025 · 5 min read
Every major outbound tool now has an AI writing feature. Paste in some company information, click generate, and a cold email appears in seconds. The emails are grammatically correct, professionally structured, and almost universally ineffective. Not because they are badly written — they are not. Because they are not written for anyone in particular.
AI makes the average email better. The problem is that "better than average" is not good enough in a market where every competitor is also using AI. When the floor rises, you need to raise your ceiling.
AI is genuinely useful for cold email in specific, constrained tasks. Given a specific fact about a prospect — a recent hire, a product launch, a LinkedIn post — AI can quickly turn that fact into a plausible first line. Given a feature of your product, AI can rephrase it as a benefit in several different registers. Given a mediocre draft, AI can suggest ways to tighten it.
These are editorial and transformative tasks, not creative ones. AI works best as a co-editor, not as an author. The author — the person who decides what observation to lead with, what problem to name, what tone to take — needs to be human.
AI without good input produces AI output — confident, fluent, and generic. "I came across your company and was impressed by what you're doing in the manufacturing space" is AI output. It sounds like personalisation but contains no information that could only have been written for this prospect. Experienced buyers recognise it immediately.
AI also struggles with the judgment call at the heart of every good cold email: which of the ten things I know about this prospect is the one worth saying? That choice requires understanding the buyer's priorities, the product's specific value, and what is likely to resonate in this particular context. No model does that reliably without explicit guidance.
Collect a specific, verifiable fact about the prospect first. Then write a first line based on that fact that connects it to the problem you solve. Use AI to generate two or three variations. Pick the one that sounds most human, tighten it, and continue the email in the same voice. Review the final email as if you were the prospect. Would you read the second sentence?
That workflow — human research, human judgment on relevance, AI for variation and editing, human for final review — produces outreach that benefits from AI without becoming dependent on it.