We sell a scan that measures whether you get found in AI answers. Then there is one question you have to ask yourself before you sell it to a client: how do we score ourselves?
On 28 August 2026 we put ai-werkers.nl through our own SitePulse.
55 percent. A C.
The eight sub-scores
| Element | Score |
|---|---|
| Visibility in AI answers | 90% |
| Technology | 86% |
| Conversational | 84% |
| E-E-A-T | 57% |
| Information gain | 55% |
| Brand footprint | 26% |
| Citability | 22% |
| Anti-slop | 20% |
That is a pleasant result to get, and I mean that.
The three highest scores are about who we are and whether the story holds up. Models know us, the technology is in order, and our texts answer real questions in plain language. That is the part you do not repair in an afternoon.
The three lowest are about form. How the texts are written and structured. That you can repair in a few weeks.
What was concretely wrong
Citability, 22 percent. Our texts were a running story. An AI prefers to pick out a single paragraph that holds up on its own, and with us every sentence needed the previous one. Useless to quote.
Brand footprint, 26 percent. There was an llms.txt that named three products while we have six. There was schema markup with invented social links in it. And there were zero reviews, while the tool to show them was already installed and simply not connected.
Anti-slop, 20 percent. Generic AI phrases. I have written a separate piece about that, because it is the most embarrassing of the three.
And one that is not among the eight scores but did come out of the report: of the fifteen test questions a customer would ask, zero produced a mention. Do they know us by name? Yes. Do they name us when someone searches without our name? No.
The nineteen contradictions
Alongside the scan we went through our own site page by page. We found nineteen places where we contradicted ourselves.
Our target group appeared on one page as 10 to 500 staff and elsewhere as 5 to 250. The WerktScan was called four different things in four places. The year we were founded differed per page. Our own company name appeared in five spellings.
For a visitor that is sloppy. For an AI crawler it is worse: it pulls facts from several pages at once and gets uncertain as soon as they clash. That is exactly why that brand footprint sat at 26.
What we have done about it
The site has been rebuilt. Static, no CMS, no plugins. Every fact is now recorded one way, in one place, and carried through identically everywhere. Every FAQ question answers itself in one to three sentences. Every figure on the site has a source, and it says whether it was measured or calculated.
And the llms.txt is now correct.
What comes next
In three months we run the same scan again and publish the grade, whether it is good news or not. Realistically we are aiming for above 75 percent, with citability, brand footprint and anti-slop each above 60.
That will be part two of this piece.
An agency that does not dare show its own grade probably has a reason. Run your own scan and see where you stand.