Voice mode lets you ask your company brain a question by speaking, and have the answer read back. It is the same brain, the same permissions and the same citations as the typed experience - only the way in and out changes.

When it is genuinely better

Broad: documentation, wikis, public channels, the CRM. The material people are meant to share.
Restricted: payments, accounting, HR. Real value, real sensitivity.
Personal: individual mailboxes and drives, visible only to their owner.

Voice is not faster than typing for short, precise questions. It is markedly better in three situations: when your hands are busy, when the question is long and rambling and you would rather think out loud than compose a sentence, and when you are walking into a meeting and want the last three things that happened with a client without stopping to read.

How to use it

Open the ask box and choose the microphone.
Speak your question the way you would ask a colleague. Long, unstructured questions are fine - the brain handles those better than most search boxes.
Read or listen. The answer appears as text with its usual sources, and can be read back aloud.

What it does not change

Voice is a surface, not a shortcut around your rules. Everything you have set up still applies: a person only ever hears what their permissions allow, every answer still cites where it came from, and voice questions count towards usage exactly as typed ones do. Speaking a question does not make it reach further than typing it.

Accuracy and accents

Transcription is good but not perfect, and it is least reliable on unusual proper nouns - company names, product codenames, surnames. If an answer looks like it addressed a slightly different question, check the transcribed text shown above it before assuming the brain misunderstood. Nine times out of ten the transcript is where it went wrong, and correcting that line re-asks without you repeating yourself.