On June 18, 2025, Google announced that AI Mode, its experimental Search interface, could now hold voice conversations. Users could tap a Live icon in the Google app, ask a question out loud, hear an AI-generated audio answer, and keep asking follow-ups. It was the next step for a feature that had rolled out to all US users only weeks earlier, in May 2025. The feature was small; the signal was not, and the details are worth unpacking.
For Search as a product, the change was bigger than adding a microphone. Voice turned the single query box into a session you could interrogate continuously — and Google adjusted not just the interface but the way content gets pulled from the web to answer. That second part is what mattered most to developers and site owners.
What the feature looked like
Per TechCrunch and The Verge, the flow worked like this: opt into the AI Mode experiment through Google Labs, open the Google app on Android or iOS, tap the new Live icon, and speak your question. The system answers in audio, relevant links appear on screen during the conversation for users to open and verify, and you continue asking follow-ups as if talking to a person.
The details around the edges made it usable rather than a demo. Search Live ran in the background, so the conversation continued while you switched apps. A transcript button let you read answers or switch to typing, and past conversations were saved to your AI Mode history for later reference. Those choices made it feel like a working session rather than a one-off question and answer. From a product design standpoint, each one lowered the barrier to speaking up: background execution meant users did not have to sit on the search page, the typing fallback meant quiet offices still worked, and saved history meant voice conversations started accumulating reviewable context.
Under the hood
Google’s Liza Ma said Search Live in AI Mode used “a custom version of Gemini with advanced voice capabilities.” Underneath the voice interface, answer quality came from the same query fan-out technique that already drove AI Mode: the question is broken into multiple sub-queries, web content is pulled from a broader and more diverse set of sources, and the results are synthesized into a single answer.
Google positioned this as an extension of Search’s existing quality systems — ranking and content evaluation carried over into the conversational flow. For publishers, it foreshadowed a problem that would only grow over time: when the answer is spoken aloud, the visibility and click behavior of links change, and the way content gets consumed changes with them. For developers who live on search traffic, query fan-out was the part worth studying more than the voice layer: one spoken question became several sub-queries, and content got matched at paragraph-level intent rather than page-level keywords.
Availability and limits
The rollout was deliberately narrow: US users in the AI Mode experiment via Google Labs, voice only. The Verge noted the test did not yet support camera sharing, and Google said capabilities like real-time camera-based questions would arrive in the coming months — features it had already previewed at Google I/O in May.
Worth separating, too: Search Live already existed in the Google app with camera support, letting users point the camera at an object and ask about it. The June news was the integration of Search Live into AI Mode, bringing continuous voice conversation to the experimental search experience. The two should not be confused, even though the names travel together in coverage.
The voice search competitive picture
Placed in mid-2025, this was a late but inevitable move: OpenAI’s ChatGPT had already built habits around Advanced Voice Mode, and Perplexity was pushing voice answers hard. Google’s advantage was distribution — the Search app’s install base and daily query volume were an entrance no standalone assistant could easily copy. The ecosystem angle mattered too: binding voice conversation to the Google app rewrote the phone’s most routine action — opening a browser to look something up — into a conversation, ground that both Apple and OpenAI were contesting at the time.
From 2026, the small test reads as the starting point of Search’s voice turn: queries became conversations, answers became audio, and links became on-screen supporting material. The change it marked for developers was clear at the time — optimizing content structure for voice interaction had started to matter as much as optimizing for clicks.
Sources
- Google’s AI Mode can now have back-and-forth voice conversations - TechCrunch
- Google tests real-time AI voice chats in Search - The Verge
AI-assisted summary compiled from the sources above, reviewed by a human before publishing.