AI is creating a new commercial role for entertainment metadata: keeping machines current.
A model can ingest a catalog during training. The catalog keeps moving after the training run ends.
New episodes arrive. Rights windows close. Titles switch services. Sports schedules change. Deep links break. Credits get corrected. Availability shifts by territory and business model.
That gap is opening a recurring business around continuously maintained metadata delivered through APIs, knowledge graphs, and increasingly AI-native connections such as Model Context Protocol servers.
AI has exposed the economic value of structured entertainment data. Runtime grounding adds another layer to the economics. The opportunity extends beyond licensing data that helps build a model. It includes licensing access to the information that keeps the model useful after it’s built.
Training Licenses the Snapshot. Grounding Licenses the Update
Gracenote sued OpenAI in March, alleging that OpenAI used copyrighted entertainment descriptions, identifiers, and related material without permission to train its AI systems. OpenAI said its models are trained on publicly available information and that its practices are grounded in fair use. Gracenote’s complaint also says the company licenses its material to AI providers for training and alleges that OpenAI’s conduct threatens that licensing market. The claims haven’t been adjudicated. The case was stayed on April 21 pending summary judgment rulings in other cases within the broader OpenAI copyright multidistrict litigation.
Gracenote’s product strategy points to another commercial relationship after the model has been built.
The company launched its Video MCP Server in September 2025, connecting LLMs to Gracenote’s continually updated entertainment knowledge base. Gracenote says the product can validate and enrich AI responses in real time while supplying program imagery, availability information, standardized content IDs, deep links, and the relationships required for conversational search and recommendations.
A training license covers a defined use of data during model development. Runtime access can create value every time an application needs information that has changed since training.
AI Raises the Value of Data That Gets Old
Some entertainment metadata has a very long shelf life.
The Godfather was released in 1972. Marlon Brando played Don Corleone. Those facts aren’t carrying much expiration risk.
The commercial information surrounding content behaves differently.
Rights expire. Availability changes. New episodes appear. Streaming destinations move. Live schedules change. Localized versions launch. Deep links stop working.
Metadata carries identity, rights, operational instructions, discovery signals, and commercial rules through the streaming supply chain. The video can remain unchanged while the information required to distribute and monetize it changes overnight.
Gracenote recently tested an ungrounded LLM against 1,300 popular TV episodes across 13 countries. The company reported that the model returned no title, description, actors, or genre for 604 episodes, equal to 46.5% of the sample. Across 100 U.S. episodes, Gracenote measured average description accuracy at 27% against its grounding data. In one Stranger Things example, the model combined details from multiple episodes.
Gracenote conducted the research using its own data as the reference point, so the results represent a company-produced test rather than an independent benchmark of the broader LLM market.
An LLM can interpret a request, generate natural language, and infer relationships from its training. An entertainment product still needs authoritative information when the answer depends on a specific episode, current availability, an accurate content ID, or a working destination.
Runtime Metadata Moves Closer to the Viewer Transaction
Metadata has always fed search, recommendations, guides, scheduling, rights systems, advertising, and distribution.
AI can move that data closer to the moment when someone decides what to watch.
A viewer can ask for the Friends episodes featuring Bruce Willis, the Christmas episodes of a favorite sitcom, or where the hell the Michigan-Rutgers game is actually streaming tonight. The language model handles the intent. The underlying data has to resolve the title, episode, person, relationship, availability, and destination.
Tubi’s integration with ChatGPT already shows how conversational interfaces can become another entry point into streaming catalogs. When the interface capturing viewer intent sits outside the service fulfilling it, clean identifiers, current availability, and accurate mappings become connective tissue between discovery and playback.
Google renewed its Gracenote partnership in February, with the company’s metadata supporting up-to-date entertainment information across Google consumer products and AI experiences. Samsung expanded its relationship with Gracenote later that month to support LLM-enabled search and discovery on Samsung smart TVs, along with AI-driven operational use cases. Financial terms weren’t disclosed.
IMDb serves the same underlying need through a different delivery model. Its commercial GraphQL API provides real-time access to title and name datasets through AWS Data Exchange. IMDb explicitly positions the API around receiving updated data as it becomes available rather than waiting for the delay associated with bulk files.
When the product depends on current information, yesterday’s database has less value.
