Austin dentists — Google Maps local pull

What was asked

Show what the system produces when an agency wants a local-services lead list — dentists, law firms, HVAC contractors, chiropractors — sourced without paying for a data vendor.

What the system did

The pipeline drives the public Google Maps UI with a headless browser (Playwright). Given a search term ("dentist") and a city ("Austin, TX"), it:

1. Loads maps.google.com/maps/search/dentist+Austin,+TX
2. Scrolls the results panel to surface up to N unique listings
3. For each listing, opens the detail card and pulls: business name, website, phone
4. Normalizes each row to the 13-column scraped_leads schema and writes them to Postgres

No API key, no paid credits, no vendor dependency — and importantly, no “fake-looking” contact info: every row here is a real Austin dental practice you can verify in Google Maps right now.

Headline numbers

• 25 practices returned
• 24 of 25 have a resolved website domain (96%)
• 24 of 25 have a company name matching the business card
• Sample includes ATX Family Dental, North Austin Dentistry, Austin Dental Works, Tech Ridge Dental, and 21 others

What a pilot engagement gets on top of this

This sample stops at the raw scrape. A real outbound retainer plugs this CSV into:

1. Email verification — run each domain through MX/SMTP/catch-all checks so you don’t send to dead mailboxes.
2. Opener generation — Claude writes a 2-sentence personalized cold email opener per lead, using the practice name, city, and dentistry-specific hooks.
3. Warmed-inbox outbound — send from a pre-warmed Gmail/Google-Workspace inbox, track opens + replies, and classify positive vs negative replies automatically.
4. Meeting booking — positive replies route to a calendar link; the system updates the outreach_contacts row with the booking.

All of the above is wired into the same database the scraper writes to — the lead list is just the front door.

Caveats

• Volume is rate-limited by design: the scraper respects ≤ 2 searches/min, ≤ 200 results/search to stay far under Google’s bot-detection thresholds. For bigger runs we either split across inputs or add IP rotation.
• Occasional profile pages are slow to load, so ~4% of rows skip enrichment (company/domain/phone may be missing). In this sample, one of 25 landed without a company/domain.
• The Maps UI is public-facing HTML, so structural changes on Google’s side can require a selector update. We treat the scraper as monitored maintenance, not a fire-and-forget tool.
• Phone numbers aren’t included in this particular export because the UI only surfaces them after a second click-through we didn’t run for the demo. A real pilot enables that step.