Why can't AI assistants read my website?

Because some websites only exist once your browser has done some work, and many of the things that read the web on behalf of AI assistants do not do that work.

Here is a real case. In July 2026 I measured a martial arts academy in Zürich. A good site: clear schedule, prices, five membership options, a beginner message repeated three times on the homepage. To a person in a browser it was complete.

I then fetched the same homepage the way a simple crawler would, without running any of the page's code. It returned 955 bytes: the page title and one line of description. About 150 characters of readable text. None of the classes, none of the schedule, none of the prices. Everything a person could see was being drawn in by JavaScript after the page loaded, and a reader that does not run JavaScript never saw any of it.

The assistants' answers matched. With web search switched off, none of the four assistants named the academy in any of 180 answers. With search switched on they named it sometimes, when what they found elsewhere, directories, reviews, a map listing, gave them enough to go on. The site itself was contributing almost nothing.

One caution on what this shows. Which assistants run JavaScript when they fetch a page is not something I measured. What I can say is that the page served 955 bytes to a reader without it, and the answers behaved as if the site were empty.

The fix was unglamorous. The site was rebuilt so that the content is present in the page as delivered, before any code runs. The same homepage now returns about 70,000 bytes of readable text to the same simple fetch. The sitemap went from 2 addresses to 22. A human visitor sees the same design as before.

I re-measured on 1 September, after the rebuild. Across the questions that named no academy, the assistants named it in 87 per cent of English answers, up from 47 per cent in July, and in 93 per cent of German answers, up from 57. The assistant that searches the web and had been unable to read the site went from naming it in 19 per cent of English answers to 94. The question about MMA, answered with the academy's name in none of 24 answers in July, was answered that way in 9 of 12. One caution that belongs with those figures: the website was not the only thing that changed between the two measurements. The summer break ended, the timetable grew, and the directory listings were worked on. The measurement shows what moved; it cannot say which change moved it. The owner reports enquiries rising from roughly 10 a month to roughly 40 over the same period.

How to check your own site in a minute. Open your homepage in Chrome. Press Ctrl+U (Cmd+Option+U on a Mac) to view the source. Search the page for your own words: a service name, a price, your opening hours. If your content is there in plain text, you do not have this problem. If the source is a few short lines and your words are nowhere in it, you do.