Google Put a New Model Behind Voice and Camera Search. Your Dealership's Answer Now Has to Survive Being Spoken Out Loud.

On September 15, Google swapped the model running its voice and camera search.

Google's announcement introduces two of them. Gemini 3.8 Live is "built for scale and cost efficiency, combining conversational intelligence with fluid dialogue and visual grounding." Gemini 3.8 Live Extended Thinking is "built for high-complexity tasks, with increased intelligence and multi-step reasoning." The first one is the one that went into Search.

Rajan Patel, VP of Engineering for Search, posted that it is already live for everyone in Search Live. Barry Schwartz picked the post up at Search Engine Roundtable and quoted the three things Patel says it brings: "more helpful responses, complete with web links to dive deeper," "fluid multilingual support (you can switch languages mid convo)," and "more natural, free-flowing interactions."

The specifications are more useful than the adjectives. Google says these models automatically detect and transition between "97 supported languages mid-conversation," process visual inputs "in near real-time," and execute "tools and API calls in the background while continuing the conversation." On the scoreboard, the Extended Thinking variant took the number one overall spot on Artificial Analysis' Speech to Speech Quality Index at 82.6, hit 68.6% on the tau-Voice agentic task benchmark, and scored 97.7% on Big Bench Audio reasoning. 9to5Google's writeup adds the detail that the high-reasoning version "reasons and speaks simultaneously," filling silence with cues like "Let me check that…" the way a person does.

Every one of those numbers is about a machine holding a conversation well enough that a human does not bail on it halfway through.

What Search Live actually is, and why most dealers have never opened it

Search Live is the microphone and camera button in the Google app. Google describes it as "an interactive, multimodal conversation in AI Mode," where you "ask your question out loud to get a helpful audio response, then continue the conversation with follow-up questions," and you can "enable your camera to add visual context."

The Google help page is blunter. Open the Google app, tap Live below the Search bar, talk. You "use your camera to show Search what you see," you get "AI-generated audio responses," and you get "helpful web links on your screen to keep exploring." It runs only in the Google app on Android and iOS, in every region and language where AI Mode is available. Google also posts the standard warning on the same page: "AI responses may include mistakes."

The history is worth three sentences, because it explains why this landed quietly.

Search Live started as a United States and India feature behind an opt-in. Google removed the opt-in and then took it worldwide on March 26, to "more than 200 countries and territories," running on a model called Gemini 3.1 Flash Live. TechCrunch noted at the time that before that date it existed in exactly two markets. So American dealers have had this surface in their market longer than anyone, and almost nobody in the business has ever tested it, because for most of its life the thing on the other end was not good enough to hold a real conversation.

That is the actual news on September 15. The surface stayed where it was. The quality of the thing answering it jumped.

The part that matters: your answer has to survive being said

For twenty-five years, a dealership's search presence had to survive being read. A shopper scanned a page of results, and being fourth still meant being seen. Then AI Overviews compressed the page, and AI Mode compressed it further, and we have written about both.

A spoken answer compresses it to one.

There is no position two in audio. Google reads a response, and the links sit on the screen underneath as the follow-up option for anybody who wants to dig. The spoken sentence is the product. The link list is the footnote.

Now add the camera, which is where this gets specific to a car business in a way it does not for most industries. From Google Lens, a shopper can "tap the Live option at the bottom of the screen to have a real-time, back-and-forth conversation about what you see." Google says the new model handles that visual input in near real time.

Point that at a physical vehicle and think about where the answer comes from. A shopper standing in your front line at 8pm on a Sunday, gate locked, sales floor dark, can hold up a phone to a unit and start asking questions about it out loud. The window sticker three feet away is the one thing in that exchange Google is not reading. The answer gets assembled out of whatever the web says about that year, make, model and trim, which means it gets assembled out of your VDP, or out of somebody else's.

