How to manage thousands of YouTube videos every year?

Using AI for bulk metadata preparation

As part of managing the LEGO YouTube channels for Europe, South America, East Asia, and the Pacific, we handle thousands of videos every year. Along with them come a lot of metadata – titles, descriptions, keywords, language settings, and other parameters that need to meet both technical platform requirements and brand guidelines. And all of this has to be done in dozens of language versions.

Metadata preparation is one of the most repetitive and sensitive parts of the publishing process, especially when dealing with such large volumes. It’s crucial to maintain consistency across markets, series, and formats, while also ensuring that the output meets SEO and branding standards.

This time, we focused on how to simplify this process – by connecting the tools we already have at our disposal. We tested how we can combine established tools with new technologies, like AI, to see where we can automate and speed up the process. We experimented with a combination of bulk upload (uploading videos in batches along with structured metadata for all at once) and AI (ChatGPT) to test how automation in metadata preparation can save time.

What led us to this

We’ve been managing LEGO’s YouTube channels since 2014 (you can read more about this in our case study). With the growing volume of content and language versions, it became increasingly clear that preparing metadata video by video is not sustainable in the long run.

Although we had already been preparing metadata in bulk for years – manually, using Excel templates, copying, and editing for each language separately – the system, though functional, was becoming more time-consuming and prone to errors.

On YouTube, most of the content we publish regularly is tied to LEGO animated series such as LEGO Ninjago, LEGO Friends, and LEGO DREAMZzz, in formats ranging from short clips (YT Shorts) to full episodes. Each of these series has its own set of text conventions, in addition to the general LEGO brand guidelines – specific character names, locations, and how we refer to the series in titles and descriptions. These names and terms need to be used consistently and accurately across all episodes and language versions, while we also adapt the titles and descriptions for good SEO and channel structure.

There are a lot of rules and requirements to follow for every video uploaded. But since they repeat in every case, the opportunity presented itself to consolidate them into a comprehensive set that can be applied automatically.

‍

How we streamlined the complex process

This approach was part of testing how to most efficiently combine new tools and technologies to automate and speed up metadata preparation, without compromising quality. The combination of machine-driven preparation, which speeds up the process, and steps that we still do manually with natural human oversight, gives us even more confidence that the output is truly high-quality and consistent.

We designed a process that combines:

• Structured data input (Excel with information about episodes),
  
  
• Custom-configured ChatGPT, which generates titles and descriptions according to predefined rules,
  
  
• And an output structure prepared for bulk upload, which can be easily reviewed, edited, and uploaded via the YouTube Content Manager.
  
  

The main benefit wasn’t the technology itself, but the way we set it up to meet the specific needs of the LEGO channels – from required formats and brand guidelines to SEO.

How does this actually change everyday work?

Speeding up and partially automating repetitive steps in the process gave us the ability to:

• Prepare metadata for multiple language versions simultaneously.
  
  
• Ensure consistency across markets and formats.
  
  
• Reduce manual edits and the risk of errors.
  
  
• Minimize reliance on external translations and proofreading, and the time we had to wait for them.
  
  

This frees up the team’s time to focus on work that adds greater value – creativity, quality control, and strategy.

What did we learn?

This experience showed us that even commonly available tools can bring real savings when connected in a smart way. It wasn’t about showcasing what AI can do; it was about figuring out where it truly helps.

For tasks that are large in volume, repetitive, and require consistency, it makes sense to find a system that simplifies and speeds up the work – without losing control over the results.

Interested in streamlining your content management?

If you're dealing with similar tasks – large amounts of content, multiple language versions, or repetitive steps – we’d be happy to take a look at your case. You might already have everything you need; it just takes thinking about how to connect it more effectively.