AI-Powered Content Discovery for OTT & CTV

Navigating the vast libraries of modern streaming services can feel like searching for a needle in a digital haystack. Viewers frequently face decision fatigue, endlessly scrolling through options without settling on anything to watch, a phenomenon often called the paradox of choice. Fortunately, artificial intelligence is stepping in to transform content discovery, using viewer data to craft a uniquely personalized experience. Still, this promising technology brings its own set of challenges and ethical considerations that industry leaders are actively addressing.

Experts from leading streaming technology firms highlight both the advantages and potential pitfalls of integrating AI into over-the-top (OTT) and connected TV (CTV) platforms. They explore how filtering methods can keep recommendations transparent, how personalization can avoid becoming overly rigid, and the role of evolving IP distribution in shaping the future of streaming.

AI’s Current Impact on Content Discovery

Companies like 24i, based in Brno, Czech Republic, are at the forefront of this shift. Their data-driven video platform leverages advanced technology to enhance user engagement and create new revenue opportunities.

Sebastian Braun, CEO of 24i, observes, “Artificial intelligence is fundamentally changing how people find something to watch. Through projects with partners like Crossmedia and RTVE Play, we’ve seen AI turn passive browsing into an intuitive, deeply engaging journey. The real value, however, lies in applying the technology thoughtfully.”

LTN, headquartered in Maryland, helps clients monetize live sports and news, launch channels quickly, and automate content workflows. Roger Franklin, LTN’s chief strategy officer, notes, “AI is making huge advances in personalization and discovery. It simplifies the user journey and helps major providers aggregate services and surface relevant suggestions across different platforms. Personalization is now a core component of the media experience, touching everything from targeted ads to customized audio and graphics in live sports. For content owners, it’s essential for maximizing return on expensive rights and keeping audiences engaged with relevant, valuable experiences. The old one-to-many broadcast model is gone, replaced by a need to deliver tailored, platform-specific live experiences to diverse audiences at scale.”

Amagi, a global SaaS provider using cloud-native solutions, connects media companies with viewers worldwide. Srinivasan KA, Amagi’s co-founder and president of global business, adds, “Today, AI algorithms power most content discovery on streaming platforms. The majority of viewing now comes from personalized suggestions rather than manual searches. These systems analyze individual viewing habits and content characteristics to predict what someone will enjoy. Generative AI, for instance, can automatically enrich metadata with multilingual summaries and keywords, making content more searchable across different regions.”

Content-Based vs. Collaborative Filtering

Recommendation systems generally fall into two categories. Content-based filtering matches a user’s past preferences with attributes of available content. Collaborative filtering identifies users with similar tastes and recommends titles based on their collective behavior. Many experts advocate a hybrid model to capture the strengths of both.

Braun explains, “At 24i, we don’t pick one method over the other. We blend various models, similarity-based, metadata-driven, and popularity-based algorithms, so viewers find both expected favorites and unexpected gems.”

SpoonLabs, a South Korean company with a U.S. focus, offers platforms for live audio and short-form video. Hayden Sang-won Ha, a machine learning engineer at SpoonLabs, says, “Collaborative filtering uses behavior patterns of similar user groups, while content-based filtering looks at specific item attributes like genre or cast. In streaming, where audiences often follow trends, collaborative filtering usually delivers stronger results.”

Srinivasan agrees, noting, “Neither approach alone is a perfect solution. A hybrid method, combining crowd wisdom with AI-driven content analysis, using enriched metadata and thematic understanding, often produces the most well-rounded recommendations.”

Solving the “Cold Start” Problem

The “cold start” issue arises with new users who lack viewing history or new content without engagement data. It also refers to performance delays in serverless architectures.

Sang-won Ha suggests, “For new users, simple onboarding steps like a short survey or genre selection can provide initial signals. For new titles, emphasizing key details, genre, cast, themes, within content-based filtering helps surface fresh content to the right viewers more effectively.”

Braun shares 24i’s approach: “We combine smart defaults, like editorial picks or trending titles, with rapid feedback loops. In the case of RTVE Play, we built a system that adjusts in real time based on user interactions, so relevance improves with every click. The goal is to shorten the journey from new user to engaged viewer.”

Srinivasan adds, “When user data is sparse, we rely on content understanding and broad trends. New users might start with popular or editor’s choice titles, possibly after indicating preferences, and the system adapts quickly as they watch and provide feedback. For new content, AI-generated metadata, summaries, keywords, genre tags, helps connect a show with the right audience from day one, even without a viewing history.”

Ethical Considerations in AI Personalization

Using viewer data to personalize content raises important ethical questions.

Braun states, “With great personalization power comes great responsibility. We design AI systems to be fair, transparent, and inclusive, avoiding algorithmic bias and ensuring all audience segments are served. Ethical AI isn’t optional, it’s essential for building lasting viewer trust.”

He continues, “Our mission is to help broadcasters and streaming platforms deliver personalized content and ads that are both effective and ethical. AI-driven discovery should connect people with content that inspires and delights, while safeguarding the trust that keeps them returning. When done right, streaming becomes less about finding something to watch and more about feeling understood and respected.”

Transparency in AI Recommendations

Building user trust requires clear communication about how recommendations are generated.

