Contextual AI Targeting 2.0 — Beyond Keywords to Scene and Sentiment Detection
For years, contextual advertising was built around a relatively simple idea: understand what a piece of content is about, then place an appropriate advertisement next to it.
An article about smartphones could carry a handset advertisement. A recipe page could feature kitchen appliances. A sports story could attract sportswear. The logic was straightforward, scalable and, importantly, did not require advertisers to know much about the individual consuming the content.
But the internet has become far more complex than a collection of pages defined by keywords.
Video, connected TV, podcasts, gaming, social content and short-form entertainment have made context increasingly visual, auditory and emotional. A single piece of content can move between subjects, tones and moments within seconds. A football match can contain excitement, disappointment, celebration and controversy. A cooking programme can shift from instruction to humour to family conversation. A news video can move from analysis to breaking information.
Keywords alone struggle to capture that complexity.
This is where contextual targeting is entering its next phase.
Contextual AI targeting is moving beyond simply identifying what content contains and towards understanding what is happening inside that content, how it feels, and whether a particular moment is appropriate for a brand.
The shift could have significant implications for video advertising, CTV, digital publishing and the broader effort to build relevance without relying on individual-level tracking.
“The next generation of contextual targeting is not asking only what the content is about. It is asking what is happening right now, and whether that moment makes sense for a brand.”
From Keywords to Contextual Intelligence
Traditional contextual targeting depends heavily on signals such as keywords, categories, topics and metadata. These remain useful, particularly across large volumes of web content, but they can miss the meaning embedded in visual and audiovisual content.
Consider a video containing the word “running”. A keyword-based system may identify running as a relevant topic. But it cannot necessarily distinguish between a marathon, a comedy scene involving someone running away, a news report about an athlete’s injury or a documentary discussing urban crime.
The word is the same. The context is not.
AI-based contextual systems can process multiple signals simultaneously. Computer vision can identify objects, locations, people, logos and activities. Natural language processing can analyse dialogue, subtitles and surrounding text. Audio intelligence can interpret speech, sound and potentially tone. Machine learning models can then combine these signals to build a richer understanding of the scene.
The result is a shift from keyword matching to semantic and multimodal understanding.
For advertisers, that difference matters because relevance is rarely determined by one word.
A luxury automobile brand may want content involving travel, design and premium experiences. A sports brand may want moments of athletic performance. A food brand may find value in scenes involving cooking, family meals or celebration.
The opportunity is to identify those moments even when the relevant signal is not explicitly written in the content metadata.
Why Video Changes the Contextual Equation
Video has made contextual targeting considerably more difficult and potentially more valuable.
A webpage may have one dominant subject. A 30-minute video can contain dozens of contextual shifts.
That creates a problem for conventional classification systems. If an entire programme is labelled simply as “comedy”, an advertiser may lose the ability to distinguish between a light-hearted scene and a moment involving violence, grief or sensitive subject matter.
AI can operate at a much more granular level.
Instead of treating the programme as one contextual unit, systems can analyse individual scenes, frames, dialogue segments and audio cues. This allows advertising decisions to become more closely aligned with the actual environment surrounding an impression.
That is particularly relevant to CTV, where advertisers increasingly want the storytelling environment to play a role in audience and media decisions.
A brand may not simply want to reach people watching a particular genre. It may want to appear around specific types of moments within that genre.
“In video, context is not a label attached to the programme. It is a sequence of moments.”
Scene Detection Adds a New Layer
Scene detection is one of the most important developments in this area.
AI models can analyse visual frames to identify objects, settings, actions and relationships between elements. Over time, these signals can be combined to understand what a scene represents.
Imagine a travel programme moving through an airport, a hotel, a restaurant and a beach. A conventional programme-level classification may describe the content as travel. A scene-level system can identify four distinct commercial environments.
That distinction creates new possibilities for advertisers.
A luggage brand could be relevant around the airport sequence. A hotel platform could be relevant around accommodation content. A food delivery service could align with the restaurant scene. A sunscreen brand could fit the beach environment.
The media decision becomes more precise without requiring the advertiser to identify the person watching.
