Home / Outbound Benchmarks / Readable text for AI crawlers

How much of your site can an AI crawler read? 116 company sites measured: Wix 0.95% text, others 5.62%

GPTBot, ClaudeBot and PerplexityBot do not execute JavaScript. We fetched 116 company homepages as a crawler would and measured the text that remains once scripts, styles and tags are removed. Wix sites: 268 KB of HTML for 2,754 characters of text. 43% of the sites had no H1 at all.

Key finding: Fetched with an AI crawler user agent and stripped of scripts, styles and tags, Wix-built homepages exposed 2,754 characters of readable text inside 268 KB of HTML (0.95%), versus 5.62% for non-Wix sites with a similar word count. 43% of the 116 sites served no H1 tag in the raw HTML, and 3% served an error page to the crawler.
Measured onSeptember 21, 2026
Sample116 company websites from one B2B prospect list
MethodOne HTTP GET per homepage with the ClaudeBot user agent, no JavaScript execution. Script and style blocks and HTML tags removed, remaining characters counted and divided by the HTML size. H1 presence checked in the raw HTML. Platform detected from the markup.
SourceVision Media internal measurement (campaign and API logs)
Readable text share in the raw HTML, 116 sites
GroupTypical HTML sizeReadable textText shareNo H1 in raw HTMLError page served to the crawler
Wix-built sites268 KB2,754 characters0.95%
Other platforms, similar word countabout 5 times less codesimilar5.62%
All 116 sites43%3%

Why it matters for AI answers

Published crawler analyses (Vercel and MERJ, more than 500 million GPTBot fetches) show zero JavaScript execution by GPTBot, ClaudeBot and PerplexityBot: they read the first HTML response and cite what is in it. Googlebot renders JavaScript, and so does Gemini through Google's infrastructure. A site can therefore rank well on Google and be nearly empty for the engines that answer instead of ranking. This is not a Wix defect: Wix ranks fine on Google in 2026 precisely because Google renders JavaScript.

How to measure your own site in one minute

  1. Fetch your homepage with curl and a crawler user agent (for example ClaudeBot).
  2. Remove script and style blocks and all tags; count what is left.
  3. If the remaining text is a title and a few hundred characters, the AI engines see a blank page: server-render the content or pre-render static HTML for the pages you want cited.
  4. Check that the raw HTML contains an H1 and the key facts you want quoted; add an llms.txt at the root with the same facts.

A note on this site

We ran the same test on our own homepage before writing this page: it was a client-rendered React application with no readable body text. The benchmark pages you are reading are static HTML for that reason.

What this measurement does not say

Questions

Do ChatGPT and Claude crawlers run JavaScript?

No, according to large-scale crawl analyses and our own fetches: they read the raw HTML. A client-rendered single-page application exposes almost nothing to them.

Related measurements

Want this run on your pipeline?

We operate multichannel outbound (cold email, LinkedIn, paid retargeting, newsletters) for B2B companies, and we measure it the way these pages do.

Book a call