AI-Enriched Metadata Drives Better CTV Content Discovery
Why are streaming services, channels, and platforms across the CTV and OTT ecosystems turning to AI/ML-enhanced metadata to remove the friction from CTV content discovery and improve user experiences? Cineverse’s Tony Huidor and SUMM8’s Jamie Mackinlay discuss how AI-enriched metadata and natural language interfaces are enhancing and transforming discovery and bringing users closer to the content they want to watch in this discussion with Integration Therapy’s Rebecca Avery at Streaming Media Connect 2026.
Matching How People Search
Avery wonders how machine learning can help with incomplete metadata, asking, “What is it about current AI practices that’s really opening up an opportunity here?”
Huidor has discovered that traditional metadata is insufficient for AI searches. Descriptive metadata doesn’t align with how people search for what to watch, he says. He gives the example of a family sitting down to watch a movie. They don’t say, “Which action movie with Tom Cruise do you want to watch?”; they say, “What are you in the mood to watch?” People connect with content on an emotional basis, he asserts, which isn’t information descriptive metadata can contain.
“So what we did with AI and machine learning was extract this rich contextual metadata on a scene-by-scene, frame-by-frame basis, and then we’ve brought that in,” Huidor shares. He talks about Cineverse’s product, cineSearch, which has an underlying dataset of domain-specific metadata for film and TV content. “We feel we have the largest dataset for AI-optimised search. We have thousands of dimensions that define a title and [encompass] mood and feelings and emotions,” Huidor explains. He gives examples of how people phrase queries: “What’s the best horror movie of the 1980s?”; “What’s the top Tom Hanks movie?”; [or] “What is the highest-rated drama from Italy?” Each query “requires different types of metadata because people search in different ways.” Cineverse extracts all of this data and uses AI to help people discover movies on a personal and deeply emotional level that’s not based on who the lead actors are or a movie’s synopsis, Huidor shares.
Managing Metadata Effectively
Rebecca Avery turns to Mackinlay to ask, “How do you see the industry trying to implement solutions like what [Huidor is] talking about?
Mackinlay calls Huidor’s take “very accurate.” Providing a natural language interface that allows viewers to pose natural questions requires good-quality data, he says, calling back to a comment from earlier in the panel by another panelist about ensuring bedrock data is accurate. “And then you’re thinking to the point about machine learning is, you’ve got now also a massive uplift of AI tags and event markers and another data[set] that could be incorporated. So you’re probably in a situation where your organisation is going to have to start linking multiple internal and external data sources together to create one organised data fabric,” Mackinlay explains.
SUMM8 is trying to do that, Mackinlay says. He notices that currently, individual vendors—content aggregators, streaming platforms, broadcasters, audience-measurement organisations, advertisers, etc.—are all trying to use the same data to gain a competitive advantage. “And that’s not possible unless you’ve got a management layer in place where you are drawing all of that data together, you’re imposing your own structure, you’ve got your IDs under control, and the data that you are using is now high-enough quality and with the right level of depth,” he explains. He’s finding that not many companies have their metadata ready to integrate with AI tools, noting they need to ask themselves the following: “Have you planned and organised your architecture in such a way that you’re actually now managing your metadata across your organisation, you’re reducing duplication, you’re improving efficiency, and you’ve got the best data available to your systems?”
The Final Word on Metadata
“Not all metadata is created equal,” Huidor agrees. “Studios have to embrace the fact that you need more types of metadata and more is more. And I think everyone approaches like, ‘Oh, I’m going to license my data from the best provider and then I’m set.’” He refers to the previous comments from another panelist that Mackinlay brought up. “It’s not about getting one set of data and you’re done. To be ready for the AI era, you need a lot of metadata, and the more the merrier because it’s going to make discovery of your content much easier if you don’t rely on one single provider.”
Join us 12-14 May 2026 for more thought leadership, actionable insights, and lively debate at Streaming Media Connect 2026! Registration is open!