Testimonial video and AI
By Syntax Motion, Vienna. Updated: 16 October 2026
A testimonial works because it is real. That is exactly why it is the format where generative AI goes wrong fastest. This piece sets out what is technically possible with a genuine customer voice, where the legal line runs and which requests we turn down.
Why the question comes up at all
A good testimonial is expensive to obtain, not to produce. Finding someone who is satisfied, has time and is willing to speak on camera takes longer than the shoot. Since tools appeared that generate a speaking person in minutes, the shortcut is obvious. It is just not a shortcut.
The line in one sentence
A real statement may be edited, an invented statement stays invented
When a real customer said something and we improve the image, sound or language of that statement, the substance remains true. When a person is produced who never existed, or words are put into a real person's mouth, it is no longer production, it is a claim about reality.
What is possible with a real customer voice
- Improve a remote recording: your customer records themselves on a phone, we bring sound, light and edit to a usable level. That saves the shoot day, not the statement.
- Replace the surroundings: the person stays real, the cluttered background is swapped for a generated environment. The same as a green screen, only cheaper.
- Translate with lipsync: the same statement in English, in that person's voice and with matching mouth movement. Requires their explicit consent, because voice and face are used.
- Recut and shorten: eight minutes of conversation become forty seconds. Permitted, as long as the cut does not reverse the meaning.
- Add scenes: generated images illustrate what is being described while the statement runs.
What we refuse
- Invented customers. A produced person describing an experience nobody had is a deception about a material characteristic. Looking convincing makes it worse, not better.
- Real people without consent. Not with a notice, not as a draft.
- Statements that were never made. A real person given a better line is the most dangerous case of all, because it lands back on that person.
This is not a matter of taste. Misleading claims about product characteristics are prohibited under unfair competition law, and since 2 August 2026 Article 50 of the EU AI Act additionally requires disclosure for convincing likenesses of real people, see AI labelling in advertising.
UGC is not the same as a testimonial
A UGC video imitates the form of a user clip: handheld, natural light, no studio. That is a design decision and permitted, as long as the ad stays recognisable as an ad. A testimonial additionally claims that this person had this experience. The form may be rebuilt, the claim may not.
The practical route
- Ask three satisfied customers for two minutes. That is all it takes.
- A remote recording on a phone, after a short briefing from us: light from the front, phone at eye level, quiet room.
- Written consent covering purpose and duration. Half a page, not a contract.
- Production: edit, sound, subtitles, environment and supporting scenes from AI.
Frequently asked questions
Can testimonials be made with AI?
A genuine customer statement may be edited, translated or illustrated with AI where the person has consented and the meaning is preserved. An invented person describing an experience that never happened is misleading advertising, however it was produced.
Does an edited testimonial have to be labelled?
Where the result is a convincing likeness of a real person, for instance an AI translation with lipsync, Article 50 of the EU AI Act requires disclosure. Editing, sound and a replaced background do not trigger it.
What about avatars as presenters?
A visibly artificial avatar presenting information is unproblematic as long as it does not pose as a customer. The line is not the avatar, it is the claim.
We only have written reviews, no video. Can anything come of that?
Yes, with the author's consent. A genuine written review can appear as a quote on screen, read by a neutral voice and attributed by name and source. What does not work is a generated person delivering it as though they wrote it.
This is a summary for practitioners, not legal advice.
Read on: All guides • UGC video • AI labelling
Satisfied customers but no video?
We show you how to get usable material with a phone, and turn it into a film.
Book a first call