AI avatars work well when your viewer already trusts you and just needs information: internal training, help documentation, localization and frequent product updates. Real presenters win whenever a video has to earn trust from someone who doesn't know you yet, which covers your homepage, sales videos and brand. The 2026 data supports that split. 52% of Americans are more concerned than excited about AI (Pew Research Center, 2026), and controlled studies find avatar videos teach about as well as human ones while people still rate the human versions as the better experience.

I should say up front where I stand. At Videodeck we only film real people: we have 10+ in house spokespeople and our own studios, and we don't use AI to generate the person on screen. So yes, I have a horse in this race.

But I get asked about Synthesia and HeyGen on almost every discovery call, and "never use avatars" would be bad advice. This is my attempt to draw the line fairly, using published data instead of vibes.

The short answer

• AI avatars are good at transferring information and weak at transferring trust. In a 447 person experiment published in Computers & Education (Netland et al., 2025), AI generated and human made teaching videos produced similar learning, but human videos scored slightly higher on learning experience.
• Audiences want to know when AI made what they're watching. 76% of Americans say it's extremely or very important to be able to tell AI content from human content, and 53% aren't confident they can (Pew Research Center, 2025).
• Labeling something as AI made tends to lower how people rate it. In experiments across the US, UK and Germany, the Nuremberg Institute for Market Decisions found identical ads were judged less natural and less useful when labeled AI generated.
• Disclosure is becoming a legal requirement, not a courtesy. Article 50 of the EU AI Act applies from 2 August 2026 and requires deployers to disclose deepfake video, and YouTube and Meta both require labels on realistic synthetic video.
• Avatar tools are cheap per video, real presenters are cheap per unit of trust. Avatar plans start around $29 a month, while a professionally produced B2B video with a real presenter typically costs $2,000 to $5,000.
• Use the Trust Test: if a stranger is watching, is about to decide whether to believe you, and the video replaces a real person, use a real person.

What the trust data says in 2026

Public concern about AI has settled at a high level, and it is growing fastest among the people most comfortable with technology. Pew Research Center has asked Americans the same question since 2021, and the share who feel more concerned than excited about AI in daily life went from 37% in 2021 to 52% in its June 2026 survey.

The group moving fastest should worry marketers. In Pew's 2026 data, 55% of adults under 30 are now more concerned than excited, up from 39% in 2024. These are the people who will be your buyers and champions for the next twenty years, and they are becoming more skeptical, not less.

The same pattern shows up globally. The KPMG and University of Melbourne global study surveyed over 48,000 people in 47 countries and found only 46% are willing to trust AI systems, while 87% want stronger laws against AI generated misinformation. The 2025 Edelman Trust Barometer found just 44% of people globally feel comfortable with businesses using AI, and trust in AI in the US sits at 32%.

People also say authenticity is the thing that earns their trust. In Getty Images' "Building Trust in the Age of AI" research, 98% of consumers agreed that authentic images and videos are pivotal in establishing trust, and almost 90% wanted to know whether an image was created with AI. Adobe's 2024 survey of 2,002 US consumers found 87% say generative AI has made it harder to tell fact from fiction online.

None of this means people hate every avatar video. It means your audience starts from suspicion, and a synthetic face asks for trust you haven't earned yet.

Do AI avatar videos actually work? What the studies show

For teaching and explaining, the honest answer is yes: the best research finds avatar videos deliver roughly the same learning as human ones. That's a real result and I don't want to wave it away.

• Same learning, slightly worse experience. Netland and colleagues at ETH Zurich showed 447 US adults a mix of human made and AI generated teaching videos. Both groups achieved a similar degree of acquired knowledge, but the human videos had a small, statistically significant advantage in learning experience.
• Same grades, lower engagement. A 108 student study in the International Review of Research in Open and Distributed Learning (2024) found instructor type had no significant effect on academic performance, but video engagement was higher with the human instructor. Students described the synthetic instructor as causing "distraction, discomfort, and disconnectedness."
• Comfort depends on the job. TechSmith's 2026 research with full time workers found an avatar in a small picture in picture window produced the best quiz scores, about 76% correct. Viewers were most comfortable with avatars in instructional, screen based content and least comfortable where a personal presence is expected, like a CEO welcome video.

