Story
August 26, 2026

Google’s Cleaner Dictation Comes With One Big Question: Did You Really Say That?

Google is rolling out Gemini 3.5 Transcribe across select products and developer tools, promising faster, multilingual voice input. Its ability to polish speech as it transcribes also puts accuracy and user intent under sharper scrutiny.

Google’s latest speech-to-text push promises to make voice input less messy, faster and more useful. But the same system that removes verbal clutter can also reshape what a speaker actually said.

The rollout began with Gemini 3.5 Transcribe, a new model Google says improves on its Chirp 3 transcription engine. It can recognize more than 85 languages, adapt to custom vocabulary, identify as many as three speakers in recorded audio, and automatically format text while stripping out fillers such as “um” and “uh.” Google calls it a tool that lets users “edit naturally with just your voice.”

On launch day, the company’s pitch was squarely aimed at developers as well as everyday dictation users. Sundar Pichai said the model could help build apps that understand “user speech / intent, even w/ multiple speakers,” highlighting language detection and specialized-jargon support. The model is available through the Gemini API, AI Studio and Antigravity, while the Gemini app on macOS and the Pixel 11’s Gboard Rambler feature are among its first consumer-facing homes.

Google’s performance claims are substantial: it says the trip from speech to finished transcription is about 70% faster, and puts its live-speech error rate at 5.5%, down from Chirp 3’s 7.32%. Chrome integration is planned “soon,” potentially extending the feature to text fields across the web.

Yet the appeal of polished transcription is also its catch. Removing stumbles and handling spoken corrections may produce cleaner prose, but it means the AI is doing more than hearing words—it is interpreting intent. One account of early testing found it effective on short passages, while warning that it “does technically change the wording of what you said.”

There was also a brief wrinkle in Google’s wider Gemini Audio messaging: after initially providing information about Gemini 3.5 Live and 3.5 Live Experimental updates, the company told The Verge that only Transcribe was being announced, without offering a new date for the other models.

Story coverage