Write a first line for every lead with AI, and check it

What it does

A first line that shows you looked at their company gets read. AI can write one for every lead in minutes. It can also write a confident line about the wrong company, in the wrong language, full of jargon you’d never use. The work is in the checking, so most of this playbook is about that.

What you need

• Your list, with each company’s website
• A tool that runs an AI prompt per lead, like Instantly or Clay
• An hour to read the lines

How it works

1. 1

   Decide what the line says

   Pick one kind of fact that’s easy to check on a website, like who they sell to or what they just launched. One kind of fact per campaign makes the lines fast to check.

2. 2

   Write the prompt with hard rules

   Prompt, run once per lead

   You're writing the first line of a short email to the founder of {{companyName}} ({{website}}).
   
   Read their website. Write one sentence, under 15 words, that says who {{companyName}} sells to, in plain words a founder would say out loud.
   
   Rules:
   - Only say what their website says. If you can't tell who they sell to, write exactly: SKIP
   - Use the company's short name, the way they write it on their own site.
   - No jargon and no category labels. Never use: solutions, platform, leverage, innovative, cutting-edge.
   - No compliments, no questions, no exclamation marks.
   - Write in English.
   
   Return only the sentence, or SKIP.

3. 3

   Test it on ten

   Run it on ten leads, open each website, and check every line. Fix the prompt until all ten are right. Most tools charge per lead, so test small before you run the list.

4. 4

   Run it, with a fallback

   Every SKIP gets a fallback line that’s true for everyone on the list. A plain true line beats a specific wrong one.

5. 5

   Read every line back

   Export them and read them all. Look for lines about a different company with the same name, banned words that got through anyway, lines that run long, other languages, and companies that have shut down. On one run of about 2,900 lines, about 1 in 12 needed a fix.

6. 6

   Check the company names too

   Databases scrape names like “Northlineinc” for “Northline”. The AI often cleans up the name inside its line, while the company name field used elsewhere in the email, like the subject, stays raw. Fix the field, not just the line.

What good looks like

Every line is true, specific and sounds like you wrote it, and you’ve read each one before it went out.

Mistakes to skip

• Skipping the read-back because the test ten looked fine. Problems show up at scale.
• Letting the AI guess. If the site doesn’t say, the line should be the fallback.
• Long first lines. One short sentence. The rest of the email still has to fit in 50 words.