What does the IAB Audience Taxonomy actually classify?
Part of Contextual targeting
The short answer
The IAB Tech Lab Audience Taxonomy is a standard list of names for audience segments, grouped at the top into demographic, interest and purchase intent. It describes groups of people, browsers or devices that a data provider has put into a segment. It does not describe pages, so it answers a different question from the Content Taxonomy.
Key takeaways
- The Audience Taxonomy standardizes segment names so that segments from different data providers can be compared, not the rules used to build them.
- Its top tier says whether a segment claims a demographic attribute, an interest or a purchase intent, and each needs different evidence.
- Content categories describe what a page is about; a page category is never proof of who reads it.
- Before using an audience label, a buyer should see the provider's disclosure of where the data came from, how old it is and whether it was modeled.
What does the IAB Audience Taxonomy classify?
Definition
- Audience Taxonomy
- An IAB Tech Lab list of standard names for audience segments, organized in tiers under three top-level groups: Demographic, Interest and Purchase Intent.
Tech Lab's taxonomies repository (opens in a new tab) states the purpose plainly: the Audience Taxonomy provides common nomenclature for audience segment names to improve comparability of data across different providers, and its Tier 1 labeling designates whether a segment describes attributes that are demographic, interest-based or purchase-intent based. The same page calls it a key element of Tech Lab's Data Transparency Standard. Version 1.1 is described there as a non-breaking update to version 1.0.
The subject of every row is an audience, meaning a group of people, households, browsers or devices that a data provider has assigned to a segment. That is the main difference from the IAB Content Taxonomy, whose subject is a page, app or video. A contextual plan uses the second. Audience labels show up beside it in data marketplaces, curated deals and bid requests, which is why planners need to tell them apart.
| Top tier | Rows | Tier 2 groups | Examples of Tier 2 groups |
|---|---|---|---|
| Demographic | 197 | 7 | Age Range, Education & Occupation, Household Data, Language, Personal Finance |
| Interest | 497 | 29 | Business and Finance, Food & Drink, Home & Garden, Travel, Video Gaming |
| Purchase Intent | 864 | 34 | Business and Industrial, Consumer Packaged Goods, Software, Travel and Tourism |
| Total | 1,558 | 70 | IDs in the file run from 1 to 1679, so numbers have gaps |
How do demographic, interest and purchase intent segments differ?
The three top tiers make three different kinds of claim. A demographic segment says something about who a person or household is: an age range, an education level, a household income band or a home location, where the file points to ISO country and region codes. An interest segment says what someone pays attention to, with names that can mirror content subjects, such as Interest, Business and Finance, Small and Medium-sized Business. A purchase intent segment says someone is in the market for a product or service, and every purchase intent row in the file carries the note See Purchase Intent Classification Extension.
| Top tier | The claim | Possible evidence behind it | What goes stale |
|---|---|---|---|
| Demographic | This person or household has an attribute | Registration forms, surveys, public records or a model | Moves, job changes and life events |
| Interest | This person pays repeated attention to a subject | Pages visited, apps used or content consumed over time | Interests fade, and one visit is weak evidence |
| Purchase Intent | This person is shopping for something now | Searches, product page visits, cart or price comparison activity | Intent ends at purchase or after a short window |
Recency matters for purchase intent segments. Tech Lab's 2022 guidance shows audience IDs carrying Purchase Intent Classification modifiers joined by a pipe character, in the form 123|PIFI2|PIPV1, so a provider can say more about the intent signal than the name alone. Ask whether a segment uses those modifiers and what each value means for that provider.
Why is a content category not proof of who reads a page?
A content category is a description of the page. Tech Lab's taxonomy implementation guide (opens in a new tab) says the Content Taxonomy simply categorizes page content and does not on its own reveal information about a user. A small business payroll article is read by owners, but also by accountants, students, job seekers and people who followed a link by accident. The page category tells a buyer what the reader chose to read in that moment, nothing more.
The same guide warns that there are few technical controls preventing taxonomy nodes being associated with individual IDs to build behavioral profiles over time. That is the path by which a content label turns into an audience label: someone logs which categories a browser visited and calls the result an interest. At that point the claim has changed object, from a page to a person, and it needs the evidence and privacy review that come with audience data.
Where do audience and content labels meet in a bid request?