This is an important philosophical change in targeting.
The industry has spent years attempting to make advertising more individualised. Contextual AI suggests another route: make the environment more intelligent.
Sentiment Becomes Part of the Context
Scene detection answers the question of what is happening. Sentiment analysis attempts to answer a different question: how does the moment feel?
This is where contextual targeting becomes more complicated.
A scene can be categorised as sports, but sports content can contain celebration, tension, disappointment, aggression or tragedy. A news article can discuss a particular industry while carrying an overwhelmingly negative tone. A social video can mention a product category while being sarcastic or critical.
Sentiment-aware contextual systems can potentially add another layer to media decision-making.
For some brands, positive emotional environments may be desirable. Others may value moments of excitement, anticipation or humour. Certain categories may want to avoid environments characterised by distress, violence or controversy.
However, sentiment detection should not be treated as a perfect science.
Human emotion is complicated. Sarcasm can be difficult for machines to interpret. Cultural references can change meaning. A serious discussion can contain humour, while an apparently positive sentence may occur within a deeply negative story.
That makes model quality, training data and human oversight important parts of contextual AI deployment.
“Emotion can make context more useful, but it can also make contextual classification more difficult. The closer AI gets to meaning, the more nuance matters.”
Brand Safety Moves From Blocking to Understanding
Brand safety has traditionally involved exclusion lists, keyword blocking and broad content categories.
The problem with keywords is that they can produce both false positives and false negatives.
A news story about a natural disaster may contain words associated with violence or tragedy even when the surrounding context is responsible journalism. Conversely, a seemingly harmless keyword may appear inside a scene that is unsuitable for a particular advertiser.
AI-driven contextual analysis can potentially make brand-safety decisions more granular.
Instead of asking whether a prohibited word appears, systems can assess the relationship between words, images, actions and the overall scene.
For example, an article discussing violence prevention is fundamentally different from content depicting violence. A documentary about financial crime is different from content promoting criminal activity.
Context is what creates that distinction.
This does not mean AI should make every brand-safety decision autonomously. Rather, it can give buyers a richer set of signals with which to define suitability.
Beyond Brand Safety: Brand Suitability
The distinction between brand safety and brand suitability is becoming increasingly important.
Safety typically asks whether an environment crosses a line. Suitability asks whether it is actually a good fit for a particular brand.
That difference opens a larger opportunity for contextual AI.
A financial services brand may technically be safe within a comedy programme but find stronger relevance around content discussing financial planning, entrepreneurship or major life decisions.
A beauty brand may not simply want “fashion” content. It may want scenes involving skincare routines, beauty tutorials or occasions where appearance and preparation are central to the story.
Suitability therefore moves contextual advertising from a defensive system into a planning tool.
Instead of using context primarily to avoid bad environments, marketers can use it to discover valuable ones.
The Rise of Multimodal Advertising Signals
The real transformation is happening because AI does not need to rely on one type of information.
Text, images, audio, speech, metadata and scene structure can all contribute to a single contextual decision.
This is known as multimodal understanding, and it is particularly important as media formats converge.
A podcast may contain spoken words but no visual layer. A YouTube video has speech, music, text overlays and images. CTV content combines video, dialogue and sound. A social post may contain a caption, video, comments and visual references.
For advertisers, the contextual signal is becoming richer because the machine can interpret more of the content itself.
This could also make contextual targeting more useful in environments where conventional cookies or user-level identifiers are unavailable.
Instead of asking, “Who is this person?” the system can ask, “What is this person watching, reading or listening to right now?”
That distinction becomes increasingly relevant as privacy expectations and regulation continue to reshape digital advertising.
CTV Could Become the Testing Ground
Connected TV is particularly well positioned for contextual AI because it combines the scale of television with the data architecture of digital media.
CTV environments contain long-form content with rich visual and audio information. That creates a large amount of contextual data that can potentially be analysed at the scene or episode level.
Advertisers can therefore begin to think about CTV not just in terms of audience segments, but also in terms of content environments.
