Why Not Just Use a Generic LLM To Improve Content Discovery? Here's What We've Learned

A question we often hear from video services is “Why should we buy video analysis solutions from Media Distillery rather than get one or more Generic LLMs and do it ourselves?” It’s a fair question and it’s one we ask ourselves all the time to make sure we’re offering added value to our customers. 

With foundation models like large language models (LLMs) now widely available, it might be tempting to try building your own discovery features on top of a general-purpose model like GPT or Gemini. In fact, we encourage you to try! What we think you will find, in line with our findings based on extensive experiments and usage, is that for many of the complex customer-facing video service applications, off-the-shelf LLMs are not (yet) up to the task. They can’t deliver reliable, predictable results that can be used automatically. Let’s look at why. 

Toddler Taming: The problem with Generic LLMs

Anyone who’s spent more than a few minutes playing with Chat GPT knows that it’s not 100% reliable. For any LLM, results are typically plagued by the following issues:

Hallucinations - they make up stuff that sounds correct but isn’t

Inconsistency - they may deliver the same good answer 8 out of 10 times they’re asked the same question, but 2 times out of 10, the response is bad

Poor reasoning - they can rely on incorrect assumptions that deliver flawed results

To put it another way: models behave like an intelligent, stubborn child: they will understand what you’re saying, but there’s no guarantee that they will listen to or follow your instructions 100% of the time. Sometimes it’s great. But when it’s bad it can be truly awful! And the more complex the task you're trying to achieve, the more prevalent these issues become. The question is whether your organization wants to invest time and resources to refine the LLMs to a point where they’re predictably delivering the level of quality you need. Or would you rather put the task in the hands of a partner who has been providing fully-automated AI-powered solutions trusted by many video services since 2018.

Deeper Understanding, At Lower Risk and the Right Price

Media Distillery’s Deep Content Understanding Platform allows us to reap the benefits of all that LLMs offer, but make sure quality meets your high standard for production. One of the ways we do this is by simplifying the problem that needs to be tackled by an LLM. We reduce the complexity of the task at hand as much as possible to drastically reduce risk of error. For example, by relying on our more traditional Machine Learning solutions that we built over the years to do the first analysis of the video. This often lets us discard large parts of the video that aren’t relevant for the task at hand. Only after that step do we leverage the extensive knowledge provided by LLMs. This combination of the best from both traditional ML and generative AI technologies enables us to reach a consistently high performance and quality threshold from our products.

An additional element that makes home-grown solutions based on generic LLMs unattractive is the operational costs and expertise required to maintain such solutions. The video services we speak to have often underestimated these aspects. It’s easy to get started with a LLM Proof-of-Concept, but do you have the in-house capacity, skills and budget to maintain a production-level solution? Or would you rather work with a partner that’s fully focused on this area and can also bring in benefits learned from working with other customers?

Solving Discovery Starts With Understanding Your Content

Great discovery experiences don’t begin with AI. They begin with understanding your content and your audience. At Media Distillery, we believe the job isn’t to build the most advanced AI for its own sake. It’s to solve real problems for video services: helping users find what to watch, reducing churn, and increasing the value of your content catalogue. That means understanding not just the language of your programs, but the structure, the context, and how viewers engage with them. It means working with content rights, editorial sensitivities, and UI constraints. And above all, it means delivering reliable, scalable outputs that improve the user experience—without adding operational burden.

It’s tempting to experiment with general-purpose LLMs for these tasks, but the risk often outweighs the reward. With the right partner, you get:

• Deep content understanding – what’s being said, who’s saying it, what matters, and when
• Viewer behavior insights – how your users search, scroll, skip, and select
• Scalable infrastructure – that processes large volumes of video and metadata in real time
• Smart workflows – that fit your tech stack and editorial needs
• Experience across deployments – so you benefit from lessons learned elsewhere

That’s what we’ve built at Media Distillery. Our Deep Content Understanding Platform isn’t just AI—it’s a purpose-built engine for discovery, shaped by years of hands-on experience with broadcasters, telcos, and streamers.

Why It Works

Because it’s designed for this domain, our approach delivers:

• Faster implementation
• Higher accuracy
• Less editorial rework
• More actionable metadata and previews
• Consistent improvements from every new customer deployment

So yes, AI plays a role. But it’s only useful if it’s delivering on your business goals—and that’s what we’re here to help you do.

Want to Know More?

Download our e-guide “Navigating OTT’s AI Hype: Five Key Takeaways from Five Years of Hands-On Experience” to see what makes specialist AI work for video. And watch out for a new e-guide “The New Playbook for Content Discovery” that’s coming soon!