How it works
A 4-stage pipeline that mirrors his manual checklist
Most automation for this kind of work just throws screenshots at GPT and hopes for the best. We extracted every signal from his Loom — toll-free numbers in the nav, "find a location" links, popup scripts, hero animations, photo quality — and split them across the cheapest stage that can detect each one.
01
Scrape
Apify Google Maps actor pulls 120–140 leads per niche × city. Captures phone, address, rating, review count, category.
cost: $0.004/lead · ~3 min
02
DOM signals
Fetch each site's HTML. Detect multi-location patterns, phone-count-in-nav, popup scripts, third-party software (HardHatPro, GlossGenius, Wix), copyright year, HTTPS.
cost: free · ~5 min
03
Screenshot
Headless Chromium captures hero region + load timing. Slow loads ('loading very slow' from his Loom) become a signal.
cost: free · ~10 min
04
Vision verdict
GPT-4.1-mini sees the screenshot WITH the DOM signals as context. Returns 3-bucket verdict + structured reasoning: visual quality, design era, photo quality, template feel.
cost: ~$0.0005/lead · ~3 min