Why every AI-built landing page looks the same (and how to stop shipping it)

Every AI-built landing page seems to share the same look. This is due to several factors. A few years ago, Tailwind's creator set a default button color that quickly became the default across many tutorials. This default color was then incorporated into numerous machine learning models that learned to recognize what a button should look like. As a result, the web is now filled with landing pages that have the same visual design, which is essentially a single person's offhand choice that has been adopted by many startups.

In April 2026, a developer named Adrian Krebs analyzed 1,590 recent Show HN launch pages and found that 22% of them had four or more characteristics that were unmistakably machine-made. Additionally, 32% of the pages showed two or three such characteristics. The findings were further confirmed in October when a study compared 15 different AI models and 15 human designers.

The AI models scored 90% in appropriateness, while the human designers scored 76%. However, the AI models scored 0.59 in creative range, while the human designers scored 0.90. This indicates that the AI-generated pages are more similar to each other than those created by human designers.

The trend of similar-looking AI-generated landing pages has been growing for several years. A paper published in 2021 found that the visual differences between website layouts had decreased by 43% between 2010 and 2019. The reason for this convergence can be attributed to the widespread use of frameworks like Bootstrap and Tailwind, which provide templates and design elements that are easily copied and adapted by developers.

Boris Müller, an interaction designer, noted in 2018 that the problem lies in the fact that templates are content-agnostic, which leads to a lack of uniqueness in the final design.

The main reason behind the similarity in these AI-generated landing pages is the way the models learn and produce their outputs. The models are trained on the distribution of human-created work and are then fine-tuned based on user ratings. This process leads to a "winner takes all" scenario, where a single, highly-rated answer is chosen and becomes the default choice for the model. This can result in a loss of creativity and uniqueness in the final output.

A potential solution to this issue was proposed by the same researchers who conducted the analysis of AI-generated landing pages. They suggested changing the objective to reward both good and rare answers. This approach resulted in a 75% increase in variety for stories and up to a 46% increase in variety for invented personas, while maintaining a similar level of quality compared to the original objective.

This indicates that encouraging more diverse outputs could help combat the problem of identical-looking AI-generated landing pages.