How Search Intent Is Classified# Link to this article
Intent is inferred deterministically from the query tokens and the SERP features Google returned. It is never an LLM guess, and an undecidable query is labelled unknown rather than forced into a bucket.
Search intent tells you what a searcher is actually trying to do, which decides what kind of page can rank. Rankato infers it from two signals it already holds for every verified ranking: the query tokens and the SERP features Google returned.
The labels# Link to this section
- Transactional — the searcher wants to act now (buy, download, book, sign up).
- Commercial — the searcher is comparing before acting.
- Navigational — the searcher wants a specific destination.
- Informational — the searcher wants to understand something.
- Unknown — neither signal was decisive.
Why "unknown" exists# Link to this section
A classifier that always produces a label produces confident nonsense on ambiguous queries. When neither the tokens nor the SERP shape is decisive, Rankato returns unknown. That is a real answer: it tells you to look at the SERP yourself rather than trust a coin flip.
Deterministic, versioned, and re-appliable# Link to this section
The classifier is a transparent rule ladder with a fixed precedence — transactional beats commercial beats informational — and it is never an LLM guess. It is stored with its model version (intent_v1) at collection time, and the same pure function can be applied on read to older rows that predate stored intent, so historical data stays comparable.
Written by Rankato