Website categorization: how sites get labels, and when a domain label misleads
Part of Lookalike websites
The short answer
Website categorization assigns content labels to a whole domain, a section or a single page. Labels are declared by the publisher, inferred by a classifier from the hostname or the text, or set by a reviewer. A domain label summarizes the site's main subject, so buyers should check which level a label describes before trusting it on a mixed site.
Key takeaways
- A category can describe three different things: the whole site, one section of it, or the page an ad will load on, and OpenRTB carries each in its own field.
- OpenRTB defines site categories as an array, so a site can carry several labels; a single label in a spreadsheet is a simplification someone made.
- Labels go stale when a site reorganizes, changes owner or adds third party sections, so each label needs a date and the URLs behind it.
- Categories work as a first filter for lookalike searches, then the articles themselves decide which candidates really resemble the seed.
How are websites assigned categories?
Definition
- Website categorization
- The assignment of one or more content labels to a website, one of its sections or one of its pages, drawn from a named vocabulary such as the IAB Tech Lab Content Taxonomy, so that buying systems and people can select or avoid inventory by subject.
A label a buyer sees comes from one of three places. The publisher can declare it, for example in the bid request its ad server sends, or in page markup: schema.org defines articleSection (opens in a new tab) for news and magazine pages and notes that articles may belong to one or more sections, such as Sports or Lifestyle. A classifier can infer it, from the hostname alone or from the text of sampled pages. Or a person can assign it after reading the site. The source of a label matters as much as the label.
| Method | How the label is made | Strength | Weakness |
|---|---|---|---|
| Publisher declared | The publisher sets categories in its ad server, bid requests or page markup such as articleSection | The publisher knows its own sections and can label each one | Self interested; a publisher may pick broad, sellable labels or leave old ones in place |
| Hostname classifier | A model maps the domain name to topics without reading pages | Fast, needs no crawl, gives the same answer each time | Cannot see sections that sit in a folder path, so every page inherits the domain's labels |
| Content classifier | A model reads the main text of sampled pages and maps them to a vocabulary | Reflects what the site actually publishes, section by section | Only as good as the sample; a few viral articles can tilt the result |
| Human review | A reviewer reads a set of articles and assigns labels by hand | Catches tone, depth and sponsored content that models miss | Slow, hard to repeat across reviewers, and dated the day it is done |
The hostname approach has a well documented example. Chrome's Topics API proposal, which Google said on October 17, 2025 it would retire along with other Privacy Sandbox technologies (opens in a new tab), chose topics based only on website hostnames (opens in a new tab). Its explainer gives the consequence plainly: example.com/tennis would only carry the topics of the more general example.com. Any label built that way describes the domain, never the section.
Why can a domain level category mislead a buyer?
A domain label is an average. On a specialist site the average can sit close to the pages themselves. On a mixed site the same average may describe few of its pages well. A regional newspaper labeled News also runs a weekend food section, a high school sports desk, an obituaries page and a coupon section. A buyer who selects the domain because of its News label gets all of them.
Third party sections make the gap wider. Google's spam policies (opens in a new tab), last updated August 28, 2026, describe site reputation abuse: third party content placed on a host site mainly to borrow the ranking signals the host earned with its own first party content. One of Google's examples is a medical site hosting a low quality casino advertising page from a third party. The same policy treats a publisher's coupon section run in partnership with a specialist provider as unlikely to draw action. Either way, the domain's category says nothing about those pages.
How many categories can one site carry?
There is no fixed limit in the standard. OpenRTB 2.6 defines cat, sectioncat and pagecat as string arrays, and the Publisher object has its own cat array too, so one site can carry several labels at each level. The cattax field names the taxonomy in use; if no cattax value is supplied, the specification says IAB Content Category Taxonomy 1.0 is assumed. A buyer who sees an old style ID in a report may be looking at that default, not at a deliberate choice.
In practice the number of labels is set by whoever built the list. A planning spreadsheet with one category column squeezes a site that covers five subjects into one label. That is a formatting decision, not a fact about the site. A better record keeps a primary label for the main subject and separate labels for each section that publishes regularly, each with article URLs that prove it.
| Field | What to record | Why it matters |
|---|---|---|
| Level | Domain, section path such as /autos/, or a single URL | A section label cannot be bought on platforms that only take domains |
| Labels | Each label with its ID and the taxonomy version, from the cattax value if supplied | IDs differ between taxonomy versions, so a bare number is ambiguous |
| Source | Publisher declared, classifier, or human review, with the tool or reviewer named | Lets the next planner judge how far to trust the label |
| Evidence | Two or three article URLs per label, with their publication dates | Anyone can reopen the pages and confirm or dispute the label |
| Share of sample | How many of the sampled articles fit each label | Separates a section the site runs every week from one stray story |
| Checked on | The date the label was last confirmed | Labels drift as sites change; an undated label cannot be trusted |
How often do site categories change?
