Google is turning the simple search box into a two-way conversation. The company announced today that Search Live—the voice-and-camera feature powered by its newest AI model—is expanding to more than 200 countries and territories at once, the widest single launch of a conversational search tool in Google’s history.
Until now, Search Live was only available in a handful of test markets. With the global release, anyone who already sees the optional AI Mode tab in the Google app can tap a microphone icon, ask a question out loud, and receive a spoken answer. Users can keep talking, refine the query, or point the camera at an object to add visual context. The entire exchange feels closer to a video call than to typing keywords into a box.
What makes Search Live different from regular voice search?
Traditional voice search is a one-shot deal: you speak, Google replies with blue links. Search Live keeps the mic open. After the first answer, the screen stays active, encouraging follow-ups such as “What about the cost?” or “Show me how to fix it.” The AI remembers the thread, so you don’t have to repeat the subject.
Another big change is multimodal input. By tapping the camera button you can show the AI what you’re looking at—say, a leaking pipe under the sink—and ask “What tool do I need?” The system blends visual recognition with web knowledge and returns both an audio explanation and links to tutorials or shopping pages.
Behind the scenes: Gemini 3.1 Flash Live
Google credits the rapid rollout to Gemini 3.1 Flash Live, a lightweight variant of its flagship large-language model. The company says the model is “inherently multilingual,” meaning it doesn’t rely on separate translation layers. Instead, it was trained on text, audio, and image data in more than 70 languages from the start. The result, according to Google, is faster response times and more natural inflection in speech.
Engineers also reduced the model’s memory footprint so that it can run on-device for simple tasks such as spelling corrections, while heavier reasoning is off-loaded to Google’s cloud. The hybrid approach keeps latency low even on mid-range Android phones.
How to start a Live session step by step
- Open the Google app on Android or iOS and make sure you are signed in.
- Look for the pulsating Live microphone icon directly under the main search bar. If you do not see it, check Settings > Search > AI Mode and opt in.
- Tap the icon, grant microphone access, and ask your question out loud.
- To add visuals, tap the camera button that appears once the session starts. Point, steady the viewfinder, and wait for the white overlay to confirm the object is recognized.
- Keep talking. Each follow-up is timestamped at the bottom so you can scroll back to an earlier answer.
- End the session by tapping the X or saying “Stop.” All audio is processed in real time and, Google says, deleted after the session unless you have chosen to save activity in your account.
Where is AI Mode available?
Google publishes a running list of supported regions. As of today, the page lists more than 200 countries, including the entire EU, India, Brazil, Japan, Nigeria, and most of Southeast Asia. Notable omissions are mainland China, Iran, and Russia, where Google’s consumer services are either restricted or blocked. The interface adapts to local dialects automatically; for example, users in Mexico receive responses in Latin-American Spanish, while Spaniards hear Castilian pronunciation.
Privacy and data handling
Because Search Live is always listening for the next question, privacy advocates have raised concerns about accidental recording. Google says the mic is activated only after the user taps the Live icon and remains on solely for the duration of the session. Audio is encrypted in transit and stored temporarily to generate the answer. Users who disable Web & App Activity in their Google account will not have audio logged at all.
Visual queries follow the same rules as Google Lens: still images may be saved for up to 24 hours to improve object recognition, but they are not used to target ads. The company has not announced plans to watermark AI-generated answers, though it does flag when a response comes entirely from the model rather than from the open web.
Early user feedback and limitations
Reporters at Search Engine Land tested the feature in English, Spanish, and Japanese. Response accuracy for factual questions—such as “Who won the 2024 Euros?”—was high, usually citing a reliable sports site. Conversational drift still occurs: when asked about shelf installation, the AI occasionally recommended products unavailable in the tester’s country. Google acknowledges that localized inventory remains a work in progress.
Another limitation is device support. Although the rollout covers 200 countries, only phones running Android 10 or i
Leave a Comment