AI Website Builders Are Quietly Making Every Site Look Alike
That new website looks familiar? Here's why AI tools may flatten everyone's taste.
What happens when AI turns plain-English requests into finished websites? Researchers studied 73 promotional sites built by graduate students with Lovable for distinct real businesses, under a graded assignment that rewarded original design. Using DINOv3 embeddings, they analyzed pairwise cosine similarity, effective diversity, and cluster structure. One representative UMAP–HDBSCAN specification assigned 88% of sites to six interpretable clusters—though sensitivity analyses yielded anywhere from three to twelve. In nine recorded sessions, preliminary AI-assisted coding documented recurring generate-and-check behavior, incomplete transmission of verbalized intentions into prompts, and infrequent explicit evaluation of originality. Among survey respondents, perceived human control, satisfaction, and perceived quality showed no detectable association with embedding-based cohort atypicality. The authors document within-cohort visual concentration and a mismatch between subjective experience and distributional distinctiveness—a condition they call authored ignorance.
- 73 students used the same AI website tool for different real businesses — and 88% of the finished sites fit into just six look-alike styles.
- Students felt creative and in control, but their sense of originality had no connection to how different their design actually was.
- Small businesses and brands risk blending together if they all rely on the same default AI designs instead of adding their own voice.
Why It Matters
If everyone builds with the same AI tools, websites, brands and businesses slowly blur into one look.