This is my whole method, published.
Run it yourself if you don't believe me.
ChatGPT rebuilds its answer from scratch every time you ask. Google serves one stored page. ChatGPT doesn't. The same question, asked twice, can name two different roofers.
So one ask is one data point. The real answer is a count. I use ten questions and ask each five times. That makes 50 fresh asks.
50 real asks. Ten homeowner questions. Five fresh asks for each one.
Logged out, fresh chat, every time. So ChatGPT can't flatter anybody. It answers like it answers a stranger.
Every answer saved as a screenshot. All 50 get proof links. Check any one against my count.
The real count gets printed. If an ask fails, I rerun it. I never report a failed run as a zero.
Your competitors get counted too. Same asks, same rules. That's how you see the whole board, not just your row.
Open ChatGPT. Start a fresh chat. These are three of the ten questions:
"My roof is leaking in [your city]. Who should I call?"
"Best roofer near me in [your city]. Give me three names."
"Getting bids on a roof replacement in [your city]. Who do you trust?"
Ask each one five times. Use a fresh logged-out chat each time. Count your name. That is the work.
historical Dallas receipt · earlier 100-Ask test · August 14, 2026
An earlier report. The count, then the local board.
Earlier receipt stack. The current test saves 50.
That is the free test. I run it. You get the report. We walk the fixes in 15 minutes.
Show me where I rank Free · 50 asks · proof links included