The categories
The four-way scheme is standard; Google's own evaluator guidelines describe similar query types under different names (know, do, website, visit-in-person).
| Label | The searcher wants | Typical page type | Journey stage (FunnelModel of the customer journey stages.) |
|---|---|---|---|
| Informational | To know or learn something | Guide, article, tutorial, definition | Awareness, consideration |
| Commercial | To evaluate options before choosing | Comparison, review, best-of, category with guidance | Consideration |
| Transactional | To do something now: buy, sign up, download | Product, service, pricing, signup page | Decision |
| Navigational | To reach a specific site or page | The brand's own page | Any |
Two refinements are often worth adding: local (the searcher wants something near them, which the local section serves) and a split of informational into quick answer (a definition, a number, a date, where a snippet or a short page serves) versus deep (a guide). Finer schemes exist; the test of any scheme is whether each label implies a different page type. If two labels would produce the same page, merge them.
Modifier rules
Most queries declare their intent in their words. A rule set applied first labels the bulk of a list:
| Pattern | Label |
|---|---|
| how, what, why, when, guide, tutorial, examples, meaning, definition, vs (concepts) | Informational |
| best, top, review, comparison, vs (products), alternatives, cheapest, for [audience] | Commercial |
| buy, price, pricing, cost, order, book, hire, download, sign up, near me, discount, deal | Transactional (near me and hire: also local) |
| brand names, product names, login, contact, hours, "site name + word" | Navigational |
| [service] + [place], near me, open now, directions | Local |
| a bare noun with no modifier | Ambiguous; results-page check |
Rules are written per market and per language: modifiers differ, and the same word carries different intent in different categories ("cost" in a B2B software list is commercial research; in a local trade list it is transactional). Build the rule set from the list itself: sort by frequency of modifiers, label the top patterns, apply, and review a sample.
Results-page signals
When rules fail, the results page (SERPThe page of results a search engine shows.) tells the truth about what the engine believes the query wants. The signals to read, at a glance and at scale through a tool that records result features:
- Shopping results and product grids: transactional or commercial.
- Ads at the top and bottom: commercial value, usually transactional or commercial intent.
- Featured snippetA highlighted extract from one page shown as a direct answer in search results., "People also ask", knowledge panel: informational.
- Map pack: local.
- Video results: informational, often how-to.
- The ranking pages' types: ten guides means informational; ten category pages means transactional; five of each means mixed.
- A single dominant brand result with sitelinks: navigational.
The search intent guide describes reading one results page in full detail, including format and depth; here the read is a quick label.
Mixed and shifting intent
Some queries are genuinely mixed: the results page shows both guides and product pages, because the population searching is split. Label these mixed, record the dominant type from the top positions, and decide at mapping time whether one page can serve the dominant intent while acknowledging the other, or whether two pages of different types will each compete for their half. The one thing not to do is build a hybrid that is neither.
Intent also shifts: a query that was informational becomes commercial as a category matures, a product name becomes generic, a seasonal event changes what a word means. Labels carry a date; the quarterly review in the keyword research hub re-checks high-value queries against the current results page. Similarity tools do not settle intent either: two queries can sit close together in meaning and still want different pages, as how embeddings group similar text explains, so labels come from rules and the results page rather than from closeness alone.
The process at scale
- Prepare the list with volume estimates, current ranking URL and position from Search Console, and any existing labels.
- Apply the modifier rules and record the label and the rule that fired, so mistakes can be traced.
- Pull results-page features for the unlabelled remainder and for every query above a value threshold, from a rank-tracking or SERP tool; label by feature rules.
- Hand-check the remainder and a random sample of the rule-labelled queries. If the sample's error rate is uncomfortable, refine the rules and re-run. The same hand-check applies when a language model applies the rules, as the AI for keyword research guide explains.
- Record the final label, the method (rule, feature, manual), the date and any notes on mixed intent.
- Hand over to clustering, where intent is the first split: queries with different intents never share a page.
