How ChatGPT Picks Which Local Business to Recommend

To close out this series, let's stop talking about tactics and follow a single question all the way through. Someone in Winter Springs has a leak, opens ChatGPT, and types: "Who's a good roofer in Winter Springs FL?" Here is roughly what happens in the next few seconds, and where a roofing company either gets named or doesn't.

Step One: The Engine Decides It Needs to Search

A question about a specific local service is one the model can't answer from memory alone, so it runs live web searches: several variations of the question plus the location. Whatever ranks for those searches becomes the candidate pool. This is the point where classic SEO still matters. If your site and your listings don't appear in the first pages for the plain-language version of the question, you're not in the pool, and nothing after this can save you.

Step Two: It Reads What It Found

The engine fetches the top results and extracts what it needs. Directory pages give it names, categories, and ratings. Review sites give it sentiment and specifics. Local news and "best of" lists give it corroboration. Your own site gives it your claims, but only if the content is in the HTML, loads quickly, and says in plain words that you replace roofs in Winter Springs. A slow page, a JavaScript-rendered page, or a page that says "solutions for every home" contributes nothing, and the engine moves on to the next candidate without a second thought. We covered this earlier, and it's the step where the most businesses silently fall out.

Step Three: It Resolves Entities

Now it has a pile of mentions and needs to turn them into a short list of distinct businesses. "Smith Roofing," "Smith Roofing LLC," and "Smith's Roofing & Repair" might be one company or three. If the name, address, and phone match across the sources, they collapse into one confident entity with all the evidence stacked on it. If they don't, the evidence gets split, and a business that should have been the obvious pick ends up looking like three weaker ones. This is the consistency problem, and it's invisible to the owner, who knows perfectly well it's all the same company.

Step Four: It Weighs Evidence

For each entity, the engine is effectively asking: how many independent sources say this business does this thing in this place, and how specific are they? A company with forty reviews that mention roof replacement by name, a listing in the local news "best of" roundup, a complete Google Business Profile with roofing as the primary category, and structured data on its site that says the same thing is going to outscore a company with a prettier website and a hundred "great job, five stars" reviews. Corroboration beats self-description, every time.

Step Five: It Writes the Answer

The engine composes a reply naming two to four businesses, each with a sentence or two of description, and cites the sources it leaned on. The description is assembled from the most quotable passages it found. If your site has a sentence that says "We replace shingle, tile, and metal roofs across Seminole County, with most residential jobs finished in two days," that sentence, or something very close to it, becomes your description. If the best it could find was a directory scrape, that's what the prospect reads.

Where This Leaves You

Five steps, five places to lose. Not in the pool: fix findability. Unreadable: fix the site. Split identity: fix consistency. Thin evidence: earn corroboration. Nothing quotable: rewrite for extraction. None of these is mysterious and none requires a large budget. They require doing unglamorous things correctly and in order, which is the same thing that has always separated the businesses that get recommended from the ones that wonder why they don't. If you want a second set of eyes on where you're falling out, that's a 30-minute call, and we'll tell you which step it is.