On 24 September 2026 Google announced Gemini 3.8 Live with Live Avatar: a real-time talking face, lip-synced and expressive, sitting on top of Gemini's live voice model. It speaks 97 languages, is available now in Gemini Enterprise for customer service and interactive walkthroughs, and can run tool calls in the background while the avatar keeps talking. Google's demo shows it checking a guest into a hotel.
Claude does not ship anything like this, and we are not going to invent a comparison. The more useful question for an Australian business is simpler. Would a face make this particular interaction better, or would it be a costly layer on top of a task that a text or voice agent already handles?
Does a talking AI avatar improve customer service?
A talking AI avatar improves customer service only when the customer is standing in front of a screen and benefits from a visible, guiding presence, such as a hotel check-in kiosk, a retail floor display or a multilingual front desk. For phone support, chat, email triage and back-office work, an avatar adds cost and testing effort without making the task itself any more likely to be completed correctly.
The reason is that customers judge a service interaction by its outcome. Was the booking changed, the refund issued, the question answered? A face can make the wait feel friendlier, but it cannot fix an agent that misreads the booking system or escalates too late. Task completion sits in the reasoning and integration layer underneath.
Here is how we would sort common use cases:
| Interaction | Customer sees a screen? | Does a face help? |
|---|---|---|
| Hotel or venue check-in kiosk | Yes | Often, for guidance and reassurance |
| Retail floor or showroom display | Yes | Sometimes, if it draws people in |
| Multilingual front-of-house desk | Yes | Yes, where staff cannot cover every language |
| Phone support line | No | No, voice alone is the interface |
| Web chat and email triage | Partly | Rarely, text is faster to scan |
| Internal tools and back-office workflows | No | No |
What Gemini Live Avatar actually does well
It is worth being fair about the launch. Three things stand out:
Real-time lip sync and expression on top of a live voice model, so the face responds as the conversation moves rather than playing pre-recorded clips.
Coverage of 97 languages, which matters for Sydney, Melbourne and Brisbane businesses serving tourists, students and diverse local communities.
Background tool calls during the conversation, so the avatar can look up a reservation or update a record without an awkward pause.
For a business that already runs on Google Workspace and has a genuine kiosk or front-desk use case, that is a real option worth piloting on its own merits.
Where Claude fits instead
Claude's strength is the work behind the interface: reading documents, calling internal systems through MCP connectors, following business rules and handing off to a person at the right moment. Most of the automation we build for Australian mid-market clients never needs a face at all. It runs in a ticket queue, an inbox, a spreadsheet or a phone line.
If you do need an embodied presence, the two are not mutually exclusive. You can keep Claude doing the reasoning and task work and put a dedicated avatar or video layer in front of it, the same way teams already pair Claude with voice platforms in phone agents built on Vapi and ElevenLabs. The interface becomes a choice you make per channel, not a reason to switch your whole stack.
A worked example: the cost of adding a face
Take an illustrative Brisbane hotel group with 6 properties weighing a check-in kiosk. These figures are illustrative only, not quotes:
Kiosk hardware, at roughly $4,500 per site: $27,000.
Integration with the property management system and payment terminal: $35,000 to $60,000, whether or not there is a face.
Extra testing for the avatar across languages, accents and noisy lobbies: $10,000 to $15,000.
Ongoing review of avatar transcripts for brand and accuracy: a few hours a week.
The integration line is the expensive part, and it is the same with or without the avatar. If the check-in agent cannot reliably find the booking, apply the right rate and issue the key card, the face just makes the failure more visible. Get the task working on a plain screen first, measure completion, then decide if a face lifts guest satisfaction enough to justify the extra spend. Our ROI calculator is a reasonable place to put rough numbers against that decision.
What not to conclude
This is not an argument that avatars are a gimmick. In the right setting they help, especially for guests who would rather follow a friendly guide than read a wall of text in a second language. Nor does it mean Google is ahead on everything because it shipped a feature first. It means the interface and the agent are separate decisions, and most budgets should go to the agent.
There are also Australian compliance points to settle before any customer-facing avatar goes live. A face that speaks for your brand needs a clear disclosure that the customer is talking to AI. Transcripts and any captured voice or video fall under the Privacy Act, so decide what is stored, where and for how long. In regulated sectors, ASIC and APRA expectations on customer communications apply to what the avatar says, just as they do to a human script.
How we would decide
Start with the task, not the demo. Write down the five most common requests the interaction must complete, test them with a text or voice agent, and track how many finish without a person stepping in. If completion is high and the setting is physical, pilot a face and compare satisfaction. For more on why completion matters more than polish, see our pieces on dictation versus task completion and Gemini 3.8 Live's voice results.
If you are weighing a customer-facing AI channel and want a second opinion on where the effort should go, book a time with us. We will look at your actual interactions and tell you whether a face, a voice or neither is the right first step. You can also see how we approach builds on our services page.
