How Google matches pages to search intent

Take [mercury]. On October 28, 2026, we captured the results page for [mercury] on google.com and for [mercure] on google.fr, then identified the entities (planet, metal, brand, place…) each of the top ten results is about. The formula of KScorerKScorer’s refres_question_rank signalΣ (1 / rank × score of the entity in the page), over the top 10 resultsEach result gives the meaning the score of its entity in the page (from 0 to 1), divided by its rank: the 1st counts for 1, the 2nd for 1/2, the 10th for 1/10. A meaning’s share is its total set against the total of all meanings compared.Engineers’ comment: “Sum of (1/web_rank * refres score of all question entities) over top 10 results.”, the system that settles between the meanings of a query, then gives each meaning’s share: every result gives a meaning the weight of the entity in the page, divided by its rank.

Eight queries around [mercury], eight answer formats

For eight more precise queries around the same word, here is what the real SERPs show (captured on October 28, 2026 on google.com, on desktop), with the broad intents they reflect and the page categories they put on top.

• mercury planet

  Wikipedia, NASA, the Smithsonian, the Planetary Society…, plus videos, short videos, a “What people are saying” block and images.

  Broad intentsIMAGEVIDEOSHORT_VIDEOINFORMATION_SOURCE_ENCYCLOPEDIA
  Page categoriesINFORMATION_SOURCE_ENCYCLOPEDIA
• mercure hotel london

  A hotel pack at the top of the page, with the city’s Mercure hotels, then the Accor group’s website, Tripadvisor, Kayak and booking sites.

  Broad intentsLOCAL
  Page categoriesRESERVATIONTRAVEL_GUIDE
• mercury retrograde 2026

  Seven of the top ten snippets show a 2026 date, at Vogue, Cosmopolitan, the Statesman, Astrostyle…: the query wants this year’s dates.

  Broad intentsFRESHINFORMATION_SOURCE_NEWS_PUBLISHER
  Page categoriesTIME_SENSITIVITYINFORMATION_SOURCE_NEWS_PUBLISHER
• mercury in fish pregnancy

  The FDA first, then the NIH twice, the EPA, two Australian government agencies and the Mayo Clinic: six of the top ten are public bodies.

  Broad intentsINFORMATION_SOURCE_GOVERNMENT
  Page categoriesINFORMATION_SOURCE_GOVERNMENTYMYL_HEALTH_OR_SAFETY
• broken mercury thermometer

  A “Discussions and forums” block with Reddit threads (“I broke a mercury thermometer”), Quora and two wikiHow tutorials.

  Broad intentsUGC_DISCUSSIONAUTHENTICITY
  Page categoriesFORUMUGC_DISCUSSIONTUTORIAL
• best mercury-free thermometer

  Popular Science, CNN Underscored and Consumer Reports next to Walmart and Target, several product blocks and a forums block.

  Broad intentsSHOPPINGINFORMATION_SOURCE_MERCHANT
  Page categoriesPRODUCT_TOPNPRODUCT_COMPARISONSHOPPING_MERCHANT_PRODUCTLISTICLE
• freddie mercury

  The knowledge panel, a carousel, short videos, a “What people are saying” block and images.

  Broad intentsVIDEOSHORT_VIDEOIMAGE
  Page categoriesINFORMATION_SOURCE_ENCYCLOPEDIANEWS
• freddie mercury bohemian rhapsody

  The song’s knowledge panel with its listening buttons, short videos, YouTube, Spotify and a “What people are saying” block.

  Broad intentsTOLEDOVIDEOSHORT_VIDEO
  Page categoriesLISTENLYRICS

For every broad intent, Intent Brain counts satisfaction in two ways:

• on the page: the satisfaction a result of that type brings. On [mercure hotel london], it is a click on the hotel pack followed by a visit that lasts;
• in the tab: the satisfaction that follows a switch to the matching tab. The user who leaves the SERP for Maps also expresses the LOCAL broad intent.

The two shares are stored in separate counters: Google knows whether a query is satisfied by a block on the page, or whether people prefer the full tab.

For every broad intent, Intent Brain breaks satisfaction down by the age of the date shown in the snippet. The 2024 leak defines that date, bylineDate, as “the date that will be shown in the snippets in web search results”. On [mercury retrograde 2026], seven of the top ten results show a 2026 date: these counters tell Google whether people are satisfied with yesterday’s page or with last year’s. Set against all dated results, they give the query’s need for freshness, the FRESH broad intent.

