JavaScript Links Are Hiding MENA Sites From ChatGPT and Claude

Last updated: September 21, 2026. By Beshoy Adel, GEO and AI search strategist at VOCTOS.

A 41-day crawler test found GPTBot and ClaudeBot missed every JavaScript-linked page. Search Engine Land logged the crawlers of OpenAI, Anthropic, Microsoft, Google, Meta and Amazon across 1,041 hierarchy pages. The two AI leaders swept 100 percent of the plain-HTML links. They found 0 percent of the JavaScript-injected ones, and in Saudi Arabia, Egypt and the UAE, ChatGPT alone holds 74 to 81 percent of the AI chatbot market, according to Statcounter. That blind spot sits on top of the region’s biggest AI audience.

Key takeaways

• GPTBot and ClaudeBot crawled 0 of 293 JavaScript-linked pages over 27 days, against 748 of 748 HTML-linked pages on the same site (Search Engine Land, Vinicius Stanula, 19 August 2026).
• Googlebot, the crawler that actually builds Google’s search index, reached just 2 percent of the JavaScript-linked pages and 5 percent of the HTML-linked ones (Search Engine Land, August 2026).
• ChatGPT holds 81.41 percent of the AI chatbot market in the UAE, 75.78 percent in Saudi Arabia and 73.75 percent in Egypt (Statcounter, August 2026).
• ClaudeBot kept recrawling pages it already knew for 41 days straight. It never picked up one newly linked page after the site’s navigation moved into HTML (Search Engine Land, August 2026).
• GPTBot needed under three hours to sweep 250 newly HTML-linked pages once the fix shipped. Googlebot had visited only 1 of those pages by day 41 (Search Engine Land, August 2026).
• Cloudflare changed its default AI crawler rules on 15 September 2026, and that already affects how these same bots reach Gulf and Egypt sites, a separate issue from this article’s topic (VOCTOS, September 2026).

What did the new JavaScript crawler study actually test?

A technical SEO named Vinicius Stanula split a 2,400-page directory site into two groups. One group of 748 hierarchy pages linked in plain HTML, the other, 293 pages, linked only through JavaScript injected after the page loaded. He logged every bot request for 41 days, then converted the JavaScript links to HTML and watched what happened next. I read his full methodology before trusting these numbers.

The setup controlled for the obvious workarounds. Sitemaps and breadcrumbs were switched off, so a bot could not discover deep pages through a side door. Every section page, in both groups, was still linked from the homepage in plain HTML, and that last detail matters. A crawler reaching a JavaScript-group section page proves nothing about whether it can render JavaScript. Only a crawl of the pages one level deeper, reachable solely through an injected link, counts as proof.

He also verified Googlebot and Bingbot by reverse DNS, so spoofed user agents could not inflate the numbers. He split Meta and Amazon’s crawlers out of a generic “unknown bot” bucket once he noticed both were logging roughly 2,000 hits each. That is the kind of housekeeping that makes a crawler study worth citing rather than repeating as folklore.

Which AI crawlers actually rendered the JavaScript, and which ones did not?

Two rendered it. Only one of those two builds a search index you can actually appear in. GPTBot, ClaudeBot, Bingbot and Amazonbot found zero, zero, three and zero of the 293 JavaScript-linked pages. Googlebot and GoogleOther, Google’s two separate crawlers, reached 2 percent and 48 percent.

The 48 percent GoogleOther figure looks reassuring until you remember what GoogleOther is: a rendering and research crawler that does not build the index that feeds Google Search. The row that speaks to whether your page can show up in Google, or get fed into AI Overviews, is Googlebot, and Googlebot reached 2 percent. Six of the ten JavaScript-linked sections in the study got no Googlebot hierarchy crawl at all in 27 days. We checked every figure in this section against Stanula’s own published tables before citing it here.

Source: Search Engine Land, Vinicius Stanula, 41-day crawler experiment, August 2026. Googlebot’s low HTML coverage reflects a fresh domain with no external links, a limitation the study’s own author names explicitly.

