AI avatar tools such as HeyGen and Synthesia turn a written script into a video of a digital presenter speaking it, in many languages and voices. In client work they are best for explainer videos, product walk-throughs, onboarding, internal training and localised versions of existing videos. They are not a substitute for real people in testimonials, emotional brand stories or anything that could mislead viewers about who is speaking.
What do HeyGen and Synthesia actually offer?
Both are commercial platforms built around the same idea: pick an avatar, paste a script, choose a voice and language, and render a video. Each offers a library of stock avatars based on consenting actors, and each lets customers create custom avatars of real people who record consent and training footage.
Features at the time of writing include translation and dubbing of existing videos with lip-sync, voice cloning with consent, screen recording and slide imports, templates, brand kits and team workspaces. Synthesia has focused heavily on corporate training and enterprise use, while HeyGen is widely used for marketing and social content. Both update their features often, so check current plans and limits before committing.
Which campaign formats suit AI avatars?
The strongest formats are informational, script-led and likely to need frequent updates or many language versions.
- Product explainers and feature updates that change every release.
- Localised versions of an existing video for several markets.
- Onboarding and how-to videos inside a SaaS product or help centre.
- Internal training and compliance modules.
- Personalised sales outreach videos at scale, where the platform supports variables.
- Event recaps or announcements when the speaker cannot film in time.
In each case, the viewer cares more about clear information than about a human connection with the presenter.
Where do avatars not belong?
We do not use avatars for customer testimonials, reviews or anything presented as a real person’s experience. A synthetic person giving a testimonial is deceptive, and in many countries it breaches advertising and consumer protection rules.
Emotional brand films, founder stories and crisis statements also need real people. Audiences notice the difference, and in those moments trust matters more than production speed. Paid social ads with avatars can work, but they often underperform authentic creator content in feeds like TikTok and Instagram Reels.
How does the process work in a typical engagement?
The process has five stages, each with a sign-off.
Discovery
We review existing video assets, target markets and channels, and agree which formats are a good fit for avatars and which are not. Legal requirements for each market are noted, including disclosure rules.
Avatar and voice setup
Clients choose a stock avatar or create a custom one. For a custom avatar of an employee or spokesperson, the person records the platform’s consent statement and training footage, and signs a written agreement covering where, how long and for what the likeness may be used.
Scriptwriting
Scripts are written for spoken delivery: short sentences, one idea per scene, no complex numbers read aloud. Product claims are checked against approved sources.
Production and editing
Videos are rendered in the platform, then edited with brand graphics, b-roll, screen captures and captions. Translated versions are reviewed by native speakers for pronunciation, terms and on-screen text.
Review and publishing
The client approves each video. AI involvement is disclosed where platform policies or law require, and files are tagged in the asset library with script, avatar and consent records.
What quality rules do we follow?
A short checklist keeps avatar videos from feeling cheap.
- Keep videos short; most explainers work best within a couple of minutes.
- Mix the avatar with screen recordings or visuals so the face is not on screen the whole time.
- Check pronunciation of brand and product names and fix them with phonetic spelling.
- Use captions in every language.
- Avoid hand gestures or framing that highlight synthetic movement.
- Test on mobile, where most social viewing happens.
How are consent and disclosure handled?
Consent is the first rule. No custom avatar or cloned voice is created without the person’s recorded and written agreement, and the agreement includes how to withdraw it. When an employee leaves, the client decides whether their avatar is retired.
Disclosure follows platform rules and law. YouTube, Meta and TikTok require labels for realistic synthetic content in many cases, and the EU AI Act introduces transparency duties for content that could be taken as real. Our default is to mention AI presentation in the description or on screen for any public video.
How do we measure results?
Measurement depends on the format. For explainers and help content, we track completion rate, support tickets on the topic and help article engagement. For localised marketing videos, we compare view-through rate and conversion against the original language version. For training, completion and assessment scores show whether the content works.
We also track production time and cost per finished minute compared with the client’s previous filmed process. Avatar video saves the most time on updates: changing a script and re-rendering takes far less effort than a reshoot.
Which mistakes do we see most often?
The first is putting the avatar on screen for the whole video. A talking head with no visuals is tiring to watch, synthetic or not, and makes small flaws more visible. The second is writing scripts as if for a brochure, with long sentences and lists of features that sound stiff when spoken.
The third is skipping native review of translations. Lip-sync can look fine while a product term is mistranslated or pronounced in a way local viewers find odd. The fourth is forgetting to update consent records when a spokesperson changes role or leaves the company.
Where to start this week
List the videos your team updates most often or would like in more languages. Pick one, ideally a product explainer, and produce it with a stock avatar on a trial plan of HeyGen or Synthesia, with captions and screen visuals. Show it to a few customers or colleagues next to the original and ask what they understood. If the information lands as well, you have found your first avatar use case.
Keep reading: Video Production · AI Marketing