Not Every Training Video Should Use an AI Avatar

Most of the conversation about AI video is framed around a single threat: whether synthetic presenters will replace the people who currently stand in front of the camera. That’s a fine question for a headline, but it isn’t much help if you’re the one who has to ship training content this quarter. The more useful question is when you should use an AI avatar and when you shouldn’t.

The answer changes depending on what you’re making. A software walkthrough and an open enrollment announcement are both training videos, but they don’t carry the same requirements. That’s why blanket policies in either direction tend to fall apart when applied to a real content calendar. 

The reality is that you don’t want everything to be an avatar, but plenty of things can be. The skill worth building is knowing the difference before you commit a production schedule to one approach.

That decision comes down to four things: what the video really costs you, what kind of content it covers, what risks each option carries, and what the research actually shows.

Key takeaways

• High-turnover content, like software updates, is often well-suited to AI avatars; sensitive human resources (HR) or benefits topics still require a human presence.
• Human on-camera instructors boost learner preference, not test performance, but that research predates AI avatars entirely.
• Camtasia’s own study found that an avatar in a picture-in-picture format increased quiz retention by about 10 percentage points over other formats. The result is promising, but not yet conclusive.
• The true cost of on-camera videos includes scheduling, reshoots, and hours of talent time, not just the production budget.
• Human talent carries turnover and liability risks, while AI avatars introduce vendor dependency and the possibility of avatar retirement. Neither option is risk-free.

The real question isn’t whether to use AI, it’s when

Ask a training team whether a given video should use an AI avatar or a real person on camera, and you’ll usually get one of two confident answers. 

The first treats AI video as a straightforward substitution, where you feed it a script, generate a presenter, and retire the studio. The second insists that learners can always tell the difference, so synthetic presenters erode trust and don’t belong in serious training. Both make the same mistake by treating “training video” as a single category when it very much isn’t.

Consider a 90-second demo of a changed settings menu alongside a video explaining a benefits change to 4,000 employees. These videos serve different purposes and carry different risks. The demo falls short when it goes stale. The benefits video falls short when its delivery undermines trust in the message. No single policy serves both well.

Once you recognize the distinction, the choice stops being philosophical and becomes a per-asset decision driven by content type and stakes. Some videos are obvious candidates for an avatar or a human presenter. The more difficult cases fall somewhere in between, and that’s where a framework is more useful than a blanket opinion.

What traditional on-camera video actually costs you

Plenty of teams underprice on-camera video because they count only the visible line items, such as talent fees, studio time, and perhaps an editor. Those costs are real, but they rarely capture the full investment.

The less visible cost is coordination. First, you need to find a window when your subject matter expert (SME), the production team, and anyone involved in approving the script are available. If a customer escalation or another priority disrupts that window, the scheduling process starts again.

There’s also a productivity cost that never appears in the video budget. When you pull an engineer or benefits manager in front of a camera, you’re taking them away from the work they were hired to do. A four- to six-hour recording session can consume most of someone’s working day before you account for preparation or rehearsal. 

The demands of the session itself also affect the presenter.

“That could be hard on the talent as well,” said Matthew Pierce. “Imagine speaking for four to six hours, trying to be on, saying things correctly, and reading a teleprompter while trying to sound like you’re not reading. So there’s a lot of cost there.”

Multiply that time across several SMEs, and the internal cost becomes much harder to dismiss. The production budget may remain modest while the organization absorbs hours of scheduling, preparation, and lost SME time.

Scheduling, shoots, and the reshoot tax

Even when everything lines up and the initial shoot goes smoothly, the footage has a shelf life that nobody puts on the production calendar. A product update ships, a policy changes, or a presenter wears a different outfit for a pickup shot three weeks later. Now you’re either living with a visible continuity break or rebooking the session.

Reshoots are the hidden tax on on-camera video. They don’t show up in the original budget because nobody plans for them, but they happen often enough that any team producing videos at scale has felt the sting. 

“With AI, you don’t have to worry about the avatar wearing a different shirt that day, whether they shaved or didn’t shave, whether their hair is different, or whether they did their makeup differently,” Pierce said. “You can redo and update any piece of the video consistently without all the hassle.” 

Every reshoot also reopens the scheduling bottleneck the team already worked through once. That fragility, more than any single line-item cost, is the strongest argument for reconsidering an on-camera-only workflow.

