Should You Use an AI Avatar in Training Videos?

Should you use an AI avatar in your training videos? Test it on one real module first.

Here is one test. In 2024, I was an internal video specialist at KnowBe4. Producer, animator, editor, designer. Part of my job was finding out whether AI tools could fit our training pipeline.

I put each tool into a real module before I trusted it with the next.

Synthesia. Good at one thing, a rough edit straight from a script. It wasn't a professional video platform, and it was too slow for real production speed.

The avatar was the bigger problem. About 78% of our 1,800+ employees took a module with it. Roughly 90% of them responded negatively in our instructional design feedback platform.

HeyGen. Synthesia's direct competitor. In our testing, a downgrade.

Google Vids. Not built as an avatar platform, but it can generate them. It showed promise for a hybrid, AI where it earns its place and real production judgment everywhere else.

Veo. Tested as a stock footage replacement. Nothing cleared the bar for professional use.

The no's mattered as much as the yes. Learning that a talking-head avatar loses 90% of its audience is something you want to know before you build a strategy on it.

What the groundwork bought

Two years later, I came back to build the instructional design team's content process. New hires, senior employees, culture, direction. The tooling question was already answered. I knew what AI could carry and what it couldn't.

The team went from three custom training videos a month to more than twelve. Within three months. No meaningful drop in quality.

The avatar failed in one module. That's why it wasn't built into everything after.