A sportswear brand might target moments of athletic performance. A travel brand might align with destination-related scenes. A consumer electronics company could look for technology-rich environments.
The same principle can extend beyond CTV into online video, digital publishers and other content ecosystems.
The result could be a new form of contextual media planning where the content itself becomes an addressable layer.
“The audience is still important, but the content environment can become a signal in its own right.”
What Happens to Keywords?
Keywords are not disappearing.
They remain fast, understandable and effective for many forms of contextual targeting. The change is that they are increasingly becoming one signal among many.
A sophisticated contextual system may use keywords to establish the subject, computer vision to understand the visual environment, speech analysis to capture dialogue and sentiment models to determine the emotional tone.
The combination is more powerful than any individual signal.
This is similar to how search evolved. A query initially provided a relatively simple signal. Search engines became more sophisticated by interpreting intent, relationships and meaning.
Contextual advertising is undergoing a comparable transition.
The question is moving from “Does this page contain the right word?” to “Does this environment create the right moment for this message?”
The Challenge of Scale and Accuracy
There is, however, a significant operational challenge.
Analysing video frame by frame and audio segment by audio segment requires considerable processing power. Doing this across millions of hours of content demands infrastructure, optimisation and reliable classification systems.
Accuracy is equally important.
An AI system that incorrectly classifies a scene can create two problems. It can exclude valuable inventory from a campaign or place an advertisement in an environment that is inappropriate for the brand.
For this reason, contextual AI should not be judged solely on how many signals it can detect.
The more important question is whether those signals improve actual media decisions.
Advertisers will need transparency around classification methods, confidence levels, taxonomy definitions and measurement. They will also need to understand how models behave across languages, cultures and different types of content.
The Human Role Does Not Disappear
AI can identify patterns at a scale that humans cannot. But advertising decisions are not purely technical.
A brand may deliberately choose to appear next to challenging cultural conversations because the context fits its positioning. Another may avoid an environment that an algorithm considers suitable because it does not fit the brand’s identity.
That is why contextual AI is best understood as an intelligence layer rather than an autonomous media strategist.
It can make more information available to planners. It can improve classification. It can identify patterns across massive content libraries. But the interpretation of what those signals mean for a brand remains a strategic decision.
This is particularly important when sentiment enters the equation. Emotional context is not universally positive or negative. Sometimes tension creates attention. Sometimes a serious environment gives a brand credibility. Sometimes the safest decision is not the most relevant one.
The technology can surface the opportunity. Strategy still determines whether the opportunity matters.
From Targeting People to Targeting Moments
The biggest implication of contextual AI may ultimately be conceptual rather than technical.
Digital advertising has spent much of its history becoming increasingly focused on individual users. Behavioural data allowed marketers to build profiles, audiences and segments around what people had previously done.
Contextual AI points towards a different model.
Instead of relying primarily on a profile of the person, advertisers can understand the environment surrounding the impression with much greater depth.
That can create a more privacy-conscious form of relevance.
A consumer does not have to be labelled as a “sports enthusiast” for a sports brand to appear around a moment of athletic achievement. Someone does not have to be identified as a traveller for a luggage advertisement to appear around travel content.
The content itself can provide the signal.
“The next evolution of contextual advertising may not be about knowing more about the consumer. It may be about understanding more about the moment.”
What Comes Next
Contextual targeting 2.0 will not replace every form of audience targeting, nor will it make behavioural signals irrelevant. The more likely outcome is a convergence.
Audience intelligence can tell marketers who they want to reach. Contextual intelligence can tell them where and when that audience may be most receptive. Creative intelligence can determine what message should appear in that environment.
When those three layers begin working together, media planning becomes less about finding an audience in isolation and more about finding the right combination of person, place, content and moment.
For advertisers, this could be particularly valuable in a media environment where privacy restrictions are increasing, content formats are fragmenting and consumers are becoming harder to reach through broad categories alone.
The promise of contextual AI is therefore not simply better targeting.
It is better understanding.
And as advertising moves deeper into video, CTV, audio and other content-rich environments, understanding the scene may become as important as understanding the screen.
“`