A taxonomy in digital advertising is a controlled vocabulary: a fixed, hierarchical list of categories, each carrying a stable identifier, that two systems can exchange without interpreting language. A publisher labels an article Personal Finance, Insurance, Health Insurance. A bidder receives that node's identifier and knows exactly which branch was meant. Nothing is guessed.

The concept exists because programmatic advertising trades in descriptions rather than objects. A bid request is a set of claims about a page, an app, a screen or a person, and the buyer has milliseconds to act on them. Free text does not survive that. One data vendor calls a segment auto-intenders, another calls a similar group in-market for cars, and no automated system can tell whether the two describe the same people. IAB Tech Lab built the industry lists to remove the ambiguity, describing their purpose as a common language for the "aboutness" of content.

How the lists are built

Four IAB Tech Lab taxonomies carry most of the traffic, and the division between them is functional. The Content Taxonomy describes what a page, app or video is about. The Ad Product Taxonomy describes the product or service advertised inside a creative. The Audience Taxonomy names segments. The Privacy Taxonomy describes how data is handled. All four sit in a public GitHub repository maintained by the Taxonomy and Mapping Working Group, described as 112 companies drawn from across the programmatic market.

Structurally these are subsumption hierarchies: a child is a narrower case of its parent, so Television sits beneath Arts and Entertainment, and depth runs to six tiers in the Audience Taxonomy. Identifiers are alphanumeric strings, though many remain sequential integers for backwards compatibility.

From version 2.0 onward the Content Taxonomy splits into two parts. Topical categories answer what the content concerns. Orthogonal attributes, called vectors, answer everything else: Content Environment, Content Purpose, Content Source and Content Form Factor, covering whether material is social or email, news or review, professionally produced or user generated, podcast or clip or game. Separation is the design point. A social post about theme parks can carry a topical label such as Family Travel alongside vector values describing the channel it arrived through, rather than forcing one category to do both jobs.

Version numbering follows a stated rule: minor releases add rows without changing structure, so Audience Taxonomy 1.1 can be read by systems built for 1.0, while major releases break. Content Taxonomy 3.0 removed parent categories, so it cannot be used alongside anything from 1.0 through 2.2.

Where the identifiers travel

In real-time bidding the transport is AdCOM, the object model underlying OpenRTB. The cat attribute carries an array of category IDs; cattax states which list they belong to. Both appear on the Ad, Site, App, Publisher, Producer and Content objects. Enumerated values run 1 for Content Taxonomy 1.0, 2 for 2.0, 3 for Ad Product 1.0, 4 for Audience Taxonomy 1.1, 5 for Content 2.1, 6 for 2.2 and 7 for 3.0, with 500 upward reserved for vendor lists. The default is 2. OpenRTB 2.6, published in April 2022, added the field so a request could declare its taxonomy rather than leaving the recipient to assume one.

Audience segments travel separately, in segment arrays tagged with their own taxonomy code. Blocked categories move the other way through bcat, which is how a publisher refuses gambling or alcohol creatives. Video passes categories through a VAST macro. Conversion payloads reuse the vocabulary: the item object in ECAPI 1.0 references the AdCOM category taxonomies, defaulting to Ad Product Taxonomy 2.0. Google Ad Manager's publisher provided signals accept data keyed to Audience Taxonomy 1.1 and Content Taxonomy 2.2, letting a publisher describe a user without sending an identifier.

Origin and evolution

Content Taxonomy 1.0 dates to 2011 and held roughly 390 categories. Version 2.0, released in November 2017, nearly tripled that to 1,180 and introduced vectors. Version 2.1 arrived in October 2020 with a sensitive data flag marking nodes that should not feed profiles, a response to the risk that content categories accumulated against an identifier become inferences about race, politics or religion. Version 2.2 followed that December, folding the eleven brand safety categories from the Global Alliance for Responsible Media framework under a Sensitive Topics parent and encoding risk as a separate vector: floor, high, medium and low.

