How we measure

A score of 43 or 78 means nothing if you do not know how it came about. So here is exactly what we do, what the numbers mean, and what we cannot measure.

What happens when you run a check

We fetch your page the way an ordinary browser does, and look at the source an AI crawler gets to see. That matters, because what you see in your browser is often more than what a crawler sees.

We also look from your home page for up to five other relevant pages, such as your product or category pages. Those count when we judge your product information, because that is rarely on a home page.

If fetching fails the first time, we try again. If we get a security check instead of your real page, we give a notice rather than a score. A wrong number is worse than no number.

The thirteen points

We check thirteen things, in four groups. Points with high impact count twice as heavily as points with average impact.

How the score is calculated

Every point gets a number from 0 to 100. Points with high impact count double. The final score is the weighted average of those. Points we could not measure are left out of the sum instead of counted as a zero.

Each point carries one of these three labels:

Why the same store sometimes gets a different score

Two checks of the same page should give the same result, and they do. If you still get a different number, something else is usually going on.

The most common reason is that a different address was checked. The home page of a store and a category page are different pages, built differently, so they score differently. That is why your report always states exactly which address we looked at.

Your site may also have changed, or been temporarily slower. To stop a one-off spike from moving your score, we measure load time several times and use the fastest.

How we measure whether AI names you

With a subscription we do more than check your site. Every month we put buying questions to AI assistants and see whether your store is in the answer. Right now those are ChatGPT and Gemini. This is how that works.

1. We look at what your store sells and write thirty questions buyers really ask about it, spread over six kinds: general, price, audience, alternative to a brand, which store, and practical questions.
2. The name of your store does not appear in those questions. Otherwise we would be measuring whether AI can repeat a name instead of whether you are named on your own.
3. Every question goes to the AI models we can query reliably at that moment. We keep the full answer.
4. Then we read those answers and take from them which stores are named, whether you are among them, in which place, and whether you are only named or actually recommended.

Named is not the same as recommended

Many answers first give a long list and then close with advice: "for this look I would mainly look at X and Y". Being in that list is named. Being in that closing advice is recommended. That difference is large, so we count the two separately. With every mention we include the quote from the answer, so you can see for yourself what our judgement rests on.

Why you do not get a percentage

You get counts: named in so many of so many questions. No percentages, because those suggest a precision that is not there. We also only count the questions in which a store could be named at all. If someone asks about a brand or a product, no store appears in the answer, not even the best one. Counting those questions would make your number prettier or uglier than it is.

Which models we ask

Today those are ChatGPT and Gemini. We want more, and the technical side is ready, but we only add a model once it answers reliably. A model that joins one month and not the next makes your numbers incomparable, and that helps nobody. Your own page always states which models actually answered that month.

We measure through the programming interface

We ask the questions through the interface developers use, with nothing built around it. That is an approximation of what someone would see in the app, not a copy of it: in the app the search results of that moment, the location and that person's history all weigh in too. So we do not claim this is exactly what a buyer sees.

This is not fully in your hands

A technical score is about your own site and you control that. Mentions are not. If a competitor makes the news or an AI model is updated, your number can change without you doing anything wrong. That is why we always show who else was named, and state with every change whether it was you or the market. Without that, a number is only confusing.

What we do with your data

We keep the results of your checks so you can look back at them and see your history. We sell nothing on and do not use your data for advertising. The privacy policy states exactly what we keep and for how long.