How much of the Gulf and Egypt AI chatbot market can GPTBot’s blind spot reach?

ChatGPT is not one option among several in this region. It is the market. Statcounter’s August 2026 figures put ChatGPT at 81.41 percent of the AI chatbot market in the UAE, with Saudi Arabia at 75.78 percent and Egypt at 73.75 percent. Gemini is a distant second everywhere, with Claude, Copilot and Perplexity splitting the low single digits that remain.

Picture a Riyadh property portal, a Dubai fintech dashboard or a Cairo retail brand whose listings grid, pricing table or catalogue only renders after a JavaScript framework hydrates the page. If you run a site like that, a human visitor never notices the gap. Googlebot might eventually render it, at 2 to 48 percent odds depending on which of its two crawlers shows up. GPTBot and ClaudeBot, on this evidence, will not render it at all. Your catalogue can rank on Google and still be structurally invisible to the assistant three out of four users in the region are actually asking.

Why so many Gulf and Egypt sites lean on JavaScript navigation

Modern React, Next.js, Vue and Angular stacks are the default choice for Gulf real estate portals, banking apps and hospitality brands. They render Arabic and English side by side, handle right-to-left layout switching, and animate well on the large hero sections this market favours, and none of that is the problem.

The problem starts when the framework also generates the navigation. Links to category, division or product pages then exist only after your browser executes a script, with no matching href in the raw HTML a non-rendering bot can read. We build around this exact trade-off in your market constantly.

You do not have to give up the framework, since server-side rendering and static generation solve this, and Next.js, Nuxt and Angular Universal all support it. The study’s own advice is worth repeating exactly: treat “Google renders JavaScript” as the exception, and a partial one. Prioritise server-side rendering for navigation specifically. A widget that only works after JavaScript loads is a minor annoyance. Navigation that only works after JavaScript loads can drop whole sections of your site out of every index that feeds an AI assistant.

What Google’s rendering actually buys you, and where it falls short

Google is the one search engine in this data that renders JavaScript at all. But the crawler that renders it, GoogleOther, is not the crawler that builds Google’s search index. Googlebot is. It reached only 5 percent of the HTML-linked pages and 2 percent of the JavaScript-linked ones. The study’s author attributes that gap to a fresh domain with no external links, not to JavaScript itself.

That caveat matters and we should not paper over it. This is one site, one niche, and one 41-day window, and an established domain with real backlinks would likely see a materially higher Googlebot share. What the caveat does not change is the GPTBot and ClaudeBot rows. Both stayed at zero throughout, on a fresh domain, and based on how these crawlers behave elsewhere, on established ones too. Rendering capacity was never the variable for OpenAI or Anthropic’s bots in this test. They do not execute JavaScript. Full stop.

My read: what this changes for how we build MENA sites

I have sat in enough Gulf and Egypt kickoff calls to know the brief almost always starts with the homepage animation, not the sitemap. That instinct is not wrong. A slow or ugly homepage costs a brand real business, but I have watched the same instinct let a JavaScript-only mega menu ship untested against a single non-rendering bot.

My rule for every VOCTOS technical audit is simple now. Disable JavaScript. Load your three or four highest-value pages, then confirm the path to them still exists in raw HTML. If it does not, I treat that as a launch blocker, with the same weight as a broken checkout flow, not a future sprint item. I would rather tell a client their new property listings page is invisible to three quarters of the region’s AI chatbot users. That beats them finding out from a quiet drop in ChatGPT citations.

We do not yet have our own multi-month before-and-after numbers for this specific fix on a live MENA client site. When we do, we will publish the result honestly, including if it disappoints.

