AI-Economy

Google launches AI avatars for customer service in 97 languages

3 min read

TL;DR Too Long; Didn’t read

Starting September 24, 2026, corporate customers in Google Gemini Enterprise can book speaking video avatars for customer service. The new Live Avatar feature supports 97 languages and responds with lip-synced facial expressions. Car dealer Cox Automotive and provider Equal AI already use it productively, according to Google. It remains unclear how and at what price German companies can access it.

A screen shows a smiling AI video avatar with moving lips, surrounded by speech bubbles in different writing systems, with a Google logo sticker on the screen bezel Image generated with GPT Image 2

Key takeaways

  • Live Avatar goes live on September 24, 2026, for corporate customers in Gemini Enterprise.
  • The video avatars automatically speak and understand 97 languages with synchronized facial expressions.
  • Cox Automotive allows customers to compare vehicles and clarify financing via avatar.
  • Equal AI processes over a million conversations daily in nine Indian languages, according to its own statements.
  • Individual avatars from personal photos remain reserved for strictly vetted approval lists.
  • Each generated image and sound carries an invisible SynthID watermark for humans.

Google is launching its video avatar feature Live Avatar for Gemini Enterprise on September 24, 2026. Companies can use it to deploy talking AI figures with synchronized facial expressions in customer service. The avatars master 97 languages and automatically recognize spoken language.

Technology connects voice dialogue with lip-synchronized video

The basis is the native language model Gemini 3.8 Live, which Google introduces in the Cloud Blog and is now expanding with video avatars. The system generates a video face in almost real-time, whose lip movements and facial expressions match the spoken text. In the background, the AI can continue to work via tool-calling – the automatic retrieval of databases or triggering of actions – without interrupting the conversation. Additionally, the model processes camera images and shared screen content in parallel with the audio track. A customer can show something during the conversation, and the avatar responds directly to it. If the user interrupts the avatar or is interrupted themselves, the system simply continues the conversation context instead of restarting. For speech recognition, 97 languages are available, which the avatar switches between without manual toggling. An extended variant with additional thinking time for more complex tasks is still in a closed preview. With this, Google aims at applications where previously only humans responded on the phone or in chat.

Retail and support test the avatars first

First companies are already using the feature productively, as Google describes in its announcement. At the car dealer Cox Automotive, an avatar takes over the vehicle search via the brand Autotrader. It compares models, highlights details directly on the screen, and guides through financing options. Customers describe in their own words what they are looking for, instead of clicking on filters. For them, the conversation is supposed to feel less like text input and more like a call with a real person. The provider Equal AI claims to handle more than a million conversations daily in nine Indian languages using the technology. This number is independently unverified. Salesforce is also integrating the avatars into its Agentforce platform, so customer service agents have a visible counterpart instead of a pure text chat. Google is specifically positioning the feature for high-volume conversation scenarios such as retail, insurance, or reception areas, for example, for automated check-in at hotels. Google names these cases as the first public reference customers for the new feature.

Access remains initially limited to business customers

Live Avatar runs exclusively through Gemini Enterprise and is therefore aimed at corporate clients, not private users of the Gemini app. A library of pre-made avatars is available for getting started. However, an individual avatar from one’s own reference images can only be created through a strict, Google-verified approval list. Google does not specify concrete rates and refers interested parties to the pricing page of Google Cloud. The feature is available through endpoints in the USA and the EU. Whether and how customers in Germany can specifically access it remains open. The offering also includes guaranteed computing capacity, contractual compliance commitments, and strict control over where data is processed. Compared to the previous version, Google also promises noticeably shorter response times and more reliable detection of conversation pauses. Each generated image and sound also carries an invisible SynthID watermark for humans. For images and videos for private users, only the visible marking can currently be turned off – the invisible watermark remains intact there as it does here.

It will be crucial whether Google also marks the talking avatars as AI in conversations with end customers. So far, the invisible watermark serves as the only proof. The initiative also falls into a phase where Gemini is already used by more than a billion people monthly. A visible face for corporate clients could directly translate this reach into paying business accounts.

Frequently asked questions

Is Live Avatar also available in Germany?

Google mentions endpoints in the USA and the EU but does not specifically address Germany. Access is exclusively through business accounts in Gemini Enterprise.

What does it cost to use Live Avatar?

Google does not provide fixed prices and refers to the general pricing page of Google Cloud. Costs are likely based on usage volume.

Can companies create their own individual avatar?

Yes, but only through a strict approval list with verification by Google. A library of pre-made avatars is available by default.

Is this different from the Gemini app for private individuals?

Yes, Live Avatar is part of the separate corporate product Gemini Enterprise and is not aimed at private users of the free Gemini app.

Do conversation partners recognize that they are speaking with an AI?

An invisible SynthID watermark is embedded in image and sound. Google does not mention any perceivable labeling in the conversation itself.

Sources (3)
  1. Google Cloud Blog: Gemini 3.8 Live with Live Avatar is now generally available
  2. Google Blog: Introducing Gemini 3.8 Live with Live Avatar
  3. TechCrunch: Google ramps up its 'AI in the workplace' ambitions with Gemini Enterprise

Your AI update for the work week

Once a week, the most important AI news – plus one practical tip to try right away. No spam, unsubscribe anytime.

← Back to the blog