SEO Vector Gap Analyser

Crawl and extract content

We fetch each URL you provide and strip away navigation, ads, and boilerplate using Mozilla Readability, leaving only the body text that search engines care about.

Generate sentence embeddings

Every sentence from every page is converted into a 384-dimension vector using the all-MiniLM-L6-v2 language model. Sentences about the same topic end up close together in this vector space regardless of the exact words used.

Find the right number of topic clusters

We run k-means for multiple values of k, then use the elbow method to find the number of distinct topic clusters that the competitor pages actually cover. No fixed number is assumed.

Measure your page against each cluster

For each topic cluster, the coverage threshold is derived from the 25th percentile of competitor sentence similarities to that centroid, and the shallow cutoff is set at 40% of average competitor depth. Both adapt per cluster, so dense topics require genuinely dense coverage to pass.

Surface missing phrases with TF-IDF

Within each gap cluster, we run TF-IDF analysis across the competitor and target page texts to find exact terms and two-word phrases that competitors use frequently but your page is missing or underusing. These appear as phrase chips on each gap card.

Claude writes targeted recommendations

The gap clusters, representative competitor sentences, and missing phrases are passed to Claude, which generates specific heading suggestions and content bullet points you can act on immediately.