The phrase “contextual targeting is back” has become a refrain in programmatic circles, and it is simultaneously true and misleading. Contextual targeting is absolutely experiencing a demand surge — DSPs report that spend on contextual segments is up significantly, publishers are investing in contextual classification infrastructure, and vendors that sat in the middle tier for years are suddenly fielding acquisition interest. But calling this a “comeback” implies that today’s contextual is the same product that adtech left behind in 2013 when behavioral audience targeting scaled. It is not.
The contextual targeting of 2010 was keyword-based and category-level. A travel advertiser placed ads on travel content. A finance advertiser placed ads on finance pages. The category designation relied on a combination of URL-level classification (this domain is a finance site) and keyword presence (this page contains words associated with travel). It was imprecise — it could not distinguish between a luxury travel article and a budget travel article, or between investment banking content and personal finance content — but it was workable as a blunt instrument.
What is emerging in 2020 is built on natural language processing, semantic analysis, and machine learning. The contextual signal available from a page today is orders of magnitude richer than what keyword parsing provided, and the ability to combine it with first-party data creates a post-cookie targeting story that is genuinely compelling — not just a fallback for when cookies disappear.
What Has Changed in Contextual Technology
The core technical shift is from string-matching to meaning. Earlier contextual technology identified pages containing specific words. Modern contextual technology attempts to understand what pages mean — the sentiment of an article, the nuanced topic coverage, the entity relationships that connect a piece of content to specific audience interests.
Natural language processing allows classification systems to read page content in context rather than in isolation. A page about “Apple” is classified differently based on whether the surrounding text discusses technology products, fruit agriculture, or stock market performance. NLP-based contextual avoids the category confusion that made keyword-based contextual imprecise.
Sentiment analysis adds an emotional dimension to topic classification. Two pages about the same topic can have radically different emotional registers — a neutral explainer about credit card debt management versus an anxiety-inducing piece about consumer debt crisis — and for advertisers trying to place ads in environments that align with their brand’s tone, that distinction matters.
Entity extraction identifies the specific named entities in content — people, places, brands, events — and builds a richer contextual fingerprint than topic categories alone. An article mentioning specific automotive brands, particular car models, and recent fuel economy regulations is contextually distinct from an article that is merely “in the auto category.”
Vendors including GumGum (VERITY platform), Peer39, and Oracle’s Grapeshot have been building these capabilities for several years. What has changed in 2020 is demand: the combination of cookie deprecation planning, ATT uncertainty, and general privacy regulation pressure is converting contextual from a “nice to have” supplementary tactic to a core targeting methodology.
OpenRTB 2.6 and Content Object Enrichment
On the infrastructure side, the OpenRTB 2.6 specification updates include meaningful additions to the bid request’s content object that give contextual signals a richer representation in the programmatic supply chain.
The content object in OpenRTB carries metadata about the content in which an ad will be served: category, producer, series, episode, genre, content rating. In OpenRTB 2.6, the content object is extended to carry the IAB Content Taxonomy v2.0 categories, which are more granular and better structured than the legacy taxonomy, and to support additional contextual metadata fields that contextual vendors can populate.
This matters because contextual data has historically been under-represented in bid requests. Publishers who classified their own inventory according to the IAB Content Taxonomy v1.0 (often very coarsely) were providing limited signal to buyers. Richer contextual metadata in the bid request — at the page level, with semantic enrichment — gives buyers much more information to target against without relying on a user-level identifier.
The IAB Tech Lab’s Contextual Targeting Working Group is actively working on standards for how contextual data is represented and passed in bid requests. Publishers investing in contextual classification today are building toward an infrastructure that OpenRTB 2.6 and its successors will support at scale.
IAB Content Taxonomy v2
The IAB Content Taxonomy v2.0 is worth attention from anyone building contextual targeting or brand suitability infrastructure. The v2 taxonomy significantly expands the category set compared to v1: from roughly 350 categories to over 600, with improved hierarchical organization and new category groups covering emerging topics.
