Answer
Review the recording and any transcription before sending; the composer lets you edit the transcript when available. With AI on, sending requests a Text AI response. With AI off, it sends a voice message to the chat. This is separate from Audio mode, which creates music. Speaking a prompt also works in Image, Video and Audio: the transcription supplies the prompt text, and its associated recording is not a generation reference or a forbidden Audio attachment. An arbitrary audio attachment does not receive this exception. In private and group chats on updated web, phone browsers and native iOS/Android, after Stop the microphone button shows upload and transcription progress without an extra loading square in the attachments row; the uploaded recording preview appears before sending. Older app builds may differ.