AI-generated video makes it possible to create characters, adapt languages and produce scenes without a conventional shoot. That agility creates real opportunities, but it also brings flaws, rights questions and one particularly sensitive risk: a synthetic representation may be mistaken for authentic evidence or testimony.
Useful does not mean infallible
The question is not whether AI can generate convincing video, but whether that video meets its objective without confusing, excluding or eroding the audience’s trust.
Why AI video is gaining ground in organisations
Generating a scene with an avatar, adapting a script into several languages or producing variants without repeating a shoot can reduce time and production barriers. In an interactive experience, video can introduce a mission, portray a customer, explain a decision or deliver feedback. Its value is not more footage; it is faster contextual content when audiovisual storytelling genuinely adds something that text cannot.
Where it can add value
Fictional customers, patients or managers present situations participants must interpret and solve.
An avatar accompanies the experience, links stages and explains rules with a consistent presence.
One message can be adapted across languages, with translation, tone and pronunciation all reviewed.
Teams can validate a script or activity before investing in more expensive audiovisual production.
Variants can be created by role, sector, product or level without rebuilding everything.
A fact or scene can be replaced more quickly, provided visual and vocal consistency are maintained.
What it still does poorly
Quality has improved, but remains uneven. Common weaknesses include gestures, hands, lip sync, continuity between shots, proper-name pronunciation and complex emotion. A result that looks convincing on a phone may feel artificial on a large event screen. Test it on the real device, in the real language and in the actual viewing context.
A character’s features, clothing or lighting may change between scenes.
A technically correct voice can sound flat, overemphatic or unsuitable for a sensitive message.
A system may invent visual details or unsupported claims when the script is not tightly controlled.
Captions, contrast, pacing, audio description and text alternatives need dedicated review.
A tool licence does not replace permission to use faces, voices, brands, music or third-party material.
Changing provider or model can alter the style and complicate future updates.
The central risk: mistaking plausibility for truth
Synthetic video can look like documentary footage. If it depicts a real person saying something they never said, recreates a patient without disclosure or presents a simulation as a testimonial, viewers may reach false conclusions. Transparency does not weaken the experience: it protects trust and helps audiences interpret the piece correctly.
- Clearly disclose when a character, voice or scene has been generated or materially altered with AI.
- Do not impersonate a real person or clone their voice without explicit permission and a tightly defined use.
- Avoid fictional testimonials that could be mistaken for real customer, employee or patient experiences.
- In health, finance, compliance or safety, require expert review of both script and final output.
- Retain the approved script, sources, permissions, tool version and generation date.
- Do not enter personal or confidential data into a platform before reviewing its safeguards and terms.
A reliable production process
1. Define the video’s job
Decide what viewers should understand or do afterwards. If a paragraph or illustration works better, do not use video by default.
2. Lock a verifiable script
Separate facts, fiction and examples. Attribute sources, remove ambiguity and prepare technical pronunciations.
3. Choose the right realism level
A clearly illustrated style can be more honest and effective than imitating real footage.
4. Secure consent and protect data
Document permissions for face and voice, limit secondary use and avoid uploading unnecessary sensitive information.
5. Generate and review in context
Check image, voice, captions, continuity, accuracy and cultural fit in the final delivery channel.
6. Tell the audience
Use a visible, understandable disclosure rather than hiding it in terms or metadata.
7. Measure and correct
Assess comprehension, perceived credibility and possible misinterpretations—not just views or completion.
A traffic-light decision guide
Fictional narrators, clearly synthetic characters, internal prototypes and scenes that do not purport to show real events.
Technical training, healthcare, recruitment, corporate communications or content featuring recognisable people and brands.
Impersonation, fabricated testimonials, statements attributed to executives, misinformation or content that could harm someone.
Pre-publication checklist
- Does video offer a concrete advantage over text, audio or illustration?
- Have all claims and figures been verified?
- Do we have permission for every face, voice, brand and asset?
- Will viewers understand what was generated or altered with AI?
- Does the piece work with captions and meet accessibility requirements?
- Has it been reviewed on the real playback device and screen size?
- Is someone accountable for approving, updating or withdrawing it?
AI with judgement
Credibility is designed before anyone clicks “generate”.
With Evenely, you can integrate AI-generated video and avatars into interactive experiences, or ask our team to define the script, style, production process and safeguards appropriate to your use case.