On September 24, Google added a video presence to Gemini 3.8 Live in its Enterprise offering. For the hotel industry, the question is what this animated face is entitled to say, do and pass on.
A face that keeps talking while it works
In the announcement, signed by Shuo-yiin Chang and CJ Zheng on behalf of the Gemini audio team, Google describes a voice paired with video generated in near real time. The avatar listens, looks and responds with facial expressions and lip synchronization. It can call tools in the background without interrupting the conversation. Google also announces switching between 97 languages without any degradation of the image.
The only hotel scene in the text is a demonstration: a guest checking into a hotel while tools work in the background. It is a presentation video, not feedback from an establishment using it. The announcement provides no measurements of latency, accuracy or error rates.
The face lends weight to the answer
A face that smiles and looks at you makes an exchange easier to read. It also makes an incorrect answer more credible. Erroneous text on a screen is read with suspicion. The same content spoken by someone who appears attentive is less readily checked. This is the author’s judgment, not Google’s, but it warrants addressing the question of authority before that of appearance.
Let us apply this to hypothetical situations, without describing an actual case. An avatar reads availability and announces it to the guest without being authorized to book it. Another mentions a rate or a change to a stay that it can neither confirm nor honor. In both cases, the wording must say what is being consulted, what is being promised and what still needs to be validated by a person. A guest will not make that distinction unaided when faced with such a fluidly animated face.
What the announced transparency covers
Google states that all audio and video outputs carry the SynthID watermark, which it describes as imperceptible. Its purpose is to allow content to be identified as generated after the fact. It does not tell the guest, at the time, that they are speaking to a synthetic presence. Providing that information is the establishment’s responsibility, on screen or verbally.
Google specifies that creating a custom avatar from a reference image is restricted to allowlisted companies. That raises the question of whose face to use: an employee’s, a real person’s or an invented character’s? The announcement does not address the rights associated with the original image.
What the establishment should test before making it available to guests
The evaluation would cover complete exchanges, including those in which no answer is available. A fluidly animated avatar that delays contact with the person able to make a decision worsens the service. The handover to the team must therefore be tested just as image quality is: when it is triggered, what is passed on and whether the guest has to repeat themselves.
Visual consistency helps an establishment present itself. But trust depends on a consistency the screen does not show: every promise made by the avatar must correspond to verified information and an identifiable person who is accountable for it.
Cette publication est également disponible en :
