ChatGPT Now Writes Your Prompt as You Talk, and OpenAI Dropped the 10-Minute Recording Limit
OpenAI said on November 18 that Live Dictation turns speech into text as you talk, on every ChatGPT plan, and that the 10-minute recording limit is gone.
ChatGPT will now type your prompt while you are still talking. In release notes dated November 18, OpenAI says Live Dictation turns speech into text as you go, so you wait less for the transcript to finish. Tap or click the waveform to show or hide that text. The company also removed the 10-minute recording limit.
The rollout is global, on every ChatGPT plan, across iOS, Android, web and the desktop app. OpenAI's line is that dictation is getting faster and more reliable. It does not publish a new accuracy number with this drop.
That last point matters because ChatGPT already has a spoken conversation mode, and the two are easy to mix up. OpenAI's Voice help page draws the split: use Voice for a live back-and-forth; use Dictation when you want to record a prompt, edit the transcript, then send it as text. Voice transcripts, the company says, are not verbatim records.
What actually changed at the mic
The old dictation path waited. You held the microphone, finished the thought, then sat through a transcription. Live Dictation streams the words into the prompt box while you are still speaking, which is the difference between a voice memo and a keyboard.
The 10-minute cap was the other constraint. Long prompts, meeting recaps, or a rambling spec no longer die at an arbitrary cutoff. OpenAI does not say how long a single take can now run, only that the old limit is gone.
The Voice Dictation FAQ still describes the underlying deal: press the microphone, the audio goes to OpenAI's models, text comes back for you to edit before it becomes a user message. Audio stays attached to the chat until you delete that chat, then OpenAI says it deletes the clip within 30 days unless a security, legal or prior training-share exception applies.
Not Voice, and not Record
Live Dictation is a prompt tool. It is not GPT-Live, the conversation mode OpenAI is pushing as Advanced and Standard Voice head toward a December 11 retirement. It is also not Record mode, the meeting-and-voice-note product that still sits on its own help page and, last time OpenAI described it, was limited by plan.
The company has been trying to make ChatGPT something you talk to instead of something you type at, from Voice on the phone to the productivity workflows it already sells. Streaming the transcript is a smaller change than a new model, and an easier one to feel: you can see the sentence form while you are still finishing it.
The edit step is the whole product. If the live text is wrong, you fix it before send. That is the part Voice, by OpenAI's own description, does not promise.
Some offers on this page may be paid placements or contain affiliate links.
Subscribe to Techpresso
Free daily newsletter, read in 5 minutes.
Subscribe free