SearchAudit.io

We Audited 301 Vibe-Coded Sites for AI Search. The Divide Isn't the Builder — It's the Domain.

SearchAudit Team
Aug 9, 2026 · 5 min read
Cover image for “We Audited 301 Vibe-Coded Sites for AI Search. The Divide Isn't the Builder — It's the Domain.”

A few months ago we audited a batch of sites built with Lovable to see whether AI search engines could actually read them. Most couldn't. The fair objection: it was a small sample, and maybe we picked bad ones.

So we went wider — and, just as important, we re-ran the test. Our new set is 301 live vibe-coded sites — mostly Lovable, plus a handful from Bolt and v0 — put through the same AI-visibility audit, which checks what AI crawlers (ChatGPT Search, Perplexity, Google AI Overviews, Claude, Gemini) rely on: can the bot reach and read your content, is there structured data, semantic HTML, a sitemap, a llms.txt. Each site gets a 0–100 score.

This isn't just the old list made longer. Of the sites we'd scored before, 199 were tested a second time — and the average held: 35.9 to 36.9. Then we added 102 sites we had never audited before, a completely fresh sample. Those 102 came back at 34.0 — right in line with everything else. That's not one dataset stretched to look bigger; it's an independent replication.

The headline didn't move.

Average score: 35.9 / 100 (median 27). 83% of sites scored under 50. 57% scored under 30 — the range where AI effectively can't read them. Only 2.7% broke 90.

Same story — on a re-test and on a brand-new sample alike. That's the important part: the problem is structural, not a fluke of who we happened to test.

Then we found the fracture line.

We split the 301 sites by a single variable: is the site on its own domain (yoursite.com), or is it still living on the builder's platform subdomain (yoursite.lovable.app)? The two groups barely overlap:

122 sites on a platform subdomain — average 24.3, median 23, 98% scored under 50, and not one broke 90. 179 sites on their own domain — average 43.8, median 36, with 8 sites above 90.

That's a 19.5-point gap, and it held in every source batch we checked separately — not an artifact of one list.

A caveat we won't skip: this is a correlation, not a cause. Moving your site to a custom domain does not, by itself, raise your AI Visibility Score. The domain is a marker. Teams who bother to connect a real domain also tend to ship server-rendered pages, add structured data, and treat the site as a product rather than a demo. The domain is the visible tip of that seriousness — not the lever. And it's no guarantee either: even on their own domain, 73% of sites still scored under 50. But every single site that broke 90 was on one.

So what are the actual levers? Across all 301 sites, the low scores trace back to the same three causes:

Client-side rendering. Lovable and similar tools render your content in the browser with JavaScript. Googlebot executes JS. Most AI crawlers don't — or do it late, or badly. They arrive, see an empty shell, and leave. Your content was never read. No llms.txt. This emerging standard is a plain-text brief telling AI models what your product is and how it's priced. Most of the sites we checked for it didn't have one. No structured data (JSON-LD). Without it, AI has to guess what your page is about — and it often guesses wrong, or skips you for a competitor that spelled it out.

Why it matters now. In 2026 a growing share of buyers ask ChatGPT or Perplexity before they ever open Google. If the AI can't read your site, it can't cite you — it cites whoever it can read. That's your competitor.

The good news: it's usually fixable in minutes, not months. The sites sitting in the 80s and 90s are the ones that get server-side rendering, a llms.txt and JSON-LD right — and yes, they're almost always on their own domain.

Want to know your score? Run a free audit at searchaudit.io — paste any URL, get your AI Visibility Score and the exact list of what's blocking you, in under 60 seconds.

(Method: 301 live vibe-coded sites — 287 built with Lovable, 14 with Bolt or v0 — audited with SearchAudit.io. This set is a fresh re-scan, not a strict superset of our earlier 210-site batch: 199 sites overlap and were re-measured, 102 are new, and 11 from the earlier batch are not included. Sites returning an empty page or a dead domain were excluded before scoring. "Own domain" vs "platform subdomain" was classified from the URL. Scores reported in aggregate; no site named individually. A correlation is reported between hosting and score; no causal claim is made.)

See what AI sees on your site

Free score & issues report. No signup.

Run a free audit