Wonmetrics.com: The same content, eight times as much for AI
The site was not in bad shape. It was simply built in a way that meant almost none of it reached an AI crawler.
Case Study
Company
Wonmetrics — one of AddonNordic's own brands. An English-language site, built as a JavaScript app behind Cloudflare.
Industry: Software
Measurement period: 25–27 August 2026
Domain: wonmetrics.com
The challenge
Measured on 25 August 2026, GPTBot got 1,016 characters of visible text out of the front page, with no structured data, and llms.txt answered 404.
The bar is over 3,000 characters, structured data present and an llms.txt that answers. The site fell short on all three.
What Geoa found
The text was built in the browser
The content was there, but only after JavaScript had run. The crawler never got that far and read a fraction of it.
No structured data
No JSON-LD tied the site to a company, so an AI had no way of working out the sender.
The sitemap was not found
The sitemap did not answer as expected, and the indexing coverage therefore stood at zero pages — not because there were no pages.
How it was fixed
Server-rendered HTML
The site was moved so the text sits in the response itself. <em>We made the move ourselves — Geoa measured before and after.</em>
Structured data added
Full JSON-LD on the front page, so the company can be resolved and credited for a mention.
The sitemap fixed
After the fix the platform found the sitemap and 11 pages, where before it showed zero.
Measured again, same method
The after figure was taken with the same command as the before figure, two days later.
Results
characters of visible text for GPTBot (25 August 2026)
characters of visible text + full JSON-LD (27 August 2026)
pages found in the indexing coverage after the sitemap fix
What kept coming back:
- The way it is built decides it, not the amount of content. The same text, eight times as much of it visible
- A sitemap that does not answer as expected looks like a site with no pages — including in our own measurement
- The fault is invisible in a browser and therefore stands for a long time if nobody measures with the crawler's signature
- The fix took two days, not two quarters — once you know what you are looking for
“We thought the site was fine because it looked fine. The difference between 1,016 and 7,986 characters is not more content — it is the same content, delivered somewhere the machine can get hold of it.”
What was used
- AI readability check
- Structured data (JSON-LD)
- Indexing coverage
- GEO Score
Related cases
EntityWatch.app: 21 characters was all AI could read
The page looked entirely right in a browser. To the crawler ChatGPT fetches pages with, the whole front page consisted of two words.
Geoa.app: From B to A in 45 days
A site built by people who make their living from AI visibility ought to have nothing left to fix. It did. 11 measurements later, the score was 5.3 points higher.
Cases: what Geoa found, and what the numbers did afterwards →
How much of your site can AI read?
Run the free GEO test and see your score across 15 technical dimensions — including how much text an AI crawler actually gets out of your front page.