In OpenRTB they travel in the same kind of container, attached to different parents. The OpenRTB 2.6 specification (opens in a new tab) says the Data and Segment objects together allow additional data about the related object, for example user or content, to be specified. The User object's data field holds additional user data, and the Content object's data field holds additional content data. The segtax community extension (opens in a new tab) adds a taxonomy ID to each Data object and says segment IDs should be considered fully proprietary when no segtax is provided.
| What to look at | Audience label | Content label |
|---|---|---|
| Parent object | user.data | site.content.data or app.content.data |
| segtax value in the extension list | 4, IAB Tech Lab Audience Taxonomy 1.1 | 5, 6 or 7 for Content Taxonomy 2.1, 2.2 or 3.0 |
| Data.name | The provider that built the segment | The provider that classified the content |
| What a match means | The browser or device was placed in a segment | The page or video was given a category |
Content categories can also arrive outside these objects, in fields such as the site's cat, sectioncat and pagecat arrays or the content object's cat array. Those describe the page. A segment ID with no segtax value is the provider's own code, so its meaning has to come from the provider's documentation.
What does an audience label review look like on a real brief?
Fictional example
A payroll software brief for Nevada restaurants
An unnamed Reno software company sells payroll and shift scheduling to small restaurants in Reno and Las Vegas. The agency builds a contextual plan from restaurant operations content: articles on tip pooling, staff scheduling for weekend rushes and opening a second location. Each site on the list carries a reason tied to its sampled articles.
A data partner then offers three segments to layer on top: Interest, Business and Finance, Small and Medium-sized Business; Purchase Intent, Business and Industrial, Human Resources; and Demographic, Education & Occupation, Self Employed. The planner records each one as a separate audience claim, apart from the contextual reasons, and asks for the provider's data label before anything is added to the plan.
The client deck keeps the two layers apart. The contextual line reads: pages about running a restaurant. The audience line reads: browsers a named provider placed in three segments, under disclosed rules. Nothing in the deck says that restaurant owners read the listed sites.
What should a buyer ask before using an audience segment label?
Tech Lab's Data Transparency Standard is the disclosure that should come with a segment. The 2022 guidance describes it as a schema of over 20 fields setting minimum disclosure requirements for any seller of data, and says the disclosures aim to clarify provenance, age, extent of modeling and segmentation criteria but do not themselves constitute a quality determination. Its inclusion methodology field separates attributes that are observed or known, declared by the audience members, derived from other fields, inferred from business rules, or modeled with an algorithm from a seed.
Six questions for any audience segment
- Ask which taxonomy row it maps to. Get the Audience Taxonomy ID and version, not only a marketing name.
- Ask how members got in. Observed, declared, derived, inferred and modeled are different strengths of evidence.
- Ask how old the data is. Purchase intent loses meaning fast, so ask for the window and the rule for leaving the segment.
- Ask what unit is counted. Individual, household, business, device and browser segments are not the same size or the same claim.
- Ask whether lookalike expansion was used. The standard's audience expansion field records whether similar IDs were added.
- Ask where the data was collected. The geography field lists countries, which matters for a plan limited to the United States or Canada.
How does PlacementGraph keep content claims apart from audience claims?
Frequently asked questions
Do two segments with the same taxonomy label contain the same people?
No. A shared label means two providers chose the same standard name for their segments. Each provider applies its own rules, sources and time windows, so the members can differ. Comparing the data labels side by side shows how far apart the two segments really are.
Can a publisher's first-party segment use the Audience Taxonomy?
Yes. Tech Lab's 2022 guidance describes publishers and their data partners signaling seller defined audiences with Audience Taxonomy IDs in user.data, with segtax identifying the taxonomy. The renamed Curated Audiences specification continues that work; our guide to curated audiences covers it.
Why do audience and content taxonomies use similar names?
Because interests can be about the same subjects that pages cover. Food & Drink, Travel and Home & Garden appear in both. The names match, but the object does not: one says a page is about travel, the other says a person was placed in a travel interest segment. Always check which taxonomy an ID belongs to.
Is the Audience Taxonomy used for Canadian plans?
It can be, with care. Its income bands are labeled in US dollars, and its location rows point to ISO codes and, for the United States, two-letter state codes. A Canadian plan should confirm how a provider handles provinces, Canadian dollars and French-language data before relying on a segment.
Sources
- Taxonomies (repository README) (opens in a new tab)IAB Tech Lab
- Audience Taxonomy 1.1 (TSV) (opens in a new tab)IAB Tech Lab
- Taxonomy and Data Transparency Standards to Support Seller-defined Audience and Context Signaling (opens in a new tab)IAB Tech Lab, 2022
- Taxonomy implementation guide (opens in a new tab)IAB Tech Lab
- Segment Taxonomies community extension (segtax) (opens in a new tab)IAB Tech Lab
- OpenRTB 2.6 (opens in a new tab)IAB Tech Lab
Researched and written by PlacementGraph Research. Read how we research, source and update these pages. Corrections: diana@placementgraph.com.