You speak a name, catch yourself, and say a different one. A moment later the screen shows only the second name, sitting in a tidy sentence. The stumble never made it onto the page.

On August 26, Google introduced Gemini 3.5 Transcribe. The company says the model converts raw audio into polished, formatted text and can clean up the verbal stumbles people make while talking. That cleanup is the part worth carrying into the weekend. It can also hide a change.

Google's developer documentation shows two modes. Verbatim keeps the fillers and the mid-sentence correction. Smart mode can drop them. In one example, a person says they should invite Alice, then corrects to Bob and Carol. The smart transcript reads: "For the meeting, I think we should invite Bob and Carol." Alice never appears.

That example lives in API documentation. People may meet similar cleanup in everyday tools. Ars Technica reports that the same model already powers Gboard's Rambler feature on Pixel 11 phones and voice input in the Gemini app on macOS. The same report says Google will expand Rambler later this year and bring the transcription feature to Chrome. Ryan Whitwam, who tested Rambler, wrote that the system does change the wording of what you said, and that may not fit every situation.

Google's earlier Rambler post also said the feature uses audio in real time and does not store it. So the recording may not be sitting there waiting for you. When a tool polishes as you speak and discards the audio, the text is what remains.

Before you send it

This weekend, pick one voice note, dictation, or spoken message. If the app kept the audio, play it once against the text and mark one change: a dropped hesitation, a missing first name, a number that got cleaned. If there is no recording, read the text out loud before you send it. When a name, amount, or decision matters, keep a voice memo in an app that saves the file.

The polished sentence can be easier to read. Keep enough of the original to know what you actually said.

Public sources