Right now: we’ll set up your AI employee for you — free of charge (worth €249)Find out how →
Case Study

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.

Case Study

Company

EntityWatch — one of AddonNordic's own brands, built as a JavaScript app (Lovable) with the content inserted by the browser.

Industry: Software
Measurement period: 13–25 August 2026
Domain: entitywatch.app

The challenge

A crawler does not run JavaScript. It fetches HTML and reads what is there. If the content is only inserted after the browser has run the code, the crawler sees an empty shell.

Measured on 13 August 2026 with curl -A "GPTBot": 21 characters of visible text. The text was, word for word, “Lovable App”. That was the entire front page as AI saw it.

What Geoa found

The front page was empty to machines

The human eye saw a finished site. The crawler got 21 characters. No browser test can show that difference — only a measurement made with the crawler's own signature.

No structured data

Without JSON-LD an AI cannot work out which company the page is about, or tie a mention to the right business.

No llms.txt

The file AI models look for first did not exist. There was neither a description of the site nor a pointer to the most important pages.

How it was fixed

1

Away from the JavaScript shell

The site was moved to server-rendered HTML, so the text sits in the response. <em>We made the move ourselves — Geoa measured it.</em>

2

The entity declared

JSON-LD was put in place, so the name, company and address can be resolved to a single business.

3

llms.txt created

The file now answers 200 and tells the AI models what the site is about and where to read on.

4

Measured again with the same command

The after figure was taken with exactly the same method as the before figure. Otherwise the two could not be compared.

Results

Both measurements are <code>curl -A "GPTBot"</code> on the front page, visible text after tag stripping. Same command, same page, twelve days apart.
21

characters of visible text for GPTBot (13 August 2026)

1,221

characters of visible text for GPTBot (25 August 2026)

58×

more text for AI to read — same site, same content

What you can do yourself in two minutes:

  1. Run the command: curl -s -A "GPTBot" https://your-domain.com and see how much text there actually is in the response
  2. Under 3,000 characters? Then there is a good chance the page is being built by the browser and not by the server
  3. It hits the whole family of JavaScript sites — not one tool, but a way of building
  4. And it is invisible in a browser. That is why it is the fault that gets to stand the longest

“The worst thing about 21 characters is that nothing drew attention to it. The site worked. The analytics worked. The only place it showed up was in a response from our own crawler measurement.”

— Martin Nymann, founder of AddonNordic — entitywatch.app is our own brand, not a customer
What was used

What was used

  • AI readability check
  • Structured data (JSON-LD)
  • llms.txt
  • GEO Score

What can AI read on your site?

Run the AI readability check on your own domain and see exactly the text an AI crawler gets out of your front page. Free, no sign-up.

Check your GEO Score free → Get started with Geoa

Get the Nordic AI visibility figures before they become common knowledge

One letter a week: what we’ve measured at Nordic companies, what changed in ChatGPT, Perplexity and Google AI Overviews, and what you can do about it. No reposting of other people’s blog posts.

How we process your email