Query Classification
Assigning search queries to labels — like intent categories — so they can be routed, mapped to content, and reported at scale.
When people type searches into a search engine, they leave behind a raw stream of keywords. Query classification is the process of turning that messy data into structured intent. It assigns search queries to specific categories, like commercial versus informational, or groups them by industry and funnel stage. This makes it possible to route queries, map them to content, and run reports at scale.
Traditionally, classification systems are rigid, trained on a fixed set of labels. But a new approach uses an open-set, zero-shot design. Instead of relying on predefined category codes, it treats labels as natural text. This means you can give the system any new list of categories at any time, and it will immediately score how well they fit each query without needing to be retrained. It can even apply multiple labels to a single query at once, using categories it has never seen before.
This model was trained on over one hundred thousand pairwise data points spanning dozens of industries. For search engine optimization, this kind of flexible classification is a game-changer. It allows marketers to plug in data directly from search consoles to map user intent, analyze gaps in search engine results, and build dashboards that track how user intent shifts over time. Ultimately, it bridges the gap between raw search data and highly relevant content engineering.
Query classification is the task of assigning search queries to labels — commercial vs informational intent, industry categories, funnel stages — so they can be routed, mapped to content and reported at scale. It's how a system turns a raw stream of keywords into structured intent.
Our own Universal Query Classifier takes this further with an open-set, zero-shot design: you hand it any list of labels at inference time and it scores which ones fit each query, with no retraining. Because labels are treated as text rather than fixed IDs, it can apply several at once, including labels it has never seen. It was trained on 114k pairwise rows across more than 40 industries using a binary relevance objective.
For AI SEO this powers query-intent mapping straight from Search Console exports, SERP intent-gap analysis, and dashboards that track shifting intent over time. It's a close cousin of query fan-out and feeds directly into relevance engineering.