AI-Built Websites, 14 Sept – 20 Sept 2026: Weekly Scan Report

By AI Website Detector Research · Published · Covers 14 Sept – 20 Sept 2026 (ISO week 2026-W38)

Key takeaways

  • We analysed 11,697 unique websites this week. 25.1% of them (2,936) showed AI involvement: 376 were built on an AI-native platform (3.2%) and 2,560 used an AI-compatible or AI-augmented stack.
  • That is 2.4 percentage points lower than the week before (27.5%, from 3,606 sites).
  • Next.js was the most detected builder or platform with 2,158 sites (18.4% of the week's scans).
  • Among AI-native builders, Lovable led with 235 sites, ahead of Framer (106).

The week in numbers

Websites analysed
11,697
AI-native builds
376 (3.2%)
AI-assisted / compatible
2,560
Traditional builds
8,761

We analysed 11,697 unique websites this week. 25.1% of them (2,936) showed AI involvement: 376 were built on an AI-native platform (3.2%) and 2,560 used an AI-compatible or AI-augmented stack. That is 2.4 percentage points lower than the week before (27.5%, from 3,606 sites).

Most detected builders and platforms

Builder / platformSitesShareWeek before
Next.js2,15818.4%629
WordPress1,18910.2%359
Elementor5544.7%148
Shopify2822.4%71
Lovable2352%93
Astro2011.7%77
PHP1971.7%48
Express.js1901.6%49
WooCommerce1361.2%79
Squarespace1201%27

Next.js was the most detected builder or platform with 2,158 sites (18.4% of the week's scans). Next.js grew the most week on week, from 629 to 2,158 sites.

AI-native builders

Among AI-native builders, Lovable led with 235 sites, ahead of Framer (106).

Where the sites are hosted

The most common hosting provider was Cloudflare (4,763 sites).

  • Cloudflare: 4,763 sites
  • Vercel: 1,719 sites
  • Netlify: 414 sites
  • GitHub Pages: 277 sites
  • AWS CloudFront: 119 sites
  • Squarespace: 117 sites

How to read this report

Every figure counts unique domains that visitors and our own checks scanned between 14 September 2026 and 20 September 2026 (UTC). Each domain is counted once per week, bots and blocked or spam domains are removed, and the classification follows the same scoring rules as every report on this site. The sample is made of sites people chose to check, so it is not a random sample of the web: read the shares as a picture of what our users scan, not of all websites.

Want the longer view? The monthly insights and the AI builder market share page put these weekly numbers in context, and the methodology explains what the detector can and cannot see.

How our detection works — see our methodology: AI Influence Score calculation, evidence tiers, and fingerprint signal types.