Where AI video and avatars earn their keep

For the clearest case for AI avatars, start with the content that goes stale fastest. Standard operating procedures, policy walkthroughs, and software UI updates are all good candidates, especially when accuracy has a short half-life and a missed detail matters. This is content that gets produced, becomes outdated, and then sits in a library looking wrong until someone finds time to redo it.

“This is what we call high turnover, and it may focus on a process, procedure, policy, or something that’s changing regularly,” Pierce said. “Organizations will need to determine individually what their threshold of frequency is, whether it’s three months, six months, or a year.” 

That cycle is where avatars and AI-generated voices earn their keep, not because they’re better presenters, but because they make updates much easier. You change the script, regenerate the narration, and publish a new version. The content stays current, and no one needs to rebook a room or clear a calendar.

It’s worth being precise about this because competitors tend to oversell the point. AI avatars don’t replace all training videos, but they can reduce production friction for the subset of videos where freshness matters more than personal presence.

The product update scenario: Hours versus minutes

The abstract argument for using AI in product update content is that “AI is faster.” The more persuasive argument is the actual difference in production time.

“You know, it might take 10 minutes to generate an avatar or something like that for a longer video, but that’s compared to hours working and preparing,” said Pierce. 

That kind of before-and-after comparison is what makes the case credible to a skeptical training team. Trainers have sat through enough vendor pitches to tune out broad promises about speed. What lands is a specific example they can map onto their own production calendar and use to see the difference.

Where human presence still matters

There’s a version of the AI video conversation that implies the only content worth putting a real person on camera for is sales and marketing, where personality drives the outcome. That undersells how much emotional weight training content can carry.

Think about a video explaining a benefits change during open enrollment or walking employees through a restructuring. These aren’t simple product demos. The information is personal, the stakes feel immediate, and the audience is looking for cues that someone real is behind the message. An avatar may communicate the facts while still making the delivery feel impersonal or disconnected from the stakes.

“I know every year our benefits slightly change and adjust,” Pierce shared as an example. “I don’t want to hear that from an avatar. I don’t want to be told by an avatar that my benefits are being cut and I’ll have to pay more money.” 

The human-connection spectrum: HR, strategy, and sales

The right way to think about this isn’t as a binary between content that “needs a human” and content that “doesn’t.” It’s a spectrum, and where a video falls depends on content sensitivity, not department. 

A routine IT security reminder and a message explaining why access to employee devices must be revoked during a layoff may both come from internal teams, for example, but they sit at completely different points on that spectrum.

“Even with training, I think there are really mixed feelings in the industry right now about avatars,” Pierce explained. ”A lot of people I talk to don’t like them. They don’t want an avatar because it feels like the uncanny valley.”

That reaction doesn’t mean avatars are unsuitable for all training. It does mean that using one is not always a neutral production choice.

Ultimately, it comes down to this: When the audience’s first reaction to the content is likely to be emotional rather than informational, a human presence is doing work that an avatar can’t yet reliably replicate.

What the research says, and what it doesn’t

There’s research on whether a visible instructor affects learning, and there’s newer research on whether AI avatars specifically change outcomes. They don’t answer the same questions, and the gap between them matters.

Preference versus performance: The Stanford findings

Research from Stanford’s Virtual Human Interaction Lab found that learners preferred having a visible on-camera instructor. That preference was clear and consistent, and it may influence a viewer’s motivation to continue watching. Performance itself, however, didn’t change. Test scores and comprehension were essentially the same whether the instructor was visible or not.

That study also predates the current generation of AI avatars, so its findings can’t be assumed to apply directly to synthetic presenters.

Camtasia’s own AI Avatar study tested AI avatars directly in relation to learning outcomes rather than preference alone. In that study, the picture-in-picture avatar format achieved the highest quiz retention of any tested format, at roughly 76%, about 10 points above other formats. Full-screen avatars, however, underperformed as viewers noticed robotic cues such as lip-sync timing and unnatural blinking.

That’s a promising early signal that how you place an avatar may matter as much as whether you use one. But it’s one study, and treating the findings as settled science would repeat the same kind of overreach this section is warning against. For now, the most accurate conclusion is that the evidence is encouraging, format-dependent, and still developing.

The risks nobody talks about

Most of the risk conversation around AI video focuses on whether the output looks “good enough.” The business-continuity risk gets far less attention, even though it can affect long-term planning just as much.

AI avatars introduce a dependency that doesn’t exist in quite the same way with a human presenter: The content is tied to a vendor’s platform, and platforms change. Switching tools, adopting a newer avatar model, or finding that a preferred avatar has been retired can mean replacing that avatar and re-rendering part of a content library. 

