Check every company on a list fits, with AI

What it does

Database filters get you close, but they can’t read a website. A company tagged “logistics” might be a freight broker with no trucks, a software company that sells to logistics, or a one-person consultancy. An AI fit check reads each company’s site and answers your fit rules one by one, with the words that prove it, so the list that goes out only has companies you’d actually sell to.

What you need

• Your sourcing spec (see write a sourcing spec)
• The list, with each company’s website
• Clay, a spreadsheet with an AI column, or Claude or ChatGPT for a short list

How it works

1. 1

   Turn your spec into yes-or-no questions

   Three to six questions a website can answer. “Do they run their own trucks?” works. “Are they a good fit?” and “Are they growing fast?” don’t. Add one last question for the lookalikes you leave out: agencies, consultancies, software companies and resellers that serve the industry rather than work in it.

2. 2

   Write the prompt

   Prompt, run once per company

   Read [WEBSITE] and answer these questions about [COMPANY]:
   1. [QUESTION 1]?
   2. [QUESTION 2]?
   3. [QUESTION 3]?
   4. Is it an agency, a consultancy, a software company or a reseller that serves this industry, rather than a company in it?
   
   For each question, answer YES, NO or UNSURE, then quote the words on their site that show it. If the site doesn't say, answer UNSURE. Don't guess from the company's name.
   
   Last line: FIT if every answer is the one I want ([THE ANSWERS YOU WANT, E.G. YES, YES, YES, NO]), NOT A FIT if any answer is wrong, and CHECK if anything is UNSURE.

3. 3

   Test it on 20 companies you’ve already judged

   Ten you know fit and ten you know don’t. If it gets more than two wrong, find the question it misread and rewrite it before you run the whole list.

4. 4

   Run the whole list

   One company per row, in Clay or a sheet with an AI column. Keep the quotes in their own column, so anyone can check an answer in seconds.

5. 5

   Read every CHECK, and spot-check the rest

   Read the unsure ones yourself: thin websites belong to plenty of good small companies. Then read 20 random FITs and 20 random NOT A FITs. If you’d overrule more than two in either pile, fix the prompt and run it again.

What good looks like

Every company on the list has a FIT, with the quotes from its own site that prove it, and you’ve read every unsure one yourself.

Mistakes to skip

• Questions a website can’t answer, like “Are they growing fast?”
• Treating UNSURE as no. Read them. Some are your best fits with a thin website.
• Skipping the test on companies you’ve already judged.