Search Intent Classifier

What the Search Intent Classifier does

This classifier labels each keyword in a list as informational, navigational, commercial or transactional using a published set of modifier rules, and shows for every keyword the exact words that decided it, a confidence figure and whether a second intent is close behind. You can override any label before exporting the list as CSV.

It is deliberately simple and entirely visible: no model, no hidden weights, nothing sent anywhere. That makes it fast on tens of thousands of keywords and easy to audit - and it means the labels are a first pass for a human to check, not a verdict.

How to use it

1. Paste keywords one per line, paste a CSV export, or drop a CSV or text file. A column called Keyword, Query or Top queries is found automatically; a column called Intent is kept as your override.
2. Add your brand names, separated by commas, so branded searches are recognised as navigational.
3. Select Classify. Each row shows the intent, the confidence and the deciding words with their weights.
4. Sort by confidence and review the low and mixed rows first. Use the drop-down in any row to override the label; the counts update immediately.
5. Download the CSV. It keeps both the rule's label and your override, so you can see later which decisions were yours.

Reading the results

Informational searches want to learn (how to, what is, why, guide). Navigational searches want a specific site or page (a brand, login, app). Commercial searches are researching a purchase (best, review, vs, alternatives). Transactional searches are ready to act (buy, price, coupon, near me, book).

Confidence is the winning intent's score divided by the total of all scores. 100% means every signal pointed the same way. A keyword is marked mixed when the runner-up scores at least three-quarters of the winner - "cheap flights paris vs rome" is both comparing and buying.

Unclassified means no rule matched. "cold brew ratio" contains no modifier at all; it is left unlabelled rather than guessed, and is a good candidate for checking the live results page by hand.

Worked example: sixteen keywords for a coffee equipment shop

Formulas and scoring rules

Score per intent

score(intent) = sum of weights of the matched words and phrases for that intentPhrases are matched longest first; a matched word is not counted again.

Question words

weight x 2 when an informational pattern starts the keyword and begins with what, why, how, who, when, where, which, does, can, should or is

Brand

navigational + 2 per brand match; + 2 more on login, app, account, website, support and similar when a brand is present

Confidence

confidence = top score / total score, rounded to two decimalsCapped at 0.5 when the only signal is a single weak word (weight under 2).

Mixed

mixed when second score >= 0.75 x top score

Limitations: what the result does not prove

• Rules read only the words in the keyword. They cannot see what the search results for it actually show, which is the best evidence of intent.
• The lexicon is English. Keywords in other languages will mostly come back unclassified unless they contain English modifiers or your brand.
• Some modifiers are genuinely ambiguous: "book" can be a noun or a verb, and "top" can be a ranking or a garment. Check any row where a single word decided the label.
• It does not estimate search volume, difficulty or value; bring those from your own keyword data and they are kept alongside the labels.

Privacy: where your data goes

Everything you paste, type or drop is processed in this browser tab. It is not uploaded, logged, stored or sent to analytics. Session recording and tag-manager scripts are switched off on this page.

Standards and sources

• Google Search Quality Rater Guidelines - understanding user intent (PDF)
• Google Search Central - Creating helpful, reliable, people-first content

Frequently asked questions

Last reviewed 19 September 2026 by the A2Z.Tools team against the sources listed above.