Embedding Centroid

Macro-semantics (site level)
Embedding Centroid is the center of gravity of all page embedding vectors on a site: the point that defines 'what the site is about' in semantic space.

The center of gravity of all page embedding vectors on a website — a point that defines 'what the site is about' in semantic space. It's calculated as the mean vector of ALL content. Site Focus is the average distance from the centroid; Site Radius is the variance. Peripheral content (sponsored articles, off-topic posts) moves away from the centroid and lowers Site Focus.

Think of the centroid as the center of a target: accuracy is how close articles hit that center. For a garden furniture store, the centroid drifted toward 'garden' instead of 'garden furniture' because content about grass and flowers pulled it in the wrong direction.

In practice, you can calculate the centroid yourself in Google Colab. Just generate embeddings for all pages and compute the mean vector. Compare your domain's centroid with the top 3 competitors' centroids to see if your site is perceived the way you want it to be.

Source: AI Semantic SEO Expert, Robert Niechciał (sensai.io)