Content Taxonomy 3.0 went to public comment on September 23, 2021 and was released in June 2022. It removed the News and Politics branch so news could be signalled through vectors alongside any topic, reworked video genres, and added handling for podcasts, games and app stores. Version 1.0, deprecated in July 2020, stayed in circulation regardless, so much of what followed was migration scaffolding: mappings in November 2023, deliverables on December 11, 2024covering connected television genres and new OpenRTB attributes named genres and gtax to replace free-text strings, and a Content 1.0 to 2.0 mapping on January 29, 2025.

The other lists grew alongside. Ad Product Taxonomy 2.0 exists because content labels were being used to describe adverts, a mismatch that lets restricted creatives slip through. Audience Taxonomy 1.1 defines more than 1,600 attribute nodes across demographic, interest and purchase-intent branches. The Privacy Taxonomy entered public comment in September 2024, built on Fides, an open-source taxonomy donated by Ethyca.

Why it matters for buyers and sellers

Three commercial jobs rest on these lists. Contextual targeting is the first, and its weight grew as identifiers weakened: Basis, working with Barometer, tags individual podcast episodes to Content Taxonomy 3.0 before bidding rather than blocking whole shows. Blocking is the second, where granularity carries revenue consequences: under a coarse list, a publisher refusing cryptocurrency adverts must block finance entirely.

Curation is the third. IAB Tech Lab's framework, formalised from work dating to early 2023, rests on seller-defined audiences, the taxonomies, data transparency standards and supply chain object transparency. Without shared vocabulary, a curated package cannot be described in terms the buyer's platform can act on. The 2025 standards roadmap framed taxonomy upgrades the same way, citing friction from free-text fields and mismatched versions as a constraint on connected television revenue.

Limitations and disputes

Adoption is the oldest complaint. A list deprecated in 2020 still required migration tooling five years later, and multiple versions run at once across trading partners, precisely the condition taxonomies exist to prevent.

Mappings between versions are imperfect by construction. The working group's guidance concedes as much, and documents the casualties: Radio had no equivalent in 2.0 and was mapped up to its parent, Hobbies and Interests; Jewelry became Women's Jewelry and Watches on the reasoning that women's fashion appears more often in the bidstream. The same guidance states that the advertiser domain field remains what sellers must ultimately use to evaluate advertisers, with categories serving as shorthand.

Even the size of the lists is disputed. IAB Tech Lab's content taxonomy page states that 3.0 expands coverage from roughly 400 categories to more than 1,500, a figure repeated in coverage of the Mixpeek donation, while its own migration documentation puts 2.0 at 1,180 against roughly 390 in 1.0. Third-party distributions of 3.1 count fewer still, some dropping deprecated rows and others keeping them.

Self-tagging is a structural weakness: the party describing inventory benefits from flattering it, and the protocol carries no verification. Katie Shell, associate product manager at IAB Tech Lab, acknowledged in an April 2026 paper that the taxonomies suffer slow update cycles, self-tagging incentive problems and an inability to express soft targeting preferences. Brand suitability shows the gap differently: no standard bid request field carries a suitability score, so risk tiers travel as vendor segments.

Privacy is the sharpest dispute. A complaint filed against Google over real-time bidding data transfers cited the Audience Taxonomy's roster of more than 1,999 characteristics assignable to individuals, alongside content codes covering bankruptcy, mental health and substance abuse. The sensitive data flags added in 2.1 address exactly this, but they are advisory.

Not the same as

A product taxonomy classifies goods for sale rather than editorial content. Google's version underpins Merchant Center feeds through the google_product_category attribute and reached structured data markup on July 7, 2026 via a CategoryCode type. Its counterpart, product_type, is a merchant's own free-text categorisation, bound to no shared list.

A brand safety framework such as the GARM floor and suitability model defines risk categories and tolerance levels. The Content Taxonomy encodes it rather than authoring it.

A venue taxonomy classifies physical screens. Digital out-of-home requests point at the OpenOOH Venue Taxonomy through venuetax and venuetypeid rather than content categories.