What this means for your dealership

Anything printed in an image is silent. A model that speaks can only speak text. Your price swoosh graphic, your payment disclaimer baked into a banner, your holiday hours posted as a JPEG, your service specials laid out in Photoshop: all of it is invisible to a voice answer. Dealers have been told for a decade that text beats images for accessibility and for crawlers. This is the same fix with a shorter feedback loop, because now you can hear the omission yourself in about ninety seconds.

The after-hours lot walk has a second participant. Plenty of stores build their whole used car merchandising strategy around the shopper who drives the lot on a Sunday. That shopper now has a research tool that works with one hand and does not require typing a stock number. Whether your VDP for that exact unit is thorough, current and readable determines what gets said back.

Ninety-seven languages, switched mid-sentence, with no Spanish page required. This one cuts both ways and deserves a careful read. A bilingual shopper can flip languages in the middle of a conversation about your store, and the model will answer in the new language using English-language sources. Your store does not need a translated website for that to happen, which sounds like free coverage. It also means your pricing conditions and your disclaimer language get translated by a machine you do not control, into a language your compliance review never saw. Anyone who runs conditional rebate offers should sit with that for a minute.

Your hours and your phone tree get spoken as fact. Voice queries cluster where hands are busy. Driving, standing under a lift, walking a lot with a kid on one hip. Those are the moments a shopper asks whether you are open, whether you take walk-in oil changes, and what the number is. Whatever your Google Business Profile says is what the model says. The gap between published hours and staffed hours stops being a website detail and becomes a spoken promise.

You cannot currently see it in your reporting. Google has not said whether Search Live conversations are broken out anywhere in the Search Console AI Mode data. We covered what that report does and does not count when it finished rolling out. Treat Search Live as unmeasured until Google says otherwise, and do not let a vendor tell you they are tracking it.

The counter-case, honestly. Google has published no usage numbers for Search Live. None. Not sessions, not queries, not a share of AI Mode. Nobody can tell you what fraction of car shopping runs through a voice conversation today, and anyone who quotes you a figure made it up. Google has also published nothing about how accurately Lens identifies a specific vehicle trim or package, which is the difference between a useful answer and a confidently wrong one about a car sitting on your lot. This was a model upgrade on an existing surface rather than a launch.

Here is why it still earns twenty minutes of your week. Every fix this demands is a fix you already owed somebody else. Text instead of images, accurate hours, complete VDPs, structured data that says what a vehicle is. All of it pays on the surfaces you can already measure, whether or not voice ever becomes a big channel for car shopping.

What to do about it

1. Open the Google app on your own phone and tap Live. Ask it, out loud, three questions a real shopper asks: whether you are open right now, whether you have a specific trim in stock, and what a service appointment costs. Write down what it says. This costs nothing and takes five minutes.
2. Go stand in front of a unit on your front line and point the camera at it. Tap Live, ask what it is and what it should cost. That is your after-hours shopper's experience, and you should know it before your competitor does.
3. Inventory every number that lives inside an image. Price, payment, down payment, APR, disclaimer, hours, specials. Move each one into real text on the page. Keep the graphic if you want it, just stop letting it be the only copy.
4. Confirm your VDPs render without JavaScript. If the vehicle details only appear after a script runs, you are relying on every reader in the chain to execute it. Some do. Plan for the ones that do not.
5. Reconcile posted hours across your Google Business Profile, your website and your phone room. Pick the hours you actually staff, then publish those everywhere.
6. Have somebody review your offer disclaimers for how they read when translated. Conditional rebates, lender-specific rates and stacked incentives are the ones that go wrong in another language.

The part worth remembering

Google spent a decade teaching everyone in this business to optimize a page for a screen. The surface it upgraded this week answers with a voice, a camera and a short list of links, and the screen is the smallest part of it.

The stores that will do well here are the ones whose information is simply true and simply readable. That is the same advantage it has always been. The difference now is that you can hear it fail from your own parking lot.

If you want a second set of eyes on how your inventory reads to a machine that can only work with text, book a walkthrough and we will run the audit with you.