Put the address in writing
Voice answers need somewhere to leave the details.
Retrospective field note for 2026-06-22, reconstructed on 16 November 2026. The archive date is not a claim of publication at the time.
Reading six prices and their cancellation rules aloud is a good way to make someone ask for the list again.
Speak the conclusion first. Put the comparison, address or exact code in writing where it can be opened and copied. Voice and text can do different parts of the same job without forcing either one to carry everything.
Streaming makes speech start sooner. It also makes corrections awkward if the answer begins before a fact is checked. For consequential information, a short pause can be better than a quick statement followed by a retraction.
Test interruption. Can the person stop the response? Does the assistant retain the correction, or restart the same explanation? A pleasant synthetic voice does not compensate for having to wait through an unwanted paragraph.
Short sentences help, but the larger edit is deciding what should be spoken at all. The interaction should end with the person knowing the answer and having the precise details available for the next step.
Google added streaming speech generation on 17 June. Source.