A conversion category taxonomy classifies events, not inventory: Campaign Manager 360 and Display and Video 360 group Floodlight activities into fourteen values across sales, leads and a residual group, a prerequisite for conversion modelling.

Recent developments

Migration became cheaper in February 2026, when Mixpeek donated an open-source mapper converting 2.x codes to 3.0 equivalents locally, combining term-frequency scoring, probabilistic ranking, nearest-neighbour search and model-based re-ranking to return confidence-scored matches. Ethan Steininger, its chief executive, called taxonomies the contract defining what systems can retrieve and act on without drifting.

Autonomous buying raised the stakes. Shell's paper of April 9, 2026 argued that integer taxonomy IDs keep agent-to-agent negotiation deterministic, since natural language leaves adjacent categories such as Green Vehicles and Car Culture indistinguishable in embedding space. Taxonomy guardrails now sit around IAB Tech Lab's buyer and seller agent hierarchies, and version 2.3 of the agentic protocol suite, released on July 30, 2026, added audience embeddings matched to Content Taxonomy standards. Whether it holds depends on adoption by both sides of a transaction, the condition every version of this problem has met since 2011.

Timeline

  • 2011: IAB releases Content Taxonomy 1.0 with roughly 390 categories
  • January 31, 2017: Content Taxonomy 2.0 opens for public review, adding more than 400 classifications across 29 tier 1 categories
  • November 2017: Content Taxonomy 2.0 released, introducing orthogonal vectors
  • July 2020: Content Taxonomy 1.0 formally deprecated
  • October 2020: Content Taxonomy 2.1 and Audience Taxonomy 1.1 released with sensitive data flags
  • December 2020: Content Taxonomy 2.2 released, incorporating the eleven GARM brand safety categories
  • September 23, 2021: Content Taxonomy 3.0 opens for a 30-day public comment period
  • February 2022: Seller-defined audience and context signalling guidance published
  • April 2022: OpenRTB 2.6 adds the cattax field
  • June 2022: Content Taxonomy 3.0 released as a breaking change
  • November 2023: Content 1.0 to Content 2.0 and Content 1.0 to Ad Product 2.0 mappings published
  • December 11, 2024: Content Taxonomy 3.1, CTV genre guidance and the genres and gtax attributes released for comment
  • September 2024: Privacy Taxonomy opens for public comment
  • January 29, 2025: Official Content 1.0 to Content 2.0 mapping released
  • August 2025: Content 2.1 to Ad Product 2.0 mappings open for comment
  • February 11, 2026: Open-source Taxonomy Mapper donated by Mixpeek
  • April 9, 2026: IAB Tech Lab publishes its position paper on taxonomy IDs for AI agents
  • July 30, 2026: Agentic protocol version 2.3 adds taxonomy-matched audience embeddings

Summary

Who. IAB Tech Lab maintains the dominant advertising taxonomies through its Taxonomy and Mapping Working Group, made up of 112 member companies. Publishers and supply-side platforms apply the labels, demand-side platforms and verification vendors consume them, and platform operators including Google publish separate lists of their own.

What. A fixed hierarchical list of categories with stable identifiers, used to describe content, adverts, audiences, venues and data handling in terms every participant resolves identically.

When. Content Taxonomy 1.0 appeared in 2011 and was deprecated in July 2020. Version 2.0 followed in November 2017, 2.1 and 2.2 in 2020, 3.0 in June 2022 and 3.1 in December 2024, with mappings and migration tooling continuing into 2026.

Where. Inside bid requests through the cat, cattax and bcat attributes, in video through VAST macros, in ad server signals such as publisher provided signals, in conversion payloads, and increasingly inside agent-to-agent protocols.

Why. Automated buying cannot negotiate over descriptions it has to interpret. Fixed identifiers turn classification into a lookup, which is what makes contextual targeting, category blocking, curation and machine-speed transactions possible, and what makes version fragmentation and self-tagging the standing failure modes.