Google has given its enterprise chatbot a face and a voice that move together in real time. The feature, called Live Avatar, sits inside Gemini Enterprise, Google's paid AI product for businesses, and lets a digital character look at you through a camera, listen, and respond with matching lip movements almost as fast as a real conversation.
It arrives exactly one week after Google launched Gemini 3.8 Live, the voice-only version of the same model. That short gap says something about the pace here: Google is moving quickly to turn a talking AI into something that looks present, not just a voice coming out of a speaker.
The pitch is customer service and guided walkthroughs. In one demo, the avatar checks a guest into a hotel while quietly pulling data from a booking system in the background, without pausing the conversation. That matters because the avatar can multitask like a front desk worker, instead of going silent while it looks something up.
The avatar also switches between 97 languages mid conversation without the lip movements falling out of sync. For a business running a global helpline, a hotel chain, an airline, or an outsourced support desk, that removes the usual delay of transferring a caller to a different language line.
Businesses get a choice of built-in avatar characters, or, with approval, can build a custom one from a single reference photo that keeps a brand's look or a real person's likeness. That turns the AI into something closer to a branded digital employee, which raises its value for marketing teams but also raises the stakes if that face ever says the wrong thing on a recorded call.
Google is not first to put a face on an AI assistant. Microsoft has been testing its own animated avatars inside Copilot Labs, and companies such as Synthesia and HeyGen have sold avatar based video tools to corporate teams for years. Google's version is built for live, two way conversation rather than pre recorded video, and sold to paying enterprise customers rather than consumers.
Every clip the avatar produces carries an invisible marker called SynthID, woven into both the audio and the video so the content can be identified as AI made later. That detail matters. Rules on AI generated media are tightening in several regions, and a tool that can put a real person's likeness onto a talking avatar is exactly the kind of feature that draws regulatory attention.
The bigger question for any business is not whether this technology is impressive, it clearly is, but how much of a front line job it can actually take over. Estimates on the pace vary widely: some analysts expect close to half of customer service work to shift to AI within the next few years, while others put the share of fully automated interactions in the low double digits over the same period. Either way, the pattern is the same. Routine, scripted conversations move to a screen with a synthetic face first, and harder, emotional, or high stakes conversations stay with humans the longest.
For now, this is a tool for companies that already pay for Google's enterprise AI products, not something the public will run into by accident. But it is a preview of what talking to a company will look like soon: a video call with a character that never gets tired, never calls in sick, and comes with a built in watermark so you can prove later that it was not a real person.