GEO
We Fetched 64 Big-Site Homepages Without JavaScript: How Many Gave AI Crawlers Real Content?
Published 2026-11-13
If an AI crawler does not run JavaScript, it only sees what the server sends in the first response. We wanted to know how many well-known sites actually put real text there, so on 2026-11-13 we fetched the homepage HTML of 73 sites without running any scripts and counted the visible words.
How we measured
We used the same hand-picked list of 73 well-known sites as in our structured data study. A script fetched each homepage with a desktop Chrome user agent on 2026-11-13 and did not execute JavaScript. 64 returned readable HTML. 9 did not (forbes.com, etsy.com, ebay.com, tripadvisor.com, yelp.com and bloomberg.com answered 403, amazon.com answered 202, and washingtonpost.com and bestbuy.com failed to load). Those 9 are excluded.
For each page we removed scripts, styles, noscript blocks, SVG and comments, stripped the tags and counted the words that remained. That is roughly what a crawler that reads raw HTML would have to work with.
Almost everyone serves real text
| Visible words in the raw HTML | Sites (of 64) |
|---|---|
| Under 100 words | 3 |
| 100–299 words | 2 |
| 300–999 words | 15 |
| 1,000+ words | 44 |
59 of 64 (92%) delivered at least 300 words, 44 delivered more than 1,000, and the median was about 1,360. Well known sites overwhelmingly render their homepage content on the server or ship it as static HTML, so a crawler that never runs a script still gets plenty to read.
The exceptions
- reddit.com: 1 word
- medium.com: 49 words
- airbnb.com: 85 words
- walmart.com: 236 words, with a single-page-app root element
- microsoft.com: 299 words
We would not call all of these empty shells. A response with one word may be a bot-challenge page returned to our script rather than the real site, and we cannot separate that from a genuinely client-rendered page. walmart.com is the clearest case of a thin response from a site that boots as a single-page app. Nothing here shows how any particular AI product handles these sites.
What this means for a small site
Large sites rarely have this problem, but small sites built as client-side apps do. We saw it first-hand in our own audits of small-business sites: one app's raw HTML held about two words (see the small-business guide). It is also the cheapest GEO item to fix: server-render or statically generate the page so the content is in the first response.
Googlebot does run JavaScript, but rendering can lag behind the first fetch. For AI crawlers the vendors say little, and several are widely reported not to execute scripts, so putting the content in the HTML removes the question. Our audit flags a page as failing when the raw HTML is an empty app shell and as a warning when only part of it is rendered later. For the access side of the same question, see our 70-site crawler study and GPTBot vs OAI-SearchBot, and you can test robots.txt rules with the AI Crawler Checker.
Limits of this study
- The sample is hand-picked and skews toward large US-based brands. It is not a random sample of the web.
- It covers the homepage only, in one fetch, as a desktop browser would identify itself.
- A low word count can mean a bot-blocking page. We cannot tell the two apart.
- It counts visible words in the raw HTML. It does not judge whether those words are useful.
- It is a snapshot taken on 2026-11-13.
Sources
Official documentation this post relies on (links checked 2026-11-13).
Frequently Asked Questions
How many big-site homepages serve real text without JavaScript?
59 of the 64 homepages we could read (92%) had 300 or more words of visible text in the raw HTML. 44 had over 1,000 words. The median was about 1,360 words.
Which homepages gave almost no text?
Three had under 100 words: reddit.com (1 word), medium.com (49) and airbnb.com (85). walmart.com (236) and microsoft.com (299) were thin but not empty. Some of these may be bot-blocking pages rather than real app shells, which this method cannot tell apart.
Do AI crawlers run JavaScript?
Googlebot renders JavaScript, though not always right away. For many AI crawlers the vendors do not document it, and several are widely reported not to run scripts. The safe assumption is that anything that matters should be in the initial HTML.
How can I check my own site?
Fetch your page with curl or view its source (not the inspected DOM) and look for your headings and body text. A page with a handful of words and a single root div is a JavaScript shell. HellSEO's audit checks this in the AI-Readable Content item.
Related Free Tools
Related Posts
We Checked 70 Major Sites: Who Blocks AI Crawlers in 2026?
We Checked 62 Big-Site Homepages: Who Publishes Entity Schema and Trust Pages?
The SMB Owner's Guide to Not Getting Buried by AI Search