Gemini pulls about half its citations from a business's own website. ChatGPT pulls about half from third-party sites. If you only optimize one, you're invisible to the other.
When someone asks an AI engine "who's the best roofer in Winchester, VA," the engine doesn't have an opinion. It retrieves sources, weighs them, and writes an answer with citations. Which sources it retrieves depends on which engine you're asking, and that has a practical consequence for any local business trying to be the answer.
Google's engines already have your website indexed and your Business Profile in hand, so they lean on both. That means your service pages, your FAQs, your schema, and the consistency between your site and your profile do most of the work. A site with a page for each service in each market, marked up properly, is what gets quoted.
ChatGPT's search uses third-party providers and cites what it finds, and roughly half of what it finds about a local business lives off your site: review platforms, directories, local news, community forums. Consistent name, address, and phone across those sources, a steady flow of reviews, and mentions on the sites your customers actually read matter as much as your own pages.
Among people who use AI for recommendations, 63% trust the answer, and 88% still verify it by checking reviews or sources. So the AI answer opens the door and your reviews walk the customer through it. A business with a great site and thin reviews gets named and then abandoned. A business with great reviews and a thin site rarely gets named at all.
Both. Structured, quotable pages and a complete Business Profile for the Google side; consistent citations and review velocity for the ChatGPT side; and monthly share-of-answer tracking so you can see which engines name you and how often. NVB Contracting's Winchester office is the first name ChatGPT gives for "best roofers in Winchester, VA," and it took both halves of that work to get there.
Sources: Cheers GEO Academy citation analysis; BrightLocal AI Trust Research (March 2026).
New clickstream data shows two out of three U.S. Google searches end without a click, and AI Overviews cut click-through by nearly 60% when they appear. The customer didn't disappear. The click did.
Cost per lead is the wrong number. Here's the cost per booked job for the three ways a roofer buys leads, with 2026 benchmarks and real client accounts.
A lead answered within an hour books at 31%. After 24 hours it's 15%. Here's what that means during storm season, and how automation closes the gap without hiring.
A free, 30-minute growth audit: your Local Services position, your share of AI answers, and what your competitors pay per lead. You keep the plan either way.