Advertising is a different story. When NielsenIQ tested AI generated ads with over 2,000 people in 2024 (about 150 of them wearing EEG caps), consumers spotted most of them, called them "annoying," "boring" and "confusing," and even high quality AI ads produced weaker memory activation than traditional ads. NIQ called the spillover a "negative brand halo."

A word on vendor studies

Be careful with performance numbers published by companies that sell avatar tools. TechSmith sells AI avatar features, so its findings are useful but not neutral. Vidyard, which also sells AI avatars, reports that among revenue teams that fully implemented AI video tools, "every one of those teams reported higher win rates."

That sentence tells you very little. Teams that fully roll out any new sales tool tend to be well run already, and self reported win rates are not a controlled test. When you read a case study promising a 400% lift from avatar outreach, ask three questions: what was the comparison, who ran the test, and who paid for it.

What about the uncanny valley?

The uncanny valley is the uneasy feeling people get when something looks almost, but not quite, human. For still images, AI has already crossed it: a 2022 study in PNAS by Nightingale and Farid found AI synthesized faces were indistinguishable from real ones and were rated more trustworthy.

Video is harder, because people read movement, timing and emotion. When Animoto surveyed US consumers in September 2025, the top giveaways people cited for AI video were robotic gestures (67%), unnatural voices (55%) and lack of emotional tone (51%), and 36% said an AI generated video would lower their perception of the brand. It's a small vendor survey, so treat it as a signal, but it matches what we hear from buyers.

Here's the uncomfortable implication. If avatars become truly indistinguishable, the trust problem gets worse, not better, because audiences start doubting every face they see. Hany Farid, coauthor of the PNAS study, makes this case well in a 2026 Berkeley talk.

AI video disclosure: what the rules require in 2026

If you publish realistic avatar video, assume you will have to label it somewhere. The rules differ by platform and region, but they point the same direction.

• EU AI Act, Article 50. The text says deployers of an AI system that generates video "constituting a deep fake, shall disclose that the content has been artificially generated or manipulated" (official AI Act Service Desk). The European Commission confirms Article 50 applies from 2 August 2026 and defines deepfakes as AI content that "resembles existing persons" and would falsely appear authentic. A digital twin of your real CEO fits that description closely.
• YouTube. Creators must disclose content that makes a real person appear to say or do something they didn't, or that shows a realistic scene that didn't happen (YouTube Help). AI scripts, thumbnails and cloning your own voice for dubs don't require disclosure. Creators who repeatedly skip it risk content removal or suspension from the Partner Program.
• Meta. Meta requires people to use its label tool when they post photorealistic video or realistic sounding audio that was digitally created or altered, and says it may apply penalties if they don't (Meta, 2024).
• FTC (United States). The FTC's 2024 rule bans testimonials that misrepresent that they are by someone who does not exist, and allows civil penalties for knowing violations. An avatar playing a "happy customer" is exactly the kind of thing this rule was written for.

I'm not a lawyer, and if you operate in the EU you should get real legal advice. The practical point is that disclosure isn't free. The NIM experiments found that simply labeling an ad as AI made it seem less natural and less useful, even though the content was identical. So the honest version of an avatar video carries a trust cost, and the dishonest version carries a legal one.

Where AI avatars are genuinely great

AI avatars shine when the viewer has already decided to trust you and just needs the information quickly. In that situation, speed and consistency matter more than presence.

• Internal training and compliance. Your employees already work for you. A clear avatar explaining the new expense policy beats a 40 slide PDF nobody opens, and the learning research above says they'll retain it about as well.
• Help center and documentation videos. Customers watching a "how to export your data" clip are already paying you. They care about accuracy and freshness, and avatars let you update the video in minutes when the interface changes.
• Localization. Turning one English training library into 10 languages is where avatar tools save serious money. HeyGen's Creator plan, for example, advertises support for 175+ languages and dialects.
• High volume product updates. If you ship weekly and want a short video for every release note, filming a human for each one is hard to justify.

