What we find while measuring.
Methodology arguments, South African engine behaviour, and the parts of the GEO literature that did not survive replication. Including the findings that cost us a feature bullet.
Language is a third of the variance — and South Africa has eleven
Afrikaans and isiZulu are live in Google's AI surfaces. Almost nobody is measuring them.
2026-07-23 · 6 minWhat a tracker can and cannot see about AI Overviews in South Africa
Best-in-class detection is about 68%. That third matters more than the two thirds.
Your CDN may be hiding your client from every AI engine
Bot-fighting defaults now block the crawlers you want. It is the most common silent failure we find.
2026-07-16 · 5 min84% of AI citations are not your client's website
The citation table is the most misread screen in the product. Here is how to read it.
2026-07-09 · 7 minNine questions to ask an AI-visibility vendor before you resell it
You are putting your logo on their numbers. These are the questions that separate an instrument from a dashboard.
Disagree with one of these?
Several of these posts argue against things we would rather were true. If you have data that contradicts one, we would genuinely like to see it — corrections get published with the correction.