Answer engine optimization · Original data
How ChatGPT chooses which businesses to recommend
Answer Engine Optimization has become one of those phrases everyone in local marketing parrots without nailing down what it actually looks like in practice. The guidance drifting around stays stubbornly vague: establish authority, earn mentions, be useful. None of it gets to what a business owner actually needs to understand: why ChatGPT names one competitor and completely ignores another when someone nearby asks for a recommendation.
We can answer that with real data, because we run the same probe our own product uses for clients. Across ten local service categories and five cities, we tracked 1,423 actual AI citations, meaning every domain ChatGPT pointed to when answering a local search question. Then we grouped those citations by the question type that generated them, and the pattern that emerged explains most of what people are wrapping up as AEO.
What the numbers show as questions get narrower
Broad questions like "best chiropractor in Denver" pulled directory sources 96% of the time. Out of 281 citations tied to that kind of query, 271 came from known directories, review sites, or aggregators, and the total spread across only 125 distinct domains. Narrow it to a specific need, something like "where can I get botox in Denver," and directories plummet to 22% of citations. The pool of distinct domains leaps to 305. Push further into specificity, a symptom question like "I have acne scars, where can I get them treated," and directories collapse to 14%, spread across 321 domains. Ask a booking-style question and directories account for a mere 11% of sources.
That's not a minor drift. It's the gap between a model leaning on a handful of aggregator sites it already trusts and a model forced to hunt down the one page that actually answers what was asked.
Why narrow questions open the door to your own site
Here's why that plays out. A broad question like "best med spa in the city" has no single obvious answer page anywhere online. No individual business publishes a page ranking every med spa in town, so the model defaults to pages that already did that labor: directories, review aggregators, local guides. A narrow question sends the model somewhere entirely different. "Best botox in this neighborhood" or "where can I get acne scars treated" points to a far smaller set of candidate pages, and if one of them happens to be written by the business itself, on its own site, answering that precise question, the model can grab it directly.
What we watched happen live this week
We watched this unfold in real time this week while testing a script for one of our own videos. We ran three versions of the same underlying question through ChatGPT, each narrower than the last. "Best med spa in Miami" returned five business cards backed by two directory sources, Discover Med Spa and a listings site called Medical Spa Locator. Not a single clinic website appeared as a source. Narrowing to "best med spa in Brickell," a specific Miami neighborhood, swapped in a different pair of sources, Tripadvisor and a local guide, and four of the five businesses named were different from the first answer. Only one business, LUX MedSpa Brickell, carried across both answers, which tells you it's sitting in more than one of these source layers simultaneously. Then we asked "best botox in Brickell," pairing the treatment with the neighborhood, and for the first time in the entire test, ChatGPT pulled a source straight from a clinic's own website. It named the page as its reason: the clinic had a page specifically about Botox in Brickell, a page that answered the exact question we asked.
That single live example is a small sample on its own, but it lines up precisely with what the 1,423-citation dataset shows at scale. The broader the question, the more the model depends on pages that already aggregate an answer for it. The narrower the question, the more room there is for a business's own page to be what the model finds and uses.
What this means for your pages
What this means for a local business owner is far more actionable than "get mentioned more." Chasing placement in every directory in your category is a legitimate strategy, but it's the strategy for winning broad questions, and broad questions are dominated by a small handful of sites you don't control. The narrow questions are the ones you can win outright, and they're also the ones your actual customers ask. Nobody walks into a med spa asking for "the best med spa." They're asking about a specific treatment, in a specific neighborhood, sometimes with a specific concern attached. Those are the pages worth writing: one page per treatment, one page per neighborhood you serve, written with enough detail that it's the obvious answer to that one narrow question and not a rehashed version of your homepage.
The five-minute check
The check for your own business takes five minutes. Ask ChatGPT the broad version of your category question and examine the sources. You'll likely see directories, and that's fine, that's the layer you compete for through review counts and directory profiles. Then ask the narrow version, the one your actual customer would type, with your specific service and your specific area attached. If your own site doesn't surface there, the fix usually isn't a better homepage. It's a page that answers that one narrow question directly, because that's the page the model is actually hunting for.