Research

Which site builders AI crawlers can read: the AI App Builder Index (November 2026)

52% of browser-rendered Lovable sites we scanned send AI crawlers an empty homepage, against 6% of server-rendered Lovable sites and 3% of Next.js sites.

By the Why Won't AI teamPublished 2 min readAsk about this page:ChatGPTClaude
On this page

We checked 130 scanned websites for one problem: an AI crawler that does not run JavaScript gets an almost empty homepage. 25 of the 130 sites (19%) had it. The share changes a lot with how the site builds its pages. 12 of 23 browser-rendered Lovable sites (52%) had it. Only 1 of 17 server-rendered Lovable sites (6%) and 1 of 31 Next.js sites (3%) had it.

Empty homepage by platform

PlatformSitesEmpty homepage for AI crawlersShare
Next.js3113%
Lovable, server-rendered (TanStack Start)1716%
Astro600%
React single-page app10330%
Lovable, browser-rendered (Vite)231252%
Other platforms (WordPress, Webflow, Replit, Base44)7343%
Platform not detected36514%
All sites1302519%

Several groups are small. Read the shares for groups under 10 sites with care. We will run this index again each month and add the new numbers here.

What "empty homepage" means

A browser runs the page's JavaScript and shows the full text. Some AI crawlers, for example GPTBot, ClaudeBot and PerplexityBot, read only the HTML that the server sends. We flag a site when a browser sees the homepage text but one or more of these crawlers gets almost none of it. Then an AI assistant has nothing from that page to quote.

What this means

The builder is not the whole story. What matters is whether the server sends the page text in its HTML. In our sample, server-rendered sites rarely had the problem. Sites that build their pages in the browser had it much more often. If you are not sure which kind your site is, a scan shows what each crawler gets.

How we got these numbers

We took the latest scan of each site that was scanned with Why Won't AI from November 11 to November 17, 2026: 130 sites. We left out our own three sites. The scan detects the platform from the page code; it could not detect a platform for 36 sites. Many of the sites are new products whose founders listed them on launch directories, so the sample leans to small, new sites. It is not a random sample of the web. We do not name any site.

Monthly update

Monthly results

On the first day of each month, we count the sites that we scanned in the month before. Each site counts once, with its latest scan.

The first monthly update appears in early December 2026.