More importantly, v2 includes a brand suitability sensitivity classification layer that aligns with GARM’s brand safety floor definitions. Categories are tagged with sensitivity scores that allow advertisers to express content avoidance preferences in terms of the taxonomy rather than custom keyword lists. A publisher that correctly classifies its inventory to v2 taxonomy can communicate brand suitability information to buyers without requiring the buyer to maintain custom block lists for every domain.
The adoption curve for v2 is still in early stages. Many publishers are still using v1 classification or no formal taxonomy at all. But v2 provides the right infrastructure for a post-cookie contextual economy, and the investment in implementing it is justified regardless of what specific cookie replacement API eventually ships.
How Contextual Layers With First-Party Data
The most compelling aspect of modern contextual targeting is not contextual alone — it is the combination of contextual signals with publisher first-party data to build what some practitioners are calling “contextual audiences.”
A publisher with logged-in users can match the contextual characteristics of content a user reads over time to that user’s authenticated identity. A user who consistently reads content about investment strategy, financial planning, and retirement preparation is demonstrating interests that are valuable to financial services advertisers — but the signal comes from content consumption, not from cross-site tracking.
This contextual audience approach does not require third-party cookies. It requires publisher first-party data and contextual classification of the publisher’s own inventory. The resulting audience signals are portable within the publisher’s authenticated environment, and with identity resolution infrastructure like LiveRamp’s Authenticated Traffic Solution, they can be extended to other publishers where the same user is authenticated.
The key difference from behavioral audience data is provenance: contextual audiences are built from what users read, not from where they browse across the entire web. That distinction matters for privacy compliance and for the types of inferences that are legally and ethically permissible.
For programmatic buyers planning a post-cookie transition, the most actionable near-term investment is building direct relationships with publishers who have strong contextual classification infrastructure and first-party authentication programs. These publishers will be the most valuable inventory sources after cookie deprecation — not because they are the most premium brands (though premium and contextual quality correlate), but because they have the data infrastructure to support targeting methodologies that work without cross-site identifiers.
FAQ
Is modern contextual targeting as effective as behavioral audience targeting? The honest answer is: it depends on the use case and the measurement approach. For brand awareness and upper-funnel campaigns, contextual targeting in well-classified environments can match or exceed behavioral targeting performance because brand-relevant context tends to correlate with relevant audience attention. For conversion-optimized performance campaigns, behavioral targeting historically outperforms contextual on last-touch metrics. The comparison becomes more complicated when you account for the fact that behavioral targeting metrics are themselves subject to attribution quality issues, and that post-cookie, the comparison will be contextual versus modeled behavioral rather than contextual versus deterministic behavioral.
How do I access modern contextual segments in my DSP? Most major DSPs have integrations with contextual data providers — ask your DSP team specifically about Peer39, GumGum VERITY, and Oracle Grapeshot segment availability within your buying platform. In some cases these are pre-built segments you can activate; in others you may need to work with the contextual vendor directly to build custom contextual taxonomies aligned to your brand’s targeting objectives. The contextual segment landscape is less standardized than third-party audience data, so provider relationships matter more than in behavioral buying.
What does it cost to implement contextual classification as a publisher? The cost varies significantly based on implementation approach. Managed service contextual classification from vendors like Peer39 or Oracle Grapeshot involves a fee for classification per page-view or per API call. Self-managed classification using open-source NLP frameworks is more technically demanding but lower in direct cost. For most publishers, the managed service approach is more practical given the technical complexity of maintaining accurate NLP classification. The cost should be evaluated against the revenue opportunity from better-classified inventory — contextual premium inventory can command meaningfully higher CPMs, which typically justifies the classification investment.
Will contextual targeting work for mobile in-app environments? Contextual targeting in mobile in-app environments is less developed than in browser environments because in-app content is harder to classify programmatically — it is often dynamic, personalized, or structured in ways that are not easily parsed by web crawlers. Some contextual vendors are developing in-app classification capabilities, and the IAB’s App Content Taxonomy provides a framework. The practical state today is that in-app contextual is less precise than web contextual, but it is improving as ATT creates demand for non-IDFA targeting approaches in mobile.