How AI writes cold email that doesn't read like AI

Most people can spot an AI cold email in the first line. "I hope this finds you well." "I was impressed by your innovative approach." Three long paragraphs, a few buzzwords, two asks at the end. The reader has seen it a hundred times this month and deletes it before the second sentence.

The model isn't the problem. The input is. An AI that knows only a name, a company and a job title has nothing true to say, so it fills the space with phrases that fit everyone. Give it real facts, clear rules and your voice, and the same model writes an email a person would send.

Why AI cold emails sound like AI

The tells are the same every time:

• Generic flattery. Praise that could be said about any company, so it reads as praise about none.
• Filler openers. A first sentence that says nothing about the reader.
• Long paragraphs. Four sentences where one would do, on a screen read between meetings.
• Buzzwords. Streamline, boost, maximize, comprehensive: words that describe every product.
• The same structure. Compliment, problem, pitch, ask. Every email, every person.

Each one is what a model does when it has to write 100 words and knows nothing worth saying.

Generic vs researched: an AI cold email to Draymount

Here is the same email written twice, by the same AI, to Ravi, VP of Sales at Draymount. In this example the sender is Maya at Ostervane, which sells onboarding for sales teams.

The first draft had a name, a company and a title to work from. The second had three facts from research, Ostervane's knowledge and four rules. It's 25 words shorter, opens with something Ravi can check is true, and asks one question.

The test is simple: swap Draymount for another company. The first draft still works, so it's generic. The second falls apart, so it was written for Ravi.

What to give the AI before it writes

Research: two or three true, specific facts. Things that happened, with a date: roles posted, an office opened, a product launched, a team that grew. Specific facts are the cure for flattery, because the AI no longer has to invent something nice to say. The fact usually becomes the opener. The post on writing a cold email first line covers how to turn one into the first sentence.

Knowledge: what you sell, who you reach and how you sound. Without it, the AI describes your product in the vaguest terms possible. With it, it can name the one outcome that matters to this reader. Write it once and every email draws on it.

Constraints: rules it can't break.

• Under 90 words for a first email.
• No sentence longer than 20 words.
• One ask, as a question.
• A banned list: "I hope this finds you well", "I came across", "quick call", "streamline", "innovative".

Examples: three to five emails you'd be glad to send. Rules say what to avoid. Examples show what good sounds like: how long your sentences run, how you open, how you sign off.

Review the drafts until they pass unchanged

Early on, read every draft before it goes out. Each time you change a draft, ask why, and turn the answer into a rule or an example in the knowledge. Fixing one email helps one person. Fixing the knowledge helps every email after it.

You'll see it working when your edits shrink: fewer words changed, then only a word here and there, then nothing. When drafts pass unchanged, your review has stopped adding anything. The post on Autopilot or Copilot covers when to make that switch.

How Leadgenerator.io writes AI cold email

Agents research each company and each person, then write every email, first email and follow-ups, from that research and your knowledge: what you sell, who to reach and how emails should sound.

On Copilot, every draft waits in a review queue, where you approve, edit or reject it, one at a time or in bulk. When drafts pass unchanged, switch to Autopilot and the agents find, write and send on their own.