I didn't want this to be true.

Earlier this year I was researching AI avatar videos for a client. These are tools where a generated presenter reads your script on camera, so you can make a training video without filming anyone. I expected the research to back up what I already believed: people learn better from a real person. It didn't. In the studies I found, learners scored about the same on tests whether a human or an avatar presented the material. I had to change my mind.

The tools are also improving quickly. On 9 October HeyGen launched its own voice model, and it says its platform is used across 85% of the Fortune 100 (PR Newswire).

If you run a training business, you've probably wondered about this. Filming takes days, and a video can be out of date as soon as a regulation changes. The worry is that learners will notice, switch off, or trust you a little less.

What the studies found about AI-generated training videos

Three controlled studies stood out. In each one, people watched the same material presented either by a human or by an AI avatar, then took a test.

• ETH Zurich (2025). 447 participants watched short teaching videos on production management, some presented by a human and some by an AI avatar. There was no significant difference in exam results (Computers & Education).

• Hacettepe and Bartın universities (2024). 108 undergraduates watched 10-minute lecture videos over two weeks, with either an AI-generated or a human instructor. The instructor type made no significant difference to their performance (IRRODL).

• Hochschule Bielefeld (2026). Engineering students watched a Synthesia avatar and a human explain fuel cells, in random order. Learning gains were similar (JMIR Formative Research).

There are limits. The Bielefeld study had 13 students, and its authors say it isn't proof the two formats are equal. The ETH Zurich participants were paid online workers, not students on a real course. But all three point the same way, and that's what made me change my mind.

Where learners did notice the difference

Test scores held up, but enjoyment didn't. In all three studies, people preferred the human.

In the ETH Zurich study, participants enjoyed the human-made videos more (Media & Learning). In the Hacettepe and Bartın study, engagement was lower with the AI instructor, and some students described it as distracting and uncomfortable. That mattered less when the topic interested them or they were watching to learn something specific. In the Bielefeld study, the human presenter scored higher on every user experience measure.

I think this is the part a training business should care about most. A test straight after a video tells you whether people took it in. It doesn't tell you whether they'll come back for module six. My guess is that fewer people finish a course they enjoy less, though none of these studies measured that.

Where I'd use an AI presenter, and where I wouldn't

The biggest gain is speed. In the ETH Zurich study, the human-made videos took several days to produce and the AI versions took a few hours (Media & Learning). When the script changes, you edit the text and generate the video again instead of booking another shoot.

So I'd use an AI presenter for content that changes often or needs to exist in several languages:

• Regulation and policy updates

• Product and software walkthroughs

• Short explainers that sit inside a longer course

• Translated versions of a course you've already filmed

I'd keep a real person wherever trust is the point. That's the welcome video, the expert whose name sells the course, coaching and feedback, and sensitive topics in healthcare or finance, and things like that.

You don't have to pick one. A course can open with you on camera and use an avatar for the modules you update every quarter. And if filming yourself is what's stopping you, it's probably cheaper than you think. We wrote about why your iPhone might be better than a studio.

What it means for you

If you've been holding off on AI video because you thought learners wouldn't learn from it, the evidence says they probably will. That was the part I got wrong.

The trade-off is enjoyment. People still prefer a real person, and that preference showed up in every study I read. I'd use avatars to keep content up to date and keep people on camera where the relationship matters.

Whichever you choose, your LMS should let you swap a video in one place without rebuilding the course around it.

If you're planning how your next course gets made, talk to our founder.

FAQ

Are AI-generated training videos as effective as videos with a human presenter?

On short-term test scores, studies have found no significant difference. Learners still rate human-presented videos higher for enjoyment and user experience.

Do learners prefer human presenters to AI avatars?

Yes. Across the studies above, people enjoyed the human versions more, even when they learned the same amount.

When should I use AI avatars in training?

They work best for content that changes often or needs translating, such as policy updates and product walkthroughs. Keep real presenters where trust and connection matter.

Will AI avatars lower course completion?

The studies didn't measure completion. Lower enjoyment is a reasonable warning sign, so it's worth tracking where learners drop off after you add avatar videos.

Can I mix AI avatars and real presenters in one course?

Yes. One option is a real person for the introduction and key moments, and an avatar for modules you update regularly.