Great content may attract subscribers, but it takes more than that to keep them engaged and coming back for more.

Empowering consumers to find what they’re looking for effortlessly is one crucial element of creating an engaging media and entertainment (M&E) experience. Unfortunately, many consumers don’t think M&E companies have cracked the code — a Deloitte survey found that almost half of consumers report abandoning an entertainment experience because they couldn’t locate the content they wanted. The same research indicates that 60% of younger audiences often turn to social media for content recommendations because its algorithms serve up content they’re interested in based on their personal preferences.

Traditional AI capabilities haven’t yet enabled M&E companies to fully realize their vision of highly personalized and seamless experiences, however, generative AI (GenAI) is now putting the opportunity within reach.

GenAI, powered by large language models (LLMs) like Cohere Command R and Embed, can streamline as much as 60 percent of the work of searching, synthesizing, and validating information from a company’s content catalog to provide deep insights about subscriber preferences and interests. This can have a profound impact on a variety of M&E strategies, from how companies manage and promote their content to how they leverage team resources.

In this article, we share four ways M&E companies are using LLMs to improve and personalize search and discovery: surfacing the right content, improving personalization and localization, enabling more dynamic engagement, and streamlining research for planning and developing new content.

Sample Use Cases for Media and Entertainment

Surfacing the Right Content with Intelligent Archives

When content isn’t organized and labeled effectively, it’s much harder for M&E companies to find and share it. The industry has long invested in AI-driven tagging and discoverability capabilities to tackle this problem. But traditional AI tools typically categorize content based only on predefined themes or specific keywords. While this approach is a start, it often falls short because it fails to identify subtle differences between similar content, such as in tone or sub-genres, that would create more intelligent content archives.

Using LLMs, content curation teams at M&E companies can gain a deeper understanding of articles, scripts, and other content filled with unstructured text to enable far more accurate and nuanced categorization. Embedding models, like Cohere’s Embed, for instance, can capture the semantics, or meanings and contexts, of words and phrases (bonus: they achieve this without receiving explicit training). For example, embedding models can achieve a higher level of precision to accurately identify an article that mentions “a look at government policies” as investigative journalism, even if it doesn’t expressly cite an “investigation.” Adding in multilingual embeddings can enable content curation teams to better tag content for distribution in specific geographies. They can, for example, identify slang or colloquialisms in an article that may indicate its relevance in certain countries.

Not only can LLMs identify nuances in data, but they can do so rapidly across massive volumes of data. By shaving off considerable time in content categorization, LLMs enable content curation teams to tap decades of archives that consumers may find valuable but would have otherwise collected digital dust. Plus, adding semantic understanding to search makes it easier to find content with similar subtle attributes, like mood or social relevance. As a result, content curators — and even marketers, production teams, and subscribers — spend far less time than the ten minutes it can take to surface the content they seek.

Soon, advanced multi-aspect embeddings (such as those available with Cohere Compass), will power even more refined classification capabilities. They’ll be able to capture insights from data with multiple dimensions, for example, article content along with its corresponding clicks and comments. This will enable organizations to better analyze long-term consumer behavior patterns, such as the time of year or day that a user accesses a particular piece of content.

Personalizing Recommendations — Even for Global Audiences

Categorizing data in ways that users find helpful is one thing, but determining exactly what a particular user values presents a different challenge. Many media consumers experience some level of browsing fatigue as they sift through curated content that doesn’t quite hit the mark. In fact, one study found that more than half of consumers surveyed said that recommended content didn’t match their interests.

Why might a consumer watch trailer after trailer of “recommended movies” without finding anything that appealed to them? Or endlessly scroll through suggested articles they received from a favorite news outlet without finding something they wanted to read? It’s because traditional machine learning systems often rely on structured and numerical data — such as user histories, ratings, and basic classification preferences — to figure out what they like. This structured data provides a good starting point, but misses the user’s “intent,” which is often hidden in unstructured data.

Using LLMs, M&E companies can incorporate far more detailed contextual information inferred from sentiment analysis, user comments, and real-time behavioral patterns to enhance existing recommendation systems. For example, LLMs can understand that a reader enjoys not just tech stories, but specifically those that highlight positive uses of new capabilities. Some M&E organizations even use LLMs to personalize marketing text, so that it appeals to a subscriber’s evolving interests. They then localize it using LLMs with multilingual capabilities. For example, one M&E company in particular expects to take localized personalization a step further by combining its own user data with external data about a region. Doing so manually would be prohibitively expensive and leave global users with a lackluster experience.

Boosting Consumer Engagement with AI Assistants and Better Content Moderation

Today’s consumers have exponentially more media options than a decade ago. One way M&E companies seek to stand out in this flooded and fragmented market is by enhancing user engagement with conversational AI assistants that understand the intent and context of user queries. One news organization is building an AI assistant to supercharge its search experience using Cohere Command R with retrieval-augmented generation (RAG). This will allow subscribers to ask questions about current news events and receive direct answers accompanied by citations and source links that help them delve deeper into the subject.

Content generated by media companies also increasingly competes for attention with user-generated content (UGC). M&E companies with a wise “can’t beat ‘em, join ‘em” mentality are catering to this trend by deploying AI assistants that enable individuals to build their own content via conversational prompts. A sports entertainment company, for instance, is creating a conversational AI assistant with Cohere Command R and Embed for users to develop player profiles based on published reports. This can foster deeper engagement and loyalty by allowing users to interact with the company’s content in more meaningful ways. It can also attract new audiences as users share their new content far and wide.

When such content is combined with community-building activities, like chats or commenting, engagement can increase even more. But the volume of content and contextual nuances of online conversations can make it difficult to effectively moderate such forums on a global basis. LLMs with multilingual capabilities provide crucial capabilities that can vastly improve content moderation efforts. For example, they can identify comments that could be viewed as culturally insensitive in a given region solely based on the context or cultural norms. As a result, companies can effectively expand engagement with global audiences while feeling confident that worldwide readers will be able to engage in a safe and appropriate manner.

Streamlining Research for Content Planning

So far, we’ve talked about making existing content easier to find, localize, promote, and search. But improved search and discovery capabilities also help M&E companies explore, ideate, and develop new content for consumers. They do so by eliminating the need to manually sift through past research, interview notes, program transcripts, and other materials in search of a key fact or trend. Conversational AI knowledge assistants can free staff — journalists, producers, content strategists, and the like — from tedious research that cuts into creative time. A movie producer, for instance, might use an AI research assistant to compare a new script with blockbuster scripts in studio archives to see if it includes elements that contributed to past success. One media outlet is building an AI assistant using Cohere Command R with RAG that will enable its journalists to quickly locate stats, quotes, and key trends from the organization’s archives.

In the near future, we anticipate that M&E companies will extend such AI knowledge assistants with agentic capabilities that can manage complex workflows for their teams, for example, by automatically reviewing, analyzing, and cross-referencing documents in public databases with internal documents, or calculating statistics from multiple third-party sources to flag newsworthy trends for journalists.

Increasingly, M&E companies are viewing GenAI as a competitive differentiator. They’re recognizing that the technology’s ability to advance content search and discovery are essential for delivering highly personalized, contextually relevant content that transforms the user experience into one that’s differentiating, delightful — and sticky.

Is your organization among those who’ve begun putting LLMs to work? If not, now is the time to get started.

Get Started!

Craig Meadow