A 41 day experiment: JavaScript links make pages invisible to AI search
A 41 day experiment published on Search Engine Land shows that how links are built is decisive for AI crawlers. The major AI crawlers found none of the pages behind JavaScript injected links during the initial phase.
How the experiment was set up
In a study by Vinicius Stanula published on Search Engine Land on August 19, 2026, a 1,062 page Brazilian business directory was split in two: 11 sections had links hardcoded in HTML and 10 sections had links injected with JavaScript.
To isolate the variable, every other discovery path was disabled. Sitemaps, breadcrumbs and hierarchy panels were switched off, leaving the links themselves as the only route to the pages.
The experiment ran for 41 days, tracking which pages the crawlers reached.
The results and the recovery test
The most striking finding concerns AI crawlers. GPTBot, ClaudeBot, Bingbot, Meta-ExternalAgent, Amazonbot and PerplexityBot found exactly zero JavaScript linked pages during the initial phase.
The Google picture differs while staying low. Only Google crawlers executed JavaScript and followed injected links, and even so Googlebot reached just 2 percent of JavaScript linked pages against 5 percent of HTML linked pages.
User demand was not enough either: ChatGPT-User could not access JavaScript protected pages despite user demand.
The recovery test is telling. After converting links to HTML on day 27, GPTBot re-crawled within 48 hours and found 250 new pages, while Google took weeks to recover any coverage.
A link is an accessibility decision rather than a design choice
The clearest conclusion of this experiment: for AI crawlers a link arriving via JavaScript may as well not exist. The page is there, the content is there, but the route to it is invisible. That is an architecture problem rather than a content quality problem. That assessment is ours.
Second, the asymmetry in how fast a fix lands. GPTBot returned within 48 hours while Google took weeks. When the mistake is corrected the gain arrives quickly on the AI side, while recovering lost crawl budget on the search side is hard. The article recommendation points the same way: fix it sooner rather than later.
Third, the Googlebot numbers being low too. Coverage of 5 percent even on HTML linked pages is a reminder of how scarce crawl budget is. The issue is not only JavaScript but the sum of discovery paths.
Fourth, the limits of the experiment. This is one site, one vertical and 41 days of observation. The findings are strong but not a universal law. Verifying it on your own site means looking at your server logs. That note is ours.
What it means for businesses in Türkiye
The assessment below is not in the source, it is our reading. The study ran on a Brazilian directory and contains no Türkiye specific data.
The equivalent here is very concrete. A significant share of corporate and ecommerce sites build product listings, category navigation and pagination with JavaScript. Lists opened through infinite scroll and show more buttons in particular leave no link in the HTML.
The consequence can be this: your product pages may not appear in AI interfaces at all, and the cause is not your content but how your listings are built.
Three practical checks. First, open your category page with JavaScript disabled in the browser and see whether the product links are still visible. Second, examine your server logs for which pages GPTBot, ClaudeBot and PerplexityBot reach. If a section they never reach exists, the cause is most likely the link structure. Third, if you use infinite scroll, leave real HTML pagination links behind it.
It is also useful to read this alongside the robots.txt debate from the same period: on one side how crawlers interpret the rules, on the other whether you leave them a path at all. Two faces of the same problem.
The UNALSOFT take
We have worked to the same principle for years on the web design side: a link is the skeleton of the site rather than a design element. Infinite scroll looks good, but without real links behind it those pages exist only for human visitors. Discussions of visibility in AI interfaces usually talk about content, while as this experiment shows the route to the content has to be open before the content is written. That is why the first place we look with clients is navigation architecture rather than copy. The visibility work on the agentic AI side rests on the same foundation.
Sources
Search Engine Land, JavaScript links and visibility in AI search
Is the route to your pages open?
Let us examine your navigation architecture and bot logs together.