Lesson 4 of 5 · 7 min read · intermediate
The contextual comeback: targeting the page, not the person
Why contextual targeting is back, how modern AI-driven contextual works, how it compares with behavioral targeting, and where it can be gamed.
Before cookies, advertising was simple: put car ads in car magazines. That idea is called contextual targeting, and it never really went away. As signal loss spread across Safari, Firefox, apps and CTV, and privacy laws raised the cost of tracking, contextual came roaring back, now powered by machine learning instead of a keyword list.
Behavioral targeting is a shop assistant who follows you around town and then pitches you running shoes because you walked past a gym. Contextual targeting is the assistant standing inside the running store. Both might sell you shoes, but only one needs to know where you have been.
Contextual targeting
- Uses the content being viewed right now
- Needs no personal ID or tracking history
- Works on Safari, Firefox, CTV and in apps
- Risk: pages can be mislabelled or gamed
Behavioral targeting
- Uses a person's past activity across sites and apps
- Needs IDs such as cookies, MAIDs or hashed emails
- Scale shrinking as signals disappear
- Risk: consent, regulation and fraudulent profiles
How modern contextual works
- Crawl or read the contentA contextual vendor or SSP fetches the page (or app screen, video transcript, podcast audio) that the ad will appear on.
- Understand itNatural language processing and computer vision identify topics, entities, sentiment and tone, not just keywords. A page about the Mumbai monsoon's effect on traffic is classified differently from one about monsoon fashion.
- Map to a taxonomyResults are mapped to categories, often the IAB Tech Lab Content Taxonomy, plus brand-safety and brand suitability labels.
- Pass signals in the bid requestCategories travel in the bid request or are packaged by the seller as seller-defined audiences or curated deals.
- Bid and learnThe DSP bids on contexts that perform, and optimisation shifts spend to the contexts that drive outcomes.
Contextual beyond web pages
| Environment | Contextual signal | Example |
|---|---|---|
| Web | Article text, headings, images | Cricket scores page in India attracts sports-drink ads |
| CTV | Show genre, episode metadata, sometimes ACR data | A cooking show in Brazil carries a kitchen-appliance ad |
| Audio | Podcast transcript and category | A personal-finance podcast in the US carries a savings-app ad |
| In-app | App category and in-app screen | A Korean mobile game shows ads for other games |
| Retail sites | Search query and product page | Searching running shoes on a Japanese marketplace shows sponsored shoe listings |
Search advertising is the purest contextual business of all: the ad matches the query you just typed. Retail media's sponsored products work the same way. That is part of why these channels weathered cookie loss so well.
Who packages the context
Context can be added at three points in the chain. Publishers can label their own pages and sell those labels as seller-defined audiences, which keeps the data on the sell side. SSPs and curators can bundle contexts from many publishers into a single Deal ID, a practice that has grown alongside curation. And buyers can use their own classifiers inside the DSP, deciding page by page. Each option shifts who controls the definition of a sports page or a parenting page, and who gets paid for it, which is why contextual is now as much a business-model question as a technical one.
The weaknesses
- Thin or ambiguous content: a homepage or a video with poor metadata is hard to classify.
- Over-blocking: crude keyword lists once blocked news about serious topics, starving quality journalism of revenue. Modern suitability tools aim to be more nuanced.
- No frequency control: without any ID, you can show the same person the same ad many times.
- Measurement is harder: without IDs, proving outcomes leans on marketing mix modeling, incrementality tests and panels.
Why it is not the whole answer
Contextual solves relevance, not identity. It cannot tell you whether someone is a returning customer, cap frequency across sites, or attribute a sale to an impression. That is why most advertisers now blend approaches: contextual for reach and privacy, first-party data and clean rooms for known customers, and aggregate measurement to tie it together. With Chrome keeping third-party cookies, some expected contextual's momentum to fade; instead, the fact that cookies are absent from so much Safari, app and CTV traffic, plus privacy regulation, has kept it firmly in the plan.
Key takeaways
- Contextual targeting matches ads to the content being viewed, needing no personal ID.
- Modern contextual uses AI to understand text, video and audio, mapped to shared taxonomies.
- It works across Safari, apps, CTV and audio where cookies do not exist.
- Contextual can be gamed by MFA sites, domain spoofing and cloaking, so verify rendered context.
- It solves relevance, not frequency or attribution, so it is usually blended with other methods.
Questions people ask
What is the difference between contextual and behavioral targeting?
Contextual targeting chooses ads based on what is on the screen right now, such as an article about hiking. Behavioral targeting chooses ads based on what a person did before, across other sites and apps, which requires tracking IDs. Contextual is more privacy-friendly and works without cookies; behavioral can be more personal but depends on consent and identifiers.
Does contextual advertising work as well as cookie-based targeting?
It depends on the goal. For reaching people interested in a topic at the right moment, contextual can perform well and costs less in privacy risk. For retargeting known visitors or capping frequency across sites, it cannot replace IDs. Results vary by campaign, so advertisers usually test contextual against ID-based segments with holdouts rather than assuming one wins.