Google's own help page on topic targeting (opens in a new tab) says that as content across the web changes over time, the pages on which ads appear can change with it. A stored label does not change by itself, so the risk is a label that stays put while the site moves. The practical answer is to re-check a label when an event suggests the site has changed, and to date every label so stale ones are visible.
Events that should reopen a site's category label
- The site changed owner, or its about page names a new parent company.
- New folders appear with content the site did not publish before, such as coupons, casino reviews or product roundups.
- The publisher started hosting sponsored or partner sections with their own bylines.
- The domain lapsed and came back with a new design and a new kind of content.
- Campaign reports show ads serving on URLs in sections nobody reviewed.
That last trigger uses data the buyer already has. Google's placement reporting (opens in a new tab) separates the group view, which shows where ads served at the domain or channel level, from the detail view, which lists specific URLs. Reading the detail view shows which sections of a domain actually took the impressions. Google notes that very low activity placements may be rolled into a single summary row, so the URL list is a sample, not a census.
How do categories help find similar sites?
Categories are useful as the first cut in a lookalike search, not the last. A seed site labeled with a narrow category points the search at the right neighborhood, and a candidate with no overlapping label at any level is unlikely to be a close match. Beyond that, labels are too coarse to rank candidates: two sites can share every label and still differ in depth, voice and audience. The lookalike websites guide covers why meaning based matching on articles does the ranking.
Using category labels in a lookalike search without being misled by them
- Label the seed at the section level. Record the section labels of the part of the seed the client likes, not the domain label of the whole seed.
- Filter candidates by overlapping labels. Keep candidates that share at least one section or page level label with the seed's section.
- Read the candidate's matching section. Open several recent articles in the section that carries the shared label and confirm the subject really matches.
- Check the label's share. A candidate whose matching label covers one article in twenty is a different kind of site from the seed.
- Flag mixed domains. When the match sits in one section of a broad site, note whether the buying platform can target that section or only the whole domain.
- Date the evidence. Save the article URLs and the date checked beside every candidate that passes.
What does a categorization check look like on a real brief?
Fictional example
A child car seat maker plans display in Georgia and Florida
An unnamed fictional maker of infant and convertible car seats briefs its agency for display in Georgia and Florida, beside content for expecting and new parents. The client's favorite seed is the family section of a regional newspaper, and the agency's domain database labels that newspaper as News.
The planner labels the seed at the section level instead: parenting, child safety and family health, with three article URLs on installing seats, choosing a stroller and newborn sleep. Filtering candidates by those section labels surfaces two parenting magazines, a pediatric nurse's blog and a lifestyle portal whose domain label is Family and Parenting.
Reading recent articles changes the picture. The portal's parenting folder holds a few older stories, while most recent posts are in a deals section run by a partner, so it goes on a review list with the reason written down. One parenting magazine has a new owner listed on its about page and a fresh product roundup folder, so its label is re-checked before it stays. The final list records each site's level, labels, source, evidence URLs and the date checked.
How does PlacementGraph judge a site by its articles rather than its label?
Frequently asked questions
Is a site's category the same as its brand suitability rating?
No. A category says what a site is about, such as parenting or automotive. Suitability says whether its content fits a brand's tolerance, such as how it treats crime or tragedy. A parenting site can be on topic and still run stories a brand would avoid, so both checks are needed.
Can a buyer target a section of a site rather than the whole domain?
Sometimes. Some buying platforms accept URL or path level placements, while others, and exchange deals built from domain lists, work at the domain level only. Check what the platform accepts first. If only the domain can be bought, the client is accepting every section of it, so review those sections too.
What should a buyer do when two sources give a site different categories?
Treat it as a sign that the site is mixed. Compare the level each source describes, since one may label the domain and the other a section. Then read recent articles and record the labels that the evidence supports, with the URLs. A disagreement between sources is useful information about the site, not noise.
Does a top level category include everything beneath it?
In Google Ads topic targeting it does: Google says that targeting a top level topic also targets all the related subtopics. That makes a broad topic a wide net. For lookalike work, label the seed at the most specific level that fits, so the search does not drift into neighboring subjects.
Sources
- OpenRTB 2.6 specification (Site, Publisher and Content objects) (opens in a new tab)IAB Tech Lab, on GitHub
- articleSection (opens in a new tab)Schema.org
- The Topics API explainer (opens in a new tab)Private Advertising Technology Community Group (W3C), on GitHub
- Update on plans for Privacy Sandbox technologies (opens in a new tab)Google Privacy Sandbox, October 17, 2025
- Spam policies for Google web search (opens in a new tab)Google Search Central, updated August 28, 2026
- About topic targeting (opens in a new tab)Google Ads Help
- Track ad placement and performance (opens in a new tab)Google Ads API documentation
Researched and written by PlacementGraph Research. Read how we research, source and update these pages. Corrections: diana@placementgraph.com.