Contextual advertising chooses which advertisement to show from the material a person is looking at - the article, video, podcast episode, app screen or search query - rather than from anything known about the person. A running shoe appears beside a marathon training plan because of what the page is about, not because of who is reading it. The approach exists because content can be read at the moment of the impression without a cookie, a device identifier or consent to profiling. That makes it the fallback wherever identifiers are missing, blocked or legally off limits. It is also the older of the two main ways to target advertising: for most of the history of print and broadcast, it was the only one.

How a page becomes a bid

Three steps sit behind a contextual impression: classification, transport and decision.

Classification happens before the auction, often long before. A crawler fetches a URL, or a publisher supplies the text, and natural language processing (NLP) extracts topics, named entities and sentiment. Newer systems add computer vision for images and transcription for audio and video. The output is mapped to a controlled vocabulary, usually IAB Tech Lab's Content Taxonomy or a vendor's own segment list. Amazon's Applied Science team described a system processing more than 8.5TB of webpage data a day, pairing offline categorisation by large language models (LLMs) with embeddings matched in real time. Frequently updated pages were re-read in near real time and static ones every 48 hours, so the quality of a verdict depends on how recently the page was visited.

Transport runs through OpenRTB, the protocol most exchanges use to send bid requests. A seller describes its inventory in the Site or App object. The page field carries the URL; cat, sectioncat and pagecat hold category codes for the site, the section and the page; keywords or kwarray carry free text. A separate field, cattax, names the taxonomy. Absent it, the specification assumes Content Taxonomy 1.0, a list IAB Tech Lab has deprecated. A recipe page might therefore send pagecat 216, Cooking under Food & Drink, with cattax 9, the code for Content Taxonomy 3.1. Connected television (CTV) requests add genres and gtax to the Content object to describe programme genre.

Decision belongs to the buyer, and few demand-side platforms (DSPs) rely on seller-declared codes alone. Contextual vendors push segments into the DSP, which checks each incoming URL against a cached classification and bids or declines, the same pre-bid mechanism that drives brand safety avoidance. Integral Ad Science (IAS) offered more than 380 contextual segments as of March 2026; Peer39, Seedtag and DoubleVerify sell comparable libraries. Platforms expose the method natively too. Google Ads documentation describes contextual targeting as matching ads to sites and videos through the keywords or topics an advertiser chooses.

Conversation is the newest kind of content. OpenAI selects ChatGPT ads from advertiser-supplied context hints, landing page content, ad title and copy, matched against the chat rather than against exact keywords.

The commercial model for the web took shape in 2002 and 2003. Applied Semantics, a Santa Monica company whose CIRCA technology analysed the meaning of web text, launched a product called AdSense that October, placing paid listings on pages according to what each page was about, according to Search Engine Watch. Google announced its own content-targeted advertising in March 2003 and bought Applied Semantics on April 23, 2003, for $102 million in cash and stock, according to Digiday. The self-service AdSense programme opened on June 18, 2003 to English-language sites, drawing on more than 100,000 AdWords advertisers.

Money then moved towards behavioural targeting, and context narrowed into a tool for avoidance. Page-level classifiers from specialists such as Peer39 and Grapeshot sold safety as much as reach; Peer39 data reached The Trade Desk by October 31, 2013, and IAS bought the semantic specialist ADmantX in November 2019. Oracle acquired Grapeshot in April 2018, reportedly for $400 million, then abandoned advertising altogether in 2024.

Privacy law revived the targeting use. The General Data Protection Regulation (GDPR) applied from May 25, 2018. IAB Europe's 34-page Guide to Contextual Advertising, released on July 28, 2021, cited research naming contextual targeting as advertisers' leading response to the loss of third-party cookies and device identifiers. Standards caught up in 2022. OpenRTB 2.6, published in April, added cattax so a request could declare its taxonomy; Content Taxonomy 3.0 followed in June with handling for podcasts, CTV and games; and IAS launched Context Control that April.

Why buyers and sellers care

Context is the one targeting input present on every impression. Identifiers are not. Chrome kept third-party cookies after Google reversed course on April 22, 2025, but Safari and Firefox traffic, much CTV inventory and any European user who refuses consent still arrive without a usable profile. That traffic can be sold only on context, which is why the price gap described below matters to publishers.

Vendors report large returns. IAS cited a 300% click-through rate increase for Samsung in March 2026, and Amazon's paper reported costs per thousand impressions (CPMs) 23% lower for brand campaigns and 15% lower for performance campaigns. Both sets of figures are self-reported and unaudited.

Regulators treat the method as lower risk. The UK Information Commissioner's Office (ICO) advised the government on May 18, 2026 to ease consent requirements for contextual targeting, frequency capping and measurement while keeping behavioural advertising inside the consent regime. For brand safety, its preferred design has the publisher classify the page and pass an abstracted signal, because a full URL combined with other data could single out an individual. In the European Union, Article 28(2) of the Digital Services Act (DSA) has barred profiling-based ads to users known to be minors since February 17, 2024, which leaves content as the practical basis for targeting them.

Where it falls short

Price is the first dispute. The Professional Publishers Association told the ICO that agencies pay "several times" more for personalised inventory than for contextual, and in some cases would not bid on contextual inventory at all. A March 2026 study by Implement Consulting Group, produced on behalf of Google, cited academic work finding a 66% efficiency loss when display spending moves from personalised to contextual ads. Vendor case studies point the other way. No independent dataset has settled the question.