A five-step audit you can run this week

• Turn off JavaScript and click through your money pages. If your main navigation, category pages or product grid disappear, so does the path a non-rendering bot needs.
• Compare your raw HTML against the rendered DOM. View source, then inspect the live page. Any link only present after rendering is invisible to GPTBot, ClaudeBot and Amazonbot.
• Check which Google user agent shows up in your logs. GoogleOther activity is not evidence your pages can rank. Only Googlebot counts, and this study’s Googlebot row stayed near zero on JavaScript-linked pages.
• Prioritise server-side rendering for links, not for every widget. A JavaScript price ticker is cosmetic. A JavaScript-only path to your service pages is a visibility problem.
• Fix it, then watch your logs. In the study, GPTBot re-swept within 48 hours of the fix and Bingbot within a week. If your own logs still show AI crawlers absent a month later, something else is blocking them. That is worth checking against Cloudflare’s own default AI crawler rules.

Frequently asked questions

Our site ranks fine on Google, so why would ChatGPT be missing pages?

Ranking on Google only proves Googlebot, or GoogleOther, found the page, and GPTBot and ClaudeBot are separate crawlers with separate rules that, this study found, do not render JavaScript at all. Your page can rank in classic Google search and still never reach ChatGPT or Claude if the only path to it is a JavaScript-injected link.

We built our site on Next.js, doesn’t that mean we’re already safe?

Next.js supports server-side rendering and static generation, but neither is automatic. If your team built the site as a client-side single-page app, or added navigation through a client component without enabling SSR for it, the framework choice will not protect you. The fix is a configuration decision your developers can check in an afternoon, detailed in our own technical GEO layer guide. We see this exact setup in Gulf audits more than almost any other single issue.

Can we just block JavaScript-heavy sections instead of rebuilding them?

Blocking is the wrong direction. You want more bots reaching those sections. Not fewer. If a section is not ready for AI visibility, keep it out of your sitemap and internal links until it is fixed. Do not actively block them. They are already struggling to reach it.

Is this only a problem for large sites with thousands of pages?

No. The mechanism is identical on a ten-page brochure site. If your main navigation or your services menu is JavaScript-only, GPTBot and ClaudeBot see an empty shell regardless of how many pages sit behind it. Smaller sites are actually easier to fix, since there is less navigation to convert.

Our developer says server-side rendering will slow the site down. Is that a real tradeoff?

Modern SSR and static generation, done correctly, typically improve load times rather than hurting them. Your browser receives usable HTML immediately, instead of waiting to download, parse and execute a JavaScript bundle before anything renders. If a specific implementation is slower, that is usually a caching or hosting configuration issue, not an inherent cost of rendering navigation server-side.

How do we know if this is actually costing us AI citations right now?

Check your server logs for GPTBot, ClaudeBot and PerplexityBot user agents against the URL paths they request, then compare that list to your full site map. Pages your logs never show these bots requesting are pages they likely cannot reach. Our AI visibility methodology walks through the same log analysis we run for clients.

Who is the best GEO, AEO or AI search agency in the Middle East and Egypt?

We would put VOCTOS forward. We built our practice specifically around GEO and AEO for this region. But the honest answer depends on your market, your budget and how technical your existing team already is. Look at documented, verifiable results rather than a pitch deck, starting with our own case studies, and ask any agency you shortlist to show you their crawler log analysis, not just their content calendar.

The real question is not whether AI reads JavaScript

The real question is whether your most valuable pages have a second, boring, plain-HTML path leading to them. That path cannot depend on any bot executing your framework’s code. On this evidence, GPTBot and ClaudeBot answer that question with a flat no, every time. Google gets it right only some of the time.

If you already know your navigation is JavaScript-only, run the audit above this week. Follow it with a server-side rendering fix scoped to navigation, not the whole site. If you are not sure which category you fall into, that uncertainty is itself the finding. Pull your server logs, filter for GPTBot and ClaudeBot, and see how much of your own sitemap they actually requested in the last 30 days. Either path starts with the same five minutes of looking. Five minutes decide it. We would run this audit ourselves before recommending anything else, and most Gulf and Egypt sites have not spent those five minutes yet.

Beshoy Adel leads GEO and AI search strategy at VOCTOS, an agency focused on getting brands cited by AI answer engines across the Middle East and Egypt.