Labelling a bathroom-fittings retailer's list
A list of several thousand queries around taps and showers is sorted by modifier. "How to fit a mixer tap" and "what is a thermostatic shower" are labelled informational by rule; "best shower for low pressure" and "mixer tap vs pillar taps" commercial; "buy chrome mixer tap" and "shower valve price" transactional; the brand names navigational. Bare nouns — "mixer tap", "shower valve", "bath taps" — go to the results-page check, which shows product grids and shopping results for the first two (transactional) and a mix of guides and grids for "bath taps" (mixed, dominant transactional). A sample of the rule-labelled set finds "cost to fit a shower" labelled transactional by the cost rule when its results page is all guides; the rule is narrowed to exclude "cost to" phrasings and re-run. The labels then split the clusters: the informational queries go to a guides plan, the transactional ones to category and product mapping.
Using the labels
- Page type: the label chooses between guide, comparison, category, product or location page.
- Prioritisation: transactional and commercial clusters nearest the business's offer usually come first; informational clusters build the authority and the links; the intent signals guide places the labels on the journey.
- Expectations: informational queries have higher volumes and lower conversion; the measurement plan reads them separately, and click-through (CTRShare of impressions that result in a click.) differs by intent and result features.
- Cannibalisation prevention: two pages with the same label and overlapping clusters is the warning sign the cannibalisation guide describes.
Building and tuning the rule set
A rule set is a table of patterns, labels and exceptions, built from the list itself rather than copied from a generic one:
- Tokenise every query and count the modifiers — the words that are not the subject nouns. The top few dozen modifiers cover most of the list.
- Label each modifier by looking at the results page for a handful of queries that carry it. The label a modifier earns is the label the engine's results support, not the one the word suggests.
- Order the rules so that stronger signals win: a transactional modifier beats an informational one when both appear ("how to buy" is transactional), and brand names beat everything.
- Add exceptions as the sample review finds them, with the reason recorded.
- Measure the rule set on a held-out sample labelled by hand; the error rate tells you whether to trust it or refine it.
A category-specific rule set is a reusable asset: the next research cycle on the same site starts from it, and the exceptions log explains the oddities a newcomer would otherwise relearn.
Intent across site types
The label distribution tells a site what kind of content it needs, and the distribution varies:
- Stores find most demand transactional and commercial, with informational queries around use, care and compatibility that the e-commerce section serves with guides linked from categories.
- Business-to-business sites find the reverse: long informational and commercial research phases, few transactional queries, and navigational demand for the product name once it is known. The journey is long, and the labels map to its stages. Many of those research queries are too rare for tools to estimate, which the guide to low-volume keyword research addresses.
- Local services find transactional and local intent dominant ("emergency plumber near me") with a smaller informational tail that builds trust before the call.
- Publishers find informational intent nearly exclusively and split it finer — news, explainer, how-to, reference — because the page types differ.
Labelling the site's existing pages by the intent they serve, and comparing that to the labelled demand, shows the mismatch most sites have: pages for the intents the team likes to write, not the intents the audience has.
The same categories carry over to video platforms with their own search bars, and the guide to getting found in TikTok search adapts them to separate search-led TikTok videos from the ones made for the feed.
Sharing labels with paid search
Paid search accounts hold intent evidence the organic team rarely sees: which queries convert, at what rate, and what the searcher did after the click. Exporting the search-terms report with conversion data and joining it to the labelled list checks the labels against behaviour — a query labelled informational that converts like a transactional one is mislabelled or mixed — and feeds the SEO versus PPC decision on which clusters each channel should carry.
Intent labels feed straight into the keyword prioritisation walk-through for the SEO traffic estimator, where each cluster's conversion rate follows its intent.
Common mistakes
- Labelling from the words alone for every query. Bare nouns and category-specific words mislead.
- Too many labels. Schemes with a dozen categories that imply the same three page types.
- One rule set for every market. Modifiers do not translate.
- Never revisiting. Intent moves; last year's labels build the wrong pages.
- Hybrid pages for mixed intent. Neither half ranks.
- Ignoring result features. The engine has already said what it thinks the query wants.