Braun notes, “Users engage more with suggestions they understand. Explainability is crucial. Imagine a small label under a recommended show saying, ‘Because you enjoyed Series X and follow Actor Y.’ In our future designs, transparency will be a default feature, helping users feel in control rather than manipulated.”

Srinivasan emphasizes, “We believe users should know why something is recommended. This means providing simple, readable cues, like ‘Because you watched X’ or noting shared actors or themes. Generative models can help create engaging explanation labels in the interface. By highlighting factors such as genre, cast, social signals, or mood, we make the recommendation process clearer and build stronger trust.”

Emerging Technologies in Streaming Personalization

Multimodal AI, capable of analyzing video, audio, and text together, is set to redefine content understanding.

Sang-won Ha explains, “Looking forward, multimodal AI will transform how we interpret content. For Vigloo’s vertical dramas, we could identify mood and subtle scene characteristics, not just genre and themes. This deeper understanding allows more precise mapping of user preferences, enabling truly granular personalization.”

Braun outlines 24i’s data strategy: “The best recommendations come from rich, well-structured data, viewing history, content metadata, contextual signals like time of day, and real-time behavior. Our work with RTVE integrates multiple data sources into a centralized AI engine, allowing precise tailoring across video, audio, news, and podcasts.”

Srinivasan details Amagi’s innovations: “AI is getting better at analyzing video and audio to autonomously create outputs like highlight reels or multilingual metadata, which can tailor the viewing experience dynamically. We’re also exploring AI-driven content presentation, automatically selecting keyframes, generating platform-specific thumbnails, and creating personalized channels. We’re moving toward a ‘zero-slate’ experience where adaptive AI ensures viewers aren’t interrupted by filler during ad breaks. The future points to holistic personalization, seamlessly blending what’s recommended, how it’s presented, and how it’s delivered to keep each viewer fully engaged.”

Generative AI and the Future of Discovery

Large language models (LLMs) and other generative AI tools are poised to play a central role.

Srinivasan says, “Generative AI enables a deeper semantic understanding of both content and user intent. We use LLM-based techniques to automatically summarize and tag content, letting the AI ‘read’ scripts or audio to extract context like plot points and characters. This allows recommendations and search to go beyond keywords, connecting viewers with content in a more nuanced way, matching mood or narrative style, for example. LLMs also enable conversational discovery; a viewer could ask an AI assistant for ‘a light-hearted crime series with witty dialogue’ and get spot-on results. In short, generative AI bridges the unstructured world of content with the viewer’s natural language, making discovery more intuitive.”

Braun envisions, “Imagine a viewer asking, ‘Show me a gripping crime drama set in the 1970s, but without excessive violence,’ and instantly receiving a curated shortlist. These conversational, intent-based interactions move beyond rigid user interfaces, making discovery feel more natural.”

He adds that 24i’s long-term vision aligns with a seamless streaming experience where “AI evolves from recommending content to orchestrating an entire entertainment journey, adapting to a user’s mood, schedule, and environment. This could mean universal recommendation engines across all subscribed services, interactive features letting viewers learn about actors or purchase items seen on screen, and human-centered design that anticipates needs before the user reaches for the remote.”

Avoiding the Personalization Echo Chamber

A major concern is that hyper-personalization could create a narrow content bubble, limiting discovery.

Sang-won Ha shares, “At Vigloo, we optimize personalization while reserving 10–20% of content for diversity. Our recommendations consider not just viewing history but also title popularity, attributes of enjoyed content, and detailed elements like genre and synopsis. This balance ensures users see aligned content while also discovering new genres or unexpected finds.”

Srinivasan notes, “Personalization goes too far when it constrains instead of expands a viewer’s content universe. If someone only sees the same types of shows, you’ve entered echo chamber territory. The goal is a balanced diet, highly relevant suggestions mixed with occasional discoveries that broaden horizons. Effective personalization should help viewers find content they love without making their world feel small.”

Braun concludes, “One of AI’s biggest risks is overfitting, showing viewers only what they already like. The art of recommendation balances familiarity with exploration. We design algorithms that deliberately inject diversity, surfacing content from different genres, creators, and cultural contexts. This keeps viewing fresh, reduces churn, and supports discovery of lesser-known titles.”

Advancements in IP Distribution

Progress in IP distribution enables new levels of customization for both viewers and advertisers.

Franklin highlights, “IP distribution has unlocked unprecedented customization, freeing content providers from satellite ‘world feed’ limitations. With IP-native workflows, a single source feed can be versioned into multiple tailored outputs, integrating local-language commentary, market-specific graphics, regional ad insertion, or even player-focused feeds. Sports leagues, for instance, are creating alternate versions of live events: one feed might highlight a star player for dedicated fans, while another delivers localized commentary and targeted ads to a specific market.”

He argues the benefits are clear: “Viewers enjoy richer, more relevant experiences that deepen engagement, while advertisers and platforms gain new revenue opportunities through hyper-targeted campaigns. Organizations like the World Surf League, Major League Volleyball, and the Tennis Channel are already using IP-native production and distribution to deliver culturally resonant, localized coverage to global audiences in real time. Intelligent, automated IP workflows make it possible to deliver infinitely customizable, cost-effective experiences that grow both audiences and revenues.”

(Source: Streaming Media)