The US antitrust trial against Google showed what these counters do in real time. A 2018 internal Google document, cited in 2023, describes the query [nice pictures] on the day of the Nice attack: “Instant Glue will suppress Image Universal. News and pictures from Nice, France is the main intent of the query.” Within hours, an image query becomes a news query.

Orbit is the Google system that sorts every page in the index into categories: what the page lets you do (book, log in…), its format (tutorial, review, forum…) and who publishes it (news, government, merchant…).
It gives every page a score from 0 to 100 for each of the 86 categories, and those scores are stored with the page in the index (the 2024 leak shows them among indexing data).
KScorer uses it to check that the top 10 pages confirm a meaning.
Categories add up: a car insurance comparison published by an insurer is at once a product comparison, a merchant page and YMYL financial content.

The four search intents the SEO trade uses to classify queries (navigational, informational, commercial, transactional) are found in these categories. So are two of the “3 Cs”, the method Ahrefs popularized for analyzing a SERP before writing: look at the top results’ content type (product page, article, category page…), content format (tutorial, comparison, list, review…) and content angle (for beginners, cheap, updated this year…). Orbit names the type and the format; the angle remains the writer’s ground.

The word SEOs use for “10 best…” articles is Google’s word too: LISTICLE. Three neighbors come with it: LISTICLE_OR_AGGREGATOR, AGGREGATOR (pages that compile content from elsewhere) and LISTICLE_EXP, an experimental variant. All four go to the logs without serving as a filter: Google tracks them closely.

Orbit, the system that sorts pages in the index into categories, can attach two options to each category:

• add_restrict makes the category filterable: Google can restrict a search to that category’s pages, as the Forums filter does. 35 categories in service have it.
• should_log makes the category measurable: Google tracks it in the statistics drawn from its logs, under the name “Orbit Doc Intents”. 19 categories have it.

The two lists barely overlap. Formats and uses (tutorial, recipe, news, booking…) serve as filters; sources and listicles are measured without serving as filters.

Everything is settled before pages are ranked. When Google understands a query, it attaches several pieces of information to it: the words that characterize it, its own Orbit category (Orbit classifies queries too) and Intent Brain’s output. Ranking systems therefore know which answer format clicks have validated before they compare pages. Take [mouse]: the animal or the computer device?

1. The querymousetyped on google.com
2. The meaningKScorerpicks the computer device, not the animal
3. The formatIntent Brainexpects products to buy: SHOPPING, with images
4. The pageOrbitfiles Target, Walmart, Amazon and eBay as merchants: INFORMATION_SOURCE_MERCHANT
5. The SERPSuperrootassembles the page: images, a popular products block, then the merchants; the animal only comes up in one “People also ask” question

1. Read the SERP as a measurement, country by country

   The blocks that dominate a SERP and the order of the filters under the search bar reflect what clicks have validated, query by query: a hotel pack for [mercure hotel london], videos for [mercury planet], forums for [broken mercury thermometer]. Aim for the format they show: captioned images, an up-to-date Google Business Profile, product pages with their merchant data, an answer drawn from experience. And check every market separately: [mercury] and [mercure] share neither the dominant meaning nor the blocks.

2. Pick the page category before writing

   Review, comparison, buying guide, top N and merchant page are five distinct categories, each measured separately. On a query that expects an official source, like [mercury in fish pregnancy], a merchant page starts at a disadvantage; on an experience query, like [broken mercury thermometer], first-hand accounts and forums answer better than a generic article.

3. Justify every listicle

   Google labels and measures listicles. A ranking that sets out its criteria, its tests and its selection method follows Google’s guidance on reviews. The ranking where the publisher puts itself first is the format Lily Ray saw fall.

4. Vendors, fire up the JEV

   While rank trackers can still capture SERPs (Google is making life hard for scrapers), they have what it takes to turn these schemas into reports. Three projects:

   • show for every keyword the number of dated snippets and the average age of their dates;
   • sort every top 10 result into Orbit’s categories (merchant page, review, comparison, forum, official site, listicle…);
   • map the blocks shown and the order of filters to Intent Brain’s broad intents (IMAGE, LOCAL, SHORT_VIDEO, UGC_DISCUSSION…), country by country.

   An SEO would then see at a glance which format and which source a query expects.

The meaning with KScorer, the format and the source with Intent Brain, the page category with Orbit: three decisions made before pages are compared, and all of them visible in the SERP.