Ad Tech Is Still Buying Video by Its Description
Picture the final minutes of a romantic comedy on a streaming service. Friends crowd into the room, and everyone raises a glass. It is exactly the moment a sparkling wine brand would pay a premium to own.
Unfortunately, that toast is invisible to most ad systems. And when nobody can prove what is in a scene, brands fall back on avoidance and block whole categories, and publishers lose revenue on their most distinctive programming.
Video is still a spreadsheet to ad tech
Most ad tech still describes video the way it did ten years ago: a title, a category, a daypart, and maybe a few metadata tags. Those signals were built for scheduling and rights management, not for understanding what is happening on screen.
An advertiser writes a brief asking for "celebration moments where people raise a glass." A publisher responds with what its systems can package: titles, genres, and IAB categories. However, the brief described a feeling, and the response described a spreadsheet. By the time it passes through agencies, DSPs, and taxonomies, it has been flattened into "Family & Lifestyle."
When nobody can prove what is laid out in a scene, the industry reaches for its bluntest tool: avoidance. Brands block whole categories, not because the inventory is unsuitable, but because no one can show them which moments are. Publishers lose revenue on some of their most distinctive content.
Multimodal AI changed the math
Multimodal AI can now read every shot and scene across a catalog. A toast, a muddy hike, and a fourth-quarter comeback all become findable across millions of minutes.
Resonance pays. In a MAGNA and Warner Bros. Discovery study, viewer excitement rose from 58% to 69% when ads aligned with the surrounding content. Kantar found that campaigns are seven times more impactful when viewers are receptive to advertising.
Precision and scale stop competing
FreeWheel's Viewer Experience Lab found that tighter program-and-ad matches pay off more: brand recall rose 20% under the loosest keyword match and 38% for an exact one. The catch is that narrower definitions leave less inventory
That trade-off comes from buying off the description. With a few keywords per program, every gain in precision shrinks the pool. When every shot is understood, a precise brief can search the whole library. The same study found a 23% recall lift even in lower-engagement shows, the long tail full of moments nobody could find.
Give the map a destination
Understanding every shot gives you a map of the catalog, but only the advertiser can supply the destination.
Two paper towel brands searching the same catalog want different things: one wants kitchen spills, the other a muddy kid on a hiking trail. Ask for a "paper towel moment," and both get cooking and food. When we asked a model for "dirty face" moments for the second brand, it returned Halloween face paint. It may be technically correct, but it’s completely useless. Once we added the brand's own visual examples, it found the right moments across the catalog.
The advertiser's brief should become the targeting signal. The technology is necessary, but intent is what makes it sufficient.
What each side should do now
NBCUniversal is already helping advertisers align their campaign intent to relevant Peacock inventory. A luxury brand campaign drove 7x more website traffic because of improved relevance. Scaling this approach requires publishers and buyers to change how they work together. Publishers should package moments with evidence, including timestamps for qualifying moments, explanations of what was excluded and why, and expected impressions. Buyers, in turn, should ask for proof of qualifying moments before a campaign runs and verification afterward. Without that evidence, advertisers are still buying the description rather than the content itself.
Every streaming catalog already holds the data advertisers want, shot by shot. The companies that can read and act on that information will be better positioned to find and own the moments that matter.
[Editor's note: This is a contributed article from Coactive AI. Streaming Media accepts vendor bylines based solely on their value to our readers.]
Related Articles
The new generation of autonomous, goal-driven AI agents are poised to reshape ad-buying, ad-decisioning, contextual targeting, and optimized ad insertion across the CTV ecosystem. On November 9, Swivel's Ayelet Palmore, IAB Tech Lab's Brad Pipkin, Vevo's Matt Burgess, Magnite's Jonathan Moffie, and Future Today's Tim Ware will join Ring Digital's Brian Ring for the Streaming Media Connect panel "Agentic AI and the Future of Streaming Ad Tech." The panel explores what agentic AI means for advertisers, platforms, and the future of programmatic workflows.
26 Oct 2026
AI's potential to transform CTV and OTT ad preparation, ad serving, ad placement, contextualization, video analysis, performance, and reporting is fairly well understood and rarely understated. But how are streaming platforms, content owners, and advertisers actually operationalizing AI today? Revry's Alia J. Daniels, Digital Culture Group's Crystal Foote, Dataxis' Ophelie Boucaud, and Sargeway's Sarge Sargent discuss how publishers are leveraging AI's advantages and also addressing some of its limitations in this clip from Streaming Media Connect 2026.
01 Sep 2026
How has AI entered the media workflow? For this new column, we'll look at different applications used in the media industry. For this issue, we'll start with asset management, asset storefronts, and localization. While some of this functionality—speech-to-text transcription, translation, voice synthesis, natural language processing, logo detection, facial recognition, and object detection—has been around for a while, the biggest improvement is that much of it is now available on workflows with live content.
15 Dec 2025