Our goal with our dataset is to only focus on the personal web. The search APIs that LLMs use rank a lot of SEO/commercial content highly that we were not interested in

To measure that we do better here, we made evaluations based on how real people recommend websites on their blogs/writing/articles. For example, we tend to surface https://every.to/p/invisible-details-of-interaction-design rather than https://userpilot.com/blog/interaction-design-principles/ when searching for advice on getting better at interaction design.