Where real people win

Real presenters win whenever the viewer is deciding whether to believe you, because a real face is evidence that a real company stands behind the claim. That covers most external marketing.

• Your homepage and main explainer. This is often the first time a buyer meets your brand. Opening with a synthetic face, which may need a label, asks a stranger to trust you before you've shown them anything real.
• Sales videos and demos. B2B buyers are evaluating whether your team is competent and honest. A real person demoing a real product answers that question. A synthetic one raises it.
• Founder and leadership messages. TechSmith's viewers were least comfortable with avatars exactly here, and an AI twin of a real executive is the case most likely to count as a deepfake under the EU definition.
• Customer stories. Only real customers can give real testimonials. Faking one with an avatar is both a trust problem and, under the FTC rule, a legal problem.
• YouTube and thought leadership. Channels grow because people get to know a person over many videos. We see this constantly in the channels we produce: the presenter becomes the reason people come back. It also matters for how AI search tools cite your videos, since they reward consistent, credible sources.

If you've never hired a production team before, here's what working with a video production agency actually looks like, step by step.

The Trust Test: a three question framework

The Trust Test is a quick way to decide whether a video can use an AI avatar or needs a real presenter. It asks three questions about the video's job, not about the technology.

1. Who is watching? Someone who already trusts you (an employee, a paying customer) or a stranger (a prospect, a candidate, an investor)?
2. What do they decide after watching? "How do I do this task?" is an information decision. "Should I believe this company?" is a trust decision.
3. What does the video replace? If it replaces a document, a slide deck or nothing at all, an avatar is an upgrade. If it replaces a real person talking to the viewer, an avatar is a downgrade.

If the answers are "someone who trusts us," "an information decision" and "a document," an avatar is a sensible choice. If any answer is "a stranger," "a trust decision" or "a real person," use a real person. Here's how that plays out across common B2B formats (for a broader look at which formats to make in the first place, see our guide to the best B2B video formats).

The cost math, honestly

On cost per video, AI avatars win by a mile, and there's no point pretending otherwise. As of September 2026, Synthesia lists a Starter plan at $29 a month with 10 minutes of video and a Creator plan at $89 a month with 30 minutes. HeyGen lists Creator at $29 a month and Business at $149 a month plus $20 per seat. Colossyan lists its Professional plan at $59 a month billed annually. Both Synthesia and Colossyan sell professionally filmed studio avatars as an add on of about $1,000 a year.

Compare that with a real presenter. Our own pricing runs from $2,000 to $5,000 per video depending on volume, which includes scripting, a custom set, a spokesperson, editing, motion graphics and social cuts. That's typical for professional B2B production with a real person on camera.

The better question is cost per outcome. A $30 avatar video that makes a prospect trust you a little less is expensive, and a $5,000 homepage video that runs for two years and lifts conversion is cheap. If you're working out how to split a budget between the two, our guide on how much to spend on video marketing walks through it, and it helps to decide up front which video KPIs will tell you whether it worked.

Marketers are clearly adding AI to the mix either way. Wyzowl's 2026 survey found 63% of video marketers have used AI video tools to help create or edit marketing videos, up from 51% the year before. That covers any AI assistance, not just avatars.

The smartest setup we see is a split library: real people for everything external and trust building, avatars or screen recordings for internal and support content. Whether you run that with your own team or a partner is a separate decision, and we've compared an in house video team vs an agency in detail.

What it is like to work with Videodeck

If your video sits on the real person side of the Trust Test, this is how we make it. Every presenter is a real human, filmed in a real studio.

