How we measure whether AI search is actually working
AI search is new enough that a lot of vendors will happily sell you a dashboard full of numbers that mean nothing. Because Gainesville skews tech-literate — a lot of these households commute into DC's tech and government economy — buyers here can smell a vanity metric. So it's worth being plain about what we actually track and what we deliberately ignore.
The metric that matters is simple to state and hard to fake: when a real person asks ChatGPT, Google's AI Overviews, or Gemini a buying question in your category near Gainesville, does your business get named, and is what the machine says about you accurate? We build a set of the actual prompts your customers use — "who's a reliable HVAC company near Gainesville, VA," "best place for a same-day dentist in Prince William County" — and we test them on a schedule across the major assistants, logging whether you appear, in what position among the recommendations, and whether the model has your services, service area, and reputation right or is repeating something stale.
We also watch the plumbing underneath that outcome, because it's what we can actually move. That means whether the AI assistants and Google's AI layer are citing your site as a source, whether your business shows up correctly in the structured places these models lean on to build answers, and whether the specific facts they repeat about you — hours, coverage area, what you specialize in — are the ones we published. When those inputs are clean and consistent, the named-recommendation outcome follows.
Here's what we don't dress up as success. We ignore raw "AI is mentioning your industry" volume, because your category being discussed says nothing about you. We ignore impression-style counts that no assistant actually exposes and that any vendor quoting them is estimating or inventing. And we ignore a single lucky screenshot where you happened to appear once — one favorable answer is anecdote, not a ranking, so we care about how consistently you show up across repeated tests and different phrasings, not a cherry-picked win.
The honest catch is that this is a moving target: the models change, and no one — us included — can guarantee a given assistant will always recommend you. What we can do is measure the real outcome on a regular cadence, show you the log, and keep improving the inputs we control. That's the difference between optimizing for AI search and just talking about it, and it's the standard we hold the AI Search work to.