A plain comparison. The same job, done two very different ways.
If you want to know whether AI answer engines are using your website, there are two fundamentally different ways a tool can find out. They are not two versions of the same thing — they answer different questions and produce different kinds of numbers. This page lays both out plainly so you can see the difference and decide which one you actually need.
No brand names. The rows are what matters — what each approach actually delivers.
| The sampling approach | The counting approach | |
|---|---|---|
| Where the number comes from | A model is asked a set of invented questions, repeatedly, and brand mentions are tallied. | Your own server records. Every figure is a count of a real retrieval by a real engine answering a real person. |
| Does the number stay the same if you re-run it? | Not necessarily — a model can answer the same question differently each time it is asked. | Yes. A count of events that happened does not change when you look at it again. |
| Engines shown separately, or blended? | Often blended into a single overall score. | Each engine on its own. The major answer surfaces are reported individually, never averaged together. |
| Detail per page? | Usually a summary or a single score for the whole site. | Page by page. Which specific pages were used, by which engine, and how often. |
| Does it show whether anyone was actually sent to you? | No — a mention in an answer is not a visit, and the two are not distinguished. | Yes. It separates being used from sending someone — including pages an engine leaned on heavily while sending no one at all. |
| Does it tell you how sure it is? | A single clean number, with its uncertainty not shown. | Every figure is labelled — confirmed, or an estimate to be read as a floor. It states where it is solid and where it is still settling. |
| What does it cost to run? | Typically an ongoing monthly subscription. | A one-time licence. No subscription. |
| Where does your data go? | Your site and prompts are processed on someone else's servers. | Nowhere. It runs on your own machine. Nothing about your site is uploaded or transmitted. |
| Does it tell you what to do next? | Some bundle content generation and sell you the fix. | It diagnoses; you decide. It points at the exact pages that need attention and stops there. The diagnosis is not a funnel to something it is selling. |
If you want a rough, directional sense of whether a model tends to mention you, the sampling approach gives you that, quickly, for a monthly fee. It is an estimate, and it is honest to treat it as one.
If you want to know — provably, from your own records — which of your pages the answer engines are actually using, how often, for which engine, and whether any of it sends you a single human being, that is a different question, and it needs the counting approach. It is the difference between a survey and a census.
This comparison was written by the people who built a tool on the counting approach — because after watching the sampling tools produce numbers that moved on their own and couldn't be checked, we wanted one that counts from the record instead.
It reads your own server records, on your own machine, and shows you which pages the answer engines actually retrieved — page by page, engine by engine, each figure labelled with how sure it is. Nothing about your site leaves your computer.
See the tool, counting live from its own logs for a plain account of exactly what it can and cannot do.