Building a local lead list with AI:
100 dentists, verified emails, cost per lead

How many dental practices in Austin can we reach by email, how good are those addresses, and what does each usable lead cost?

The rest, about $0.03, was the model calls. The leads tool searches Google Maps and visits each practice's website for emails, phones and social links, so one call gives a list that is already enriched.

Phones are easy, emails aren't. 99 of the 100 practices have a phone number on Google Maps, and all of them link a social profile. Only 67 list an email on their website, and those 67 share 44 addresses: one oral-surgery group runs 14 locations from the same two inboxes. Counting places instead of inboxes would have overstated the list by half.

Half the emails are risky. Of the 44 addresses, 22 were deliverable, 19 risky and 3 bad. The risky ones were mostly catch-all domains (13), which accept any address so you can't tell if a person reads it, and mail servers that didn't answer (9). 22 of the 44 were role addresses such as info@ or office@, normal for a small practice.

21 practices are actually reachable. By practice: 31 have a deliverable email, 34 only a risky one, 2 only a bad one and 33 none. In unique businesses, that's 21 with a deliverable inbox: a clean, honest list size from a search of 100.

Cost per verified lead: under 1¢. $0.19 for the whole run, or $0.009 per business with a deliverable inbox, the personalised first lines included.

We asked gpt-5.6-luna, a fast and inexpensive model, for one sentence per practice that uses one true detail from its data, and nothing else. With only the Maps listing to go on, it was accurate and flat:

"[Practice] is listed in [neighborhood] with a 4.9 rating across 258 reviews."

So for five practices we added their five newest Google reviews ($0.017) and asked again, preferring anything about booking, scheduling or the front desk, since that's what the seller cares about. The difference:

"A recent review says your staff accommodated a fast-track appointment for emergency treatment."

"A patient said booking a last-minute appointment online was very easy."

"A patient praised your entire front office and clinical team for creating an atmosphere of ease and assurance."

We checked each line against its review: all of them are true to what the reviewer wrote.

With an API key in LOVELY_API_KEY and the run_tool and ask helpers from the market research case study:

places = run_tool("google-maps-leads", {"searchTerms": ["dentist"], "location": "Austin, TX, USA", "radius": 15, "maxResults": 100, "websiteFilter": "withWebsite"}, 100)

emails = sorted({e.strip().lower() for p in places for e in (p.get("emails") or [])})
checked = {r["email"]: r["status"] for r in run_tool("email-verify", {"emails": emails}, len(emails))}

leads, seen = [], set()
for p in places:
    good = [e.strip().lower() for e in (p.get("emails") or []) if checked.get(e.strip().lower()) == "good"]
    if good and good[0] not in seen:
        seen.add(good[0])
        leads.append({"id": len(leads), "name": p["title"], "email": good[0], "rating": p.get("rating"), "reviews": p.get("reviewCount"), "categories": p.get("categories")})

lines = ask("One sentence per practice using one true detail from its data. Return a JSON array of {id, line}.", leads, model="gpt-5.6-luna")

Change the search term and the city for any local market. For a step-by-step version that saves a CSV, see export Google Maps leads to CSV; for the tools behind it, lead generation.