Intent Classifier

Sorts a keyword list into informational, navigational, transactional and commercial. Scores all four rather than stopping at the first match, shows every signal it found, and tells you when it is not sure instead of guessing confidently.

Live v2.0 · Updated September 2026 Free · No login
Classifier Idle
Every keyword is scored against all four intents, not matched against the first rule that fires. Anything the tool is unsure about is marked for review rather than guessed at.
Runs locally · nothing sent to a server

How it works

It is pattern matching, not a model, and it runs entirely in your browser. What changed in v2.0 is how the patterns are combined.

Every intent is scored, and the strongest wins. The previous version checked rules in a fixed order and stopped at the first hit, with brand names checked first. That returned ahrefs vs semrush as navigational at 90% confidence, when it is plainly a comparison. On keywords carrying a brand name plus any other signal, it was wrong far more often than not, and always at high confidence.

Signals now carry weights. vs is a strong commercial signal. login is a strong navigational one. A brand name on its own is weak, and only becomes strong on a short query with no commercial or transactional marker, because brands appear constantly inside comparisons and questions.

Confidence is calculated, not assigned. It comes from the winning score and the margin over the runner-up, so a contested keyword reports low confidence and gets marked for review. The previous version returned a fixed number per rule regardless of how many signals matched.

Every classification shows the signals behind it, and you can move any keyword to a different bucket. Overrides are kept in the CSV export, so a corrected list stays corrected.

Frequently asked

No, and it does not need to be. It is weighted pattern matching running in your browser, which handles the large majority of keywords correctly and is transparent about the rest. Every result shows the signals that produced it, which a model would not.
The winning intent’s score plus how far clear it was of the runner-up. A keyword with one strong unambiguous signal scores near 95. A keyword where two intents scored similarly comes back low and is marked for review, because a close call is genuinely a close call. In v1.0 this number was a constant per rule and carried no information at all.
Because most keywords containing a brand are not about reaching that brand’s website. ahrefs vs semrush is a comparison, semrush pricing is transactional, and how does google rank pages is a question. A brand only signals navigation strongly on a short query with no other intent marker, which is what semrush login looks like.
Look at the live results for them, then use the move buttons to put them in the right bucket. Bare noun phrases like entity extraction tool genuinely are ambiguous, and the honest thing for a tool to do is say so rather than assign one at 90%. Your overrides are kept in the CSV.