No. None of the major AI crawlers — GPTBot, ClaudeBot, PerplexityBot, Meta's and Amazon's bots — execute JavaScript. They read raw HTML, and whatever the server sends is all they will ever see. We audited 800 of the web's top domains to measure what that costs: 1 in 15 serves them nothing at all.
What we measured
On 25 August 2026 we fetched the homepage of the top 800 apex domains from the Tranco research ranking twice each, using the QuanticData SEO audit API: once as a plain HTTP client that does not execute JavaScript — the exact view an AI crawler gets — and once fully rendered in a real browser. For every site we compared word counts, titles, canonicals and whether any readable content exists at all in each view. 374 domains answered HTTP 200 on both fetches and form the study sample; the rest are DNS infrastructure, bot walls or fetch errors, and we excluded them rather than guess.
The headline numbers
| Finding | Share of top sites (n = 374) |
|---|---|
| Readable content exists only with JavaScript | 6.4% |
| Loses more than 80% of its words without JavaScript | 11.5% |
| Loses more than half of its words without JavaScript | 17.4% |
| Serves a different page title to no-JS clients | 13.4% |
| Median words ratio, no-JS vs rendered | 0.99 |
Two things are true at once. The median top site is fine: server-rendered, same words either way — the 0.99 ratio says most of the web never needed the browser. And a heavy tail is not fine at all: one site in six sheds more than half its content for any client that does not run JavaScript, which in 2026 includes every major AI crawler.
The biggest sites are the worst
Split the sample by rank and the pattern inverts what you might expect. Among the top 100 domains, 8.1% serve content only with JavaScript; in the 101–400 band it drops to 6.7%, and in the 401–800 band to 5.1%. The largest properties on the web — the app-first companies with the engineering budgets to render client-side — are precisely the ones an AI crawler is most likely to read as an empty shell. Smaller sites, still built on server-rendered stacks, are more legible to machines than the giants.
Names you know that vanish without JavaScript
| Site | Words without JS | Words rendered |
|---|---|---|
| reddit.com | 0 | 376 |
| duckduckgo.com | 42 | 2,018 |
| tumblr.com | 5 | 1,585 |
| jetbrains.com | 0 | 1,584 |
| fontawesome.com | 0 | 1,233 |
| europa.eu | 8 | 585 |
| icloud.com | 0 | 137 |
| roblox.com | 0 | 125 |
The title field has its own comedy. TikTok's rendered homepage ends up with an empty title while the no-JS view keeps "TikTok - Make Your Day". Yandex answers no-JS clients with "Are you not a robot?". Microsoft's server greets a no-JS fetch in Spanish while the rendered page is English — 13.4% of the sample serves a different title depending on whether you run JavaScript, which matters when that title is what gets quoted in an AI answer.
What AI crawlers actually do with JavaScript
This is the site-side half of a picture whose crawler-side half is already documented. Vercel's analysis of crawler traffic found that none of the major AI crawlers render JavaScript — OpenAI's and Anthropic's bots fetch JS files but never execute them. An industry analysis by searchviu put it at 69% of AI crawlers unable to execute JavaScript, with no rendering queue to catch up later, unlike Googlebot's two-pass indexing. Put the two halves together: crawlers that will not render, meeting a top-100 web where one site in twelve has nothing to read without rendering. That intersection is the content simply missing from AI answers.
What to do about it
If any part of your funnel depends on being read — by search engines, by LLMs building answers, or by the agents that browse on users' behalf — the checklist is short:
- Measure your own gap. Fetch your key pages with JavaScript off and count what survives. Our SEO audit endpoint does exactly the dual fetch-and-diff used in this study ($0.0012 per URL), and the free website-to-Markdown tool shows what a no-JS reader extracts from any page.
- Server-render what matters. Product names, prices, docs, the words you want quoted — in the HTML, not behind hydration. Frameworks make this a flag, not a rewrite.
- Give machines a text lane. Prerendered fallbacks, sitemaps that are actually complete, and Markdown twins of key pages cost little and are what no-JS readers consume whole.
- Watch the title diff. If your no-JS title, description or canonical differs from the rendered one, that is the version AI systems will cite.
Methodology and limits
Sample: top 800 apex domains from the Tranco list of 25 August 2026, obvious CDN/tracker infrastructure filtered out; homepage only, one dual fetch per site through US residential exits. 374 domains returned HTTP 200 on both the no-JS and the rendered fetch and form every percentage above. Of the rest, 151 hosts do not serve a public homepage (DNS/CDN infrastructure that Tranco ranks highly), and 73 failed or timed out on at least one fetch — including hard bot walls, which means our numbers, if anything, understate how much of the top web a crawler cannot read. Single-day, homepage-only measurements are a floor, not a ceiling: interior pages of app-like sites are typically more JavaScript-dependent than their homepages, not less. The full per-domain dataset is downloadable as CSV (374 rows) — reuse it with attribution.