That usually isn’t catastrophic. In many cases, it’s a planned swap-and-regenerate process. It is still a maintenance cost that teams adopting AI video for the first time may not account for.

A full vendor shutdown is also possible, but far less common than routine model changes. That wouldn’t likely be the main concern here, though it is worth accounting for. 

Employee turnover versus vendor dependency

Human talent carries its own version of this risk, and it can be less predictable. When an employee leaves or is terminated, you may need to pull their videos entirely, renegotiate usage rights, or rerecord with someone new. This can force content offline with little warning.

Ultimately, neither option is risk-free. Any team that treats AI avatars as a way to eliminate production risk is simply trading one set of vulnerabilities for another. Acknowledging that upfront is what separates a real production strategy from a vendor pitch.

Production choices that reduce the downside

The most practical response to AI video’s current limitations is to make production choices that account for where the technology is right now while still leaving room for where it’s headed.

Picture-in-picture placement is the safer default. Full-screen avatars invite scrutiny that the technology doesn’t always withstand. 

For example, small stiffness in the avatar’s expression or movement that a viewer might not notice in a corner overlay can become distracting at full scale. Pairing screen recordings or slides with a smaller avatar window keeps the focus on the material while still giving learners a visual anchor.

Voice variety matters, too. Reusing a single AI voice across an entire training library can start to flatten the experience for learners. 

“There’s novelty the first couple of times. We have one voice that sounds like Burt Reynolds and one that sounds like a football announcer,” Pierce said. Those are novel, but if you keep using the same voices over and over, the novelty fades.” 

Rotating voices, even within the same series, introduces enough variation to keep the content from feeling monotonous. It’s a small production decision that can make a noticeable difference across a library of dozens or hundreds of videos.

Choosing deliberately, not defaulting to either extreme

You don’t need to pick a side in the AI avatar debate. The goal is to match the production method to the content.

Camtasia’s survey data supports that approach. Viewers were most comfortable with AI avatars in screen-based instructional content, where consistency matters more than personal presence. They were least comfortable with leadership messages, culture-setting communications, and sensitive topics requiring emotional nuance and trust.

High-turnover, procedural content is where avatars can reduce friction. Sensitive, people-focused content is where a real person is more likely to strengthen trust. Teams that make those choices deliberately, asset by asset, will produce better training videos.

Camtasia brings AI-generated narration, avatars, screen recording, and editing tools into one toolkit. See all products.

Frequently asked questions

Does research actually prove AI avatars improve learning outcomes?

Partially. Earlier research on on-camera instructors found that a visible presenter improved learner preference but didn’t change test performance or comprehension. However, that study predates AI avatars entirely. Camtasia’s own AI Avatar study has since tested this more directly: the picture-in-picture avatar format achieved the highest quiz retention of any format tested, at roughly 76%, about 10 points above other formats, while full-screen avatars underperformed. The evidence isn’t absent, but it is format-dependent and still developing compared to decades of research on human presenters.

How do you decide which training content works for AI avatars versus real people?

Look at how often the content changes. High-turnover material like software walkthroughs, procedures, and policy updates is a strong fit for AI voices or avatars because you can regenerate it in minutes. Sensitive topics like benefits changes, layoffs, or strategic messaging still need a real person delivering the message directly.

What does AI video actually save beyond the cost of hiring on-camera talent?

The bigger savings show up in time and logistics, not just talent fees. Traditional shoots require scheduling talent, completing multiple takes, and reading from a teleprompter for hours or across multiple days. When something changes after filming, AI-generated narration lets you revise the script and regenerate the affected content in minutes rather than rebooking a shoot.

Are AI avatars actually safer than human on-camera talent from a risk standpoint?

Not entirely. They shift the risk elsewhere. Human talent carries turnover risk: an employee leaves or is terminated, and their videos can become a liability or need rerecording. Avatars carry vendor risk instead. Switching AI tools or adopting newer avatar models can mean an existing avatar gets retired and needs replacing. A vendor going fully out of business is possible, but far less common than routine tool or model changes.

Should AI avatars appear full-screen or in a smaller window during training videos?

Picture-in-picture placement is the safer choice right now. Full-screen avatars draw more attention to any stiffness or uncanny valley effect in the technology’s current state. Pairing that placement with some variety in AI voices also helps, since reusing the same voice across an entire training library can start to feel flat to learners.