How a project runs

1. A 30 minute call. We learn what you sell, who buys it and where the videos will live. You can book one here.
2. A content plan and scripts. We write them, you approve them. Nothing gets filmed until the words are right.
3. Your presenter and your set. You pick from more than 10 in house spokespeople, and we build a set in your brand colors in our own studios.
4. Filming and editing. The first version lands in 5 to 10 business days, with motion graphics, product screens, captions and thumbnails.
5. Revisions until it is right. Revisions are unlimited and each round takes about two days. You own every final file.

Why software brands choose us

• Real people on camera. We never use AI avatars. Buyers can tell, and trust is the whole point of a video.
• Prices you can see before the call. Our plans are published: $2,500 a video at 4 a month, $2,300 at 8 and $2,000 at 12, or $5,000 for a single video.
• Speed without chaos. Owning the studio and keeping the team in house is what makes 5 to 10 business days realistic.
• Consistency. Many of our clients work with us month after month, because one video is a campaign and forty videos is a channel.
• We think past the edit. Titles, thumbnails, YouTube optimization and social cuts come with every video, so it gets found after launch day.

An education video we produced for Hometap, filmed on a warm home office set, explaining a genuinely confusing money decision in plain English.

"They are super good at putting your product in a good light and can deliver the videos super fast."

Malte Scholz, co founder, CEO and CPO, airfocus

You can see more on our work page and read what clients say on our wall of love.

FAQ

Are AI avatar videos effective?

Yes, for teaching and explaining. A 2025 experiment with 447 adults in Computers & Education found AI generated and human made teaching videos produced similar learning, though viewers rated the human videos as a slightly better experience. For ads and sales, the evidence is weaker: NielsenIQ found AI generated ads produced weaker memory activation than traditional ads.

Do people trust AI generated video?

Less than they trust real people, on average. Only 46% of people globally are willing to trust AI systems (KPMG and University of Melbourne, 2025), and 52% of Americans are more concerned than excited about AI (Pew, 2026). Trust is higher when viewers already know the brand and the content is purely informational.

Do I have to disclose that a video uses an AI avatar?

Often, yes. YouTube requires disclosure for realistic synthetic content, Meta requires labels on photorealistic video that was digitally created, and EU AI Act Article 50 requires deepfake disclosure from 2 August 2026. Even where it isn't required, disclosing is the safer choice for brand trust.

Can people tell when a video uses an AI avatar?

Many people think they can, but confidence is low overall. 53% of Americans are not confident they can tell AI content from human content (Pew, 2025). Viewers who do spot AI video most often point to robotic gestures, unnatural voices and flat emotion.

Synthesia vs a real actor: which is cheaper?

Synthesia is far cheaper per video, with plans from $29 a month, while professional production with a real presenter usually costs $2,000 to $5,000 per video. The real actor is the better value for videos that need to build trust, like homepage explainers and sales videos, because those assets run for a long time and drive revenue.

Should our CEO use an AI digital twin?

For internal updates to employees, it can be reasonable if clearly labeled. For customers, investors or candidates, we'd advise against it: viewers are least comfortable with avatars where a personal presence is expected, and an AI copy of a real person closely matches the EU AI Act's definition of a deepfake, which must be disclosed.

Will AI avatars replace real presenters?

They'll replace a lot of internal and support video that used to be slides or text. They're unlikely to replace real people in trust building content, because as synthetic video gets more convincing, a verifiably real person becomes more valuable, not less.

The bottom line

Use AI avatars where your viewer already trusts you and needs information fast, and put real people wherever you're asking a stranger to believe you. Run each video through the Trust Test and the answer is usually obvious.

If your homepage, sales or YouTube videos fall on the "real person" side, that's what we do at Videodeck: scripted, filmed and edited with real spokespeople in our own studios, in 5 to 10 business days. Take a look at our work or see our pricing, and if avatars are the right call for part of your library, I'll happily tell you that too.

More from Heleana

• What Working With a Video Production Agency Actually Looks Like, From Brief to Final File
• How Much Should a B2B Company Spend on Video in 2026?
• The B2B Video Formats That Actually Work in 2026
• In House Video Team vs Agency vs Freelancers: What It Really Costs in 2026