vs
We measured 114 public Replit sites (median score 40/100) but did not have a sample for Base44, so the numbers below cover one side only. The rest of the page compares how each builder can be recognised and what its sites usually get wrong.
| Replit | Base44 | |
|---|---|---|
| Type | AI builder (site written from a prompt) | AI builder (site written from a prompt) |
| Default hosting | *.replit.app | *.base44.app |
| Sites in our sample | 114 | not measured |
| Median site score | 40/100 | not measured |
| Sites with a critical problem of their own | 18.4% | not measured |
| Call a Supabase database from the browser | 0% | not measured |
| No main heading (H1) | 86.8% | not measured |
| No meta description | 41.2% | not measured |
| Zoom disabled on phones | 79.8% | not measured |
| Builder badge left on | 3.5% | not measured |
Highlighted: the better of the two on that row. Lower is better for every percentage.
ReplitHow to spot it: Look at the address: *.replit.app or *.replit.dev is Replit hosting.
What usually goes wrong: The default description is what people see in Google. If the description still says "Update this description to reflect the app", that is the sentence under your title in search results, and in every link preview.
Base44How to spot it: Look at the address: *.base44.app is Base44 hosting.
What usually goes wrong: The app shell is empty for crawlers. Base44 renders everything in the browser. Search engines and link previews see a page with no text, so the description you wrote in the editor may be all that reaches them.
Usually, yes. Each leaves its own traces: Replit on *.replit.app and in its scripts, Base44 on *.base44.app and in its own markers. Paste the URL in the scanner above and it names the builder with the evidence it found.
Neither decides it alone. In our sample the biggest SEO gaps were a missing main heading and a missing meta description, both fixable on either platform. Compare the rows above, then scan your own site to see which of them apply to it.
Not by being AI-built. The problems we see most are defaults nobody changed: titles, badges, blocked zoom, and database keys used without access rules. Each is a short fix once you know it is there.
Numbers from The AI-built web, October 2026: public sites on each builder's own hosting, scanned once. Read the method and its limits there.