OpenAI is rolling out Live Dictation in ChatGPT, which it announced on X on 18 November. Your words now appear on screen while you speak, instead of after you stop recording. The 10-minute limit on a single dictation is gone too. According to reports on the rollout, every ChatGPT plan gets it worldwide, on iOS, Android, the web and the desktop app.
Your words appear while you talk, and a tap on the waveform hides them
Until now, dictation in ChatGPT took two steps. You recorded, pressed stop and waited for the transcript. With Live Dictation, the text builds up as you speak, so there is little left to wait for when you finish. Tap or click the waveform to show or hide the running text. You can still edit the transcript before you send it.
Dictation is the microphone button that turns speech into a typed message. It is separate from ChatGPT Voice, the mode for spoken conversations. This update is about dictation only.
The 10-minute cutoff is gone, and OpenAI names no new maximum
Atty Eleti, who works on voice features at OpenAI, wrote on X that the 10-minute dictation limit was removed as part of the upgrade. For anyone who dictates at length, that clears a real obstacle. A meeting recap, a site visit report or the first draft of a memo no longer has to be split into chunks. OpenAI has not said whether a new upper limit applies, so try a long recording before you build a routine around it.
Dictation audio stays with the chat until you delete it
Longer dictations mean more recorded audio, so the storage rules matter more than before. OpenAI’s Voice Dictation FAQ says dictation audio is kept for as long as the chat is in your history. When you delete the chat, OpenAI deletes the audio clip within 30 days, unless it has to keep it for security or legal reasons. Audio you already shared for training, once separated from your account, is not deleted with the chat.
Can you stop OpenAI from training on your dictation?
Yes. On a personal account, OpenAI may train on dictation audio if you chose to share audio to improve its models. Go to Data controls in the ChatGPT settings and switch off “Include your audio recordings”. Or turn off “Improve the model for everyone” completely. That stops audio sharing for both dictation and voice chats. For business plans, OpenAI says it does not train on inputs or outputs by default.
Dictate client notes in a business workspace, not on a personal account
For the organisations DataNorth works with, this makes ChatGPT a realistic tool for long spoken notes. Think of a consultant recapping a client meeting in the car, or a project manager reporting from a site visit. Notes like these usually name clients, patients or colleagues. On a personal account, that information falls under OpenAI’s consumer rules and can end up in training data if audio sharing is on.
So check the data controls on the personal accounts that staff use for work, and agree which details may be dictated at all. Where people need to dictate client information, give them a ChatGPT Business or Enterprise workspace instead. And test a long recording in your own language and jargon first, because OpenAI does not publish a list of languages for dictation.
For more information, visit the official announcement of Live Dictation on ChatGPT’s X account.