Classification error is the second. Keyword lists cannot read tone, and news pays for it: Vodafone reported in July 2025 that revised settings cut its block rates by 41% and opened 10% more news inventory, partly by moving to exact-match keywords so that a blocked "die" stopped catching "diet", figures supplied by Vodafone and DoubleVerify. The Media Rating Council (MRC) ruled on October 18, 2025 that property-level checks may not be sold as brand safety unless images, video and audio are analysed too.

Coverage is the third. A March 2026 analysis by Peer39, cited by Gracenote, found only 40% of CTV bid requests carried usable programme-level signals. Paywalled pages and apps with little text give crawlers less to read.

Relevance is a proxy, not a measurement. Zack Rosenberg, chief executive of Qortex, told AdExchanger that Nike advertising performed about four times better against cooking content than basketball content on his platform, arguing that the task is to match the audience consuming the content rather than its subject. The incentives can also be gamed: the acronym MFA originally meant made for AdSense, thin pages built to harvest contextual clicks.

Nor is the label a legal safe harbour. A bid request still carries an IP address and device data unless they are stripped. On September 17, 2026 the Regional Court of Cologne rejected Snap's argument that ads chosen from My AI chatbot prompts were merely contextual, holding that processing prompt text engaged Article 9 of the GDPR because prompts can contain sensitive information.

Not the same as

Behavioural advertising selects from a profile assembled across sites and over time. Ad personalisation covers its forms; Google's non-personalised mode is, in effect, contextual plus coarse location.

Brand suitability uses the same classification to avoid content rather than to seek it. Excluding tragedy coverage is suitability; buying recipe pages is targeting.

Search advertising matches a declared query. The EU Council's draft consent exemption treats a single search query as context, but search is sold, priced and measured as a separate channel.

Semantic targeting is a technique, not a category: one way of producing contextual signals, alongside keyword matching and embeddings.

Recent developments

Chatbots are the live test. OpenAI began showing ChatGPT ads in the United States on February 9, 2026, matched on the current conversation, past chats and previous ad interactions. Its European launch in August 2026 started contextual only, excluding past chats and memories until users opt in to personalisation.

Streaming metadata is consolidating around identifiers. The Trade Desk gained Gracenote's show-level content IDs on August 27, 2026, letting buyers target and curate below the level of genre.

European law is moving too. A Council of the EU compromise text dated September 3, 2026, published by noyb on September 21, would exempt from consent the measurement of advertising "based solely on the immediate content displayed during an individual visit to a single web page or on the basis of a single search query", provided no profiling or link to a user's past or future activity is involved. The text still faces negotiation with the European Parliament. The Cologne judgment runs the other way, narrowing what can be called contextual once the content is something a user wrote.

Timeline

  • October 2002: Applied Semantics launches AdSense, placing paid listings on pages by analysing their text
  • March 2003: Google announces content-targeted advertising for publishers
  • April 23, 2003: Google acquires Applied Semantics
  • June 18, 2003: Google opens the self-service AdSense programme
  • October 31, 2013: Peer39 contextual data available in The Trade Desk
  • April 2018: Oracle acquires Grapeshot
  • May 25, 2018: GDPR applies across the EU
  • November 2019: IAS acquires ADmantX
  • July 2020: IAB Tech Lab deprecates Content Taxonomy 1.0
  • July 28, 2021: IAB Europe publishes its Guide to Contextual Advertising
  • April 2022: OpenRTB 2.6 adds the cattax field; IAS launches Context Control
  • June 2022: Content Taxonomy 3.0 released
  • February 17, 2024: DSA ban on profiling-based ads to known minors applies to all online platforms
  • September 30, 2024: Oracle advertising products, including contextual intelligence, reach end of life
  • December 2024: Content Taxonomy 3.1 released
  • February 11, 2025: Amazon publishes its technical paper on LLM-based contextual targeting
  • April 22, 2025: Google says Chrome will keep third-party cookies
  • October 18, 2025: MRC restricts brand safety claims by property-level verification
  • February 9, 2026: ChatGPT ads launch in the United States
  • March 2026: Google-commissioned Implement study published; Peer39 finds 40% of CTV bid requests carry programme signals
  • May 18, 2026: ICO advises easing consent rules for contextual targeting
  • August 2026: ChatGPT ads reach Europe on a contextual-only basis
  • August 27, 2026: The Trade Desk adds Gracenote show-level IDs
  • September 3, 2026: Council compromise text adds a consent exemption for measuring contextual ads
  • September 17, 2026: Regional Court of Cologne rules against Snap over My AI prompts used for ads

Summary

Who. Publishers and supply-side platforms describe inventory; contextual and verification vendors such as IAS, DoubleVerify, Peer39 and Seedtag classify it; DSPs including Display & Video 360, The Trade Desk and Amazon DSP act on the signals; IAB Tech Lab maintains the taxonomies and OpenRTB fields; regulators including the ICO and EU institutions decide how far the method escapes consent rules.

What. A way of selecting advertisements from the content being viewed - a page, video, programme, app screen, search query or chat - rather than from data about the viewer, built on classification mapped to shared or proprietary taxonomies.

When. Automated web matching took commercial form with Applied Semantics' AdSense in October 2002 and Google's content-targeted ads in March 2003. It was eclipsed by behavioural targeting from the mid-2000s and revived after GDPR took effect in May 2018 and cookie restrictions spread.

Where. In bid request fields such as page, pagecat, cattax and genres, in DSP pre-bid segments, in platform keyword and topic settings, across web, app, CTV, audio and, since 2026, chatbot inventory.

Why. Content can be read on every impression without an identifier or profiling consent. Its value is contested: publishers report it fetches a fraction of personalised prices, classifiers misread tone, CTV signals are sparse, and courts have begun to test whether content a user types can be treated as context at all.