Gracenote, the content intelligence business unit of Nielsen, unveiled the next phase of its artificial intelligence product roadmap on September 9, 2026, from New York. The plan rests on three additions that sit on top of the company's existing Video Model Context Protocol Server: a dedicated Sports MCP Server, an entertainment-focused small language model, and task-oriented agents. According to Gracenote, the components are designed to be composable, meaning they can be used on their own or combined for different use cases, which the company says reduces the need to build each AI implementation from scratch.

In Short

Gracenote keeps a giant, checked catalog of facts about movies, TV shows and sports, and it is now selling more tools that let AI systems look up those facts instead of guessing. This matters because AI often invents details when it answers questions, and wrong data flows straight into what viewers see and what advertisers buy against. The practical change is that a company building an AI feature can plug into Gracenote's verified data, a smaller specialist model, or automated helpers rather than trusting a general chatbot's memory.

Three layers, one stack

The announcement maps to three layers of what the company calls the AI stack, each tied to a priority Gracenote says it hears repeatedly from customers: deeper audience engagement, sharper programming analysis, and more efficient media operations.

The first layer covers domain-specific data access and prebuilt tools. This is the part that is available today. The Video MCP Server connects a customer's preferred AI models to Gracenote's source-verified metadata, which the company states covers more than 55 million titles alongside continually updated viewing availability data. Underpinning that catalog is a global entertainment knowledge graph that maps relationships among programs, cast and crew. Persistent, portable content identifiers link records across sources and services, so a single program can be recognized consistently no matter where it appears.

The planned Sports MCP Server broadens that foundation into live sport. According to Gracenote, it will carry data on live, upcoming and recent events, including schedules, scores, standings, statistics and where-to-watch information. Read together, the company frames the two servers as the basis for more reliable conversational search and personalized recommendations, helping audiences find content, locate games and get answers about specific teams and athletes.

The second layer is specialized reasoning, and this is where the entertainment-focused small language model enters. Gracenote positions the model for high-volume workloads where low latency and predictable costs matter most. Paired with the company's data and tools, potential applications named in the announcement include evaluating acquisition and licensing opportunities, identifying catalog gaps, and analyzing availability windows.

The third layer is task execution, handled by agents. According to Gracenote, agent capabilities will combine AI models and the company's tools with customer-defined business logic and guardrails, automating discrete workflows that face audiences or run operations behind the scenes. The announcement lists content identity resolution, catalog reconciliation, feed enrichment, image selection, schedule maintenance and availability updates among the candidate jobs. One worked example goes further: a rights-monitoring agent could scan the open web for full episodes tied to a customer's catalog, match them to Gracenote identifiers, and flag potential violations for legal review.

The accuracy problem the roadmap is built around

The company grounds the entire plan in a measured failure rate rather than a general worry. Gracenote's June 2026 study evaluated the output of a large language model relying solely on its training knowledge across 2,600 popular movie and TV titles in 13 countries. For 506 of those titles, close to one in five, the model fabricated information for every metadata attribute the researchers examined. That figure works out to 19.5% of the sample, a result PPC Land reported when the study was published, noting that the comparison pitted an ungrounded client drawing only from training data against one querying Gracenote's video dataset through a Video MCP Server.

The industry vocabulary for that behavior is hallucination, the situation where a model produces fluent, authoritative-sounding output that does not correspond to fact. The distinction Gracenote draws is between a model working from a fixed snapshot of training data and one fetching structured, continuously updated records at the moment a question is asked. Sport sharpens the problem further, because live schedules change, scores are produced in real time, and league structures shift across seasons, so a model trained months earlier has no reliable way to answer a question about tonight's fixtures.

Consumer research the company has published points the same direction. An April 2026 Gracenote study found that three in four United States consumers verify AI chatbot answers about entertainment rather than taking them at face value, a trust gap that lands precisely on the search and discovery experiences the MCP servers are meant to feed.

Two executives frame the strategy

Jared Grusd, chief executive officer of Gracenote, set out the company's reading of the trajectory. "AI adoption in media will be evolutionary. Its impact will be revolutionary," he said. "Over the next few years, the industry will progress from optimizing established processes to solving challenges previously too difficult or costly to address - and ultimately to redesigning how programming is acquired, managed and delivered. Gracenote's role is to provide the trusted content intelligence infrastructure that enables incremental advances to compound into industrywide transformation."

Tyler Bell, senior vice president of AI products at Gracenote, argued that no single assistant can do the whole job, which is the reasoning behind splitting the roadmap into separate parts. "Media companies cannot capture AI's full value with a single assistant designed to do everything," he said. "Search, catalog matching and schedule quality are fundamentally different problems, each with its own data, latency and oversight requirements. Putting AI into production requires clearly defined tasks, measurable outcomes and human checkpoints where they matter most. Gracenote's established content foundation, deep media expertise and purpose-built tools uniquely position us to help our customers realize the transformational benefits of AI."