Deeper Integration Raises the Value of the Data Relationship
When metadata feeds a live AI experience, the supplier relationship expands beyond delivering fields. Coverage determines which questions the system can answer, freshness affects whether those answers remain accurate, identifiers resolve the correct asset, availability connects discovery to something the viewer can actually watch, and deep links turn that answer into an action.
That moves metadata closer to runtime infrastructure. Buyers still care about data quality, but update cadence, geographic coverage, provenance, uptime, interoperability, and integration with internal systems become part of the product.
OpenAI added remote MCP server support to its Responses API in May 2025, allowing developers to connect models to outside data and tools through a standardized protocol. Gracenote built its Video MCP Server around the same open standard.
Easier connectivity can expand the market for specialized data by reducing the technical friction required to plug trusted sources into AI applications. It also shifts more of the competition toward the underlying data itself. When multiple suppliers can connect through a common protocol, differentiation depends more heavily on coverage, freshness, enrichment, rights-cleared assets, relationships, availability, identifiers, and the operational work required to keep those records accurate.
The Premium Sits in the Hard-to-Maintain Data
AI doesn’t turn every entertainment fact into a premium product.
Basic title information exists from multiple sources. Some of the industry’s most useful metadata infrastructure is intentionally built to circulate widely.
EIDR provides persistent identifiers and essential metadata through a public audiovisual registry. Its registry metadata is unlicensed and freely usable downstream, while participating organizations pay annual fees for services including registrations, unlimited API usage, support, deduplication, and governance participation.
A common identifier becomes more useful as more companies recognize it, while a continuously maintained enrichment layer creates value through the work required to keep information complete, reconciled, current, and usable across systems.
“Jaws is a movie released in 1975” doesn’t require much proprietary infrastructure.
Determining which version of Jaws a particular viewer can watch right now, under which business model, through which destination, with the correct content mapping across every system involved in the transaction requires considerably more.
The stronger recurring opportunities sit around metadata with expensive characteristics: frequent changes, editorial enrichment, rights information, difficult entity relationships, conflicting source records, global localization, current availability, and links that need to keep working.
Machine Consumption Creates More Ways to Package Metadata
Metadata companies already sell data through feeds, APIs, identifiers, enrichment services, imagery, rights information, availability products, recommendation tools, and operational systems. AI expands how those capabilities can be packaged and consumed.
One customer may need a bulk catalog for internal ingestion, while another needs a real-time API, structured relationships for conversational discovery, availability and deep links, or permission for an AI application to query selected portions of a knowledge graph without receiving the underlying dataset.
The same information can support different products depending on how frequently it changes, how the customer accesses it, and what the application needs to do with the answer. Access, maintenance, verification, and service quality can therefore carry commercial value alongside the underlying record.
For suppliers, that expands the use cases for data they already spend heavily to collect, normalize, enrich, and maintain. For buyers, it sharpens the difference between information that can sit inside a static catalog and information that benefits from a persistent external connection.
The Streaming Wars Take
The companies maintaining entertainment data still have to identify titles, reconcile records, enrich catalogs, track availability, map relationships, manage imagery, and keep information current across markets. AI increases the number of products that can consume that work and shortens the distance between the data supplier and the viewer experience.
The opportunity won’t distribute evenly across the metadata stack. Shared identifiers and stable facts have different economics than current availability, rights intelligence, deep links, proprietary enrichment, and continuously maintained knowledge graphs.
Streaming companies also need control of the canonical rules that define their own catalogs. Specialist suppliers can extend, enrich, verify, and connect that foundation at a scale most individual media companies won’t replicate efficiently.
The upcoming TSW Guide to Metadata will examine how catalog quality, ownership, rights, discovery, advertising, and AI all depend on the same underlying data discipline.
So while the AI interface gets the attention, keeping its answer attached to reality is becoming a business of its own.
The Streaming Wars is intentionally ad-free
We don’t run display ads. Not because we can’t, but because we don’t believe in them.
They interrupt the reading experience. They cheapen the work. And they burn advertisers’ money on impressions nobody actually wants.
So we chose a different model.
We say the things people in this industry are already thinking but don’t say out loud. We connect the dots beyond the headline and focus on explaining why things matter to the people working in this business.
If you believe industry coverage can exist without clutter and interruption, you can support it here → SUPPORT TSW.
Support is optional. But it directly funds research and continued coverage — and helps prove this model can work.
Support TSW →
