CLUSTERING

CLUSTERING is an embedding task type that optimizes vectors for topical grouping, used in keyword clustering pipelines. The Gemini API's task_type=CLUSTERING parameter generates vectors differently than with SEMANTIC_SIMILARITY — vectors optimized for grouping achieve better separation in vector space, producing cleaner clusters with a higher Silhouette Score. In SEO, it's used when you want to split 500+ keywords into topical groups, with each cluster becoming a potential content unit (article or page).

In the clustering pipeline, the keyword clustering skill uses this task type to generate embeddings, then the K-means algorithm splits them into clusters. When performing clustering tasks, task_type=CLUSTERING should be used rather than SEMANTIC_SIMILARITY to achieve optimal results. This distinction is mathematical: different task types produce vectors with different geometric properties.