How the pieces got here

The infrastructure this roadmap extends did not appear with the September 9 announcement. Gracenote first launched its Video MCP Server on September 3, 2025, connecting large language models to its entertainment database to validate and enrich their responses. The Model Context Protocol itself originated at Anthropic in November 2024 and was later donated to a Linux Foundation fund; it standardizes how a language model discovers and calls external tools, and advertising infrastructure adopted it quickly, a pattern PPC Land has tracked as platforms across the ecosystem exposed their systems to third-party agents through MCP servers.

The distribution footprint has widened in parallel. Gracenote's metadata now reaches connected TV platforms directly: Tubi and FOX One gained access to Gracenote metadata spanning 55 million titles, extending the same catalog the MCP servers query into consumer-facing apps.

On the advertising side, the metadata layer has moved steadily into the buying stack. Gracenote launched Content Connect on December 4, 2025, giving agencies, brands, supply-side platforms and demand-side platforms direct access to standardized program-level metadata for CTV targeting. Before that, Index Exchange became the first supply-side platform to embed Gracenote contextual intelligence in September 2025, carrying content identifiers and brand-safety controls into programmatic infrastructure. More recently, the taxonomy reached a demand-side platform for the first time, when The Trade Desk gained show-level CTV data in what the companies described as Gracenote's first DSP deal, letting buyers apply Gracenote content identifiers to set targeting that moves beyond genre.

The sports angle has a documented backstory

The Sports MCP Server is the most concrete new piece, and Gracenote has been building toward it. In December 2025 the company expanded its On Sports platform, linking sports documentaries and shoulder programming to live events across 160 leagues in more than 50 countries, a move framed around rising streaming sports viewership. Discovery friction in sport is measurable: Gracenote research found that 26% of sports fans cannot find the games they want to watch, the kind of gap that live schedule and where-to-watch data is meant to close.

Supply of sports content is growing faster than the channels carrying it. Gracenote analysis published on August 20, 2026, found sports programming on free ad-supported streaming television expanding roughly two and a half times faster than the number of sports channels distributing it, with individual games and events up 37.5% as channel counts rose 13.8%. That growth runs into a data quality problem the company has quantified before: an October 2025 Gracenote survey found that 68% of sports programming on FAST channels lacked imagery, 55% was missing original air date information, and 39% had no episode title, gaps that block both contextual advertising and reliable discovery.

Why this matters for the marketing community

For media buyers and platform operators, the roadmap is less a single product than a signal about how AI features will be assembled in the near term. The composable framing matters commercially. Rather than a media company training its own model or trusting a general assistant's memory, the pitch is to plug verified data, a specialist model, and automated agents into existing workflows as separate, swappable parts. That approach maps onto a broader argument Nielsen has been advancing beyond entertainment metadata: the company recently extended the same grounded-versus-ungrounded case into advertising intelligence with Ad Intel AI, exposing a real-time view of 5.5 million brands through an MCP interface.

The advertising relevance is direct. Program-level metadata is the same raw material that powers contextual CTV targeting, and the accuracy of that metadata determines whether a buyer can reliably match an ad to the right show or exclude the wrong one. As sports supply expands and home screens become addressable inventory, the incentive to close metadata gaps grows, and the entity that standardizes those identifiers occupies a structurally central position between content owners, platforms and advertisers. The roadmap keeps Gracenote in that position while the discovery layer shifts from keyword search toward AI-driven answers.

There are limits worth stating plainly. The Sports MCP Server, the small language model, and the agents are planned rather than shipped; only the Video MCP Server is described as available today. The accuracy figures come from vendor-published research, drawn from a company that sells the data grounding it recommends, so the direction of the finding and the direction of the commercial interest align. Gracenote's own executive framing acknowledges the gating factor: putting AI into production, in Bell's words, requires clearly defined tasks, measurable outcomes and human checkpoints. Whether the roadmap's later stages deliver on latency and cost in practice is not something the announcement resolves.

Where the company shows the work

Gracenote said it will display the Video MCP Server, offer an advance look at the Sports MCP Server, and take part in a panel titled "Driving True Transformation vs Incremental Change with AI" at IBC2026, running September 11 to 14 at RAI Amsterdam. The company describes itself as standardizing how the global media and entertainment ecosystem indexes content and associated metadata, with coverage it puts at more than 55 million titles and over 80,000 channels and catalogs, spanning more than 70 languages across over 80 countries.

Timeline

Summary

Who: Gracenote, the content intelligence business unit of Nielsen, with statements from chief executive officer Jared Grusd and senior vice president of AI products Tyler Bell.

What: An expansion of its AI product roadmap adding three components on top of the existing Video MCP Server: a dedicated Sports MCP Server, an entertainment-focused small language model, and task-oriented agents, all designed to be used independently or combined.

When: Announced on September 9, 2026, from New York, ahead of demonstrations at IBC2026 running September 11 to 14.

Where: New York, with the products aimed at media companies, streaming platforms and advertisers globally, drawing on a catalog the company puts at more than 55 million titles across over 80 countries.

Why: To address the accuracy, latency and cost barriers that keep general-purpose large language models out of production, grounding AI features in source-verified metadata rather than a model's training data, after research found an ungrounded model fabricated all metadata for 19.5% of tested titles.