OpenAI has rolled out an upgraded version of dictation in ChatGPT, and the change addresses a long-standing complaint about the old system. Spoken words now appear on screen as you say them, rather than after a pause while the app processes the recording. The company has also dropped the 10-minute cap that used to cut off long voice sessions.
The new Live Dictation feature is rolling out worldwide to every ChatGPT tier, free and paid, on iOS, Android, the web and the desktop apps. When it reaches an account, nothing extra needs setting up: you tap the microphone icon and start talking.
The old setup was a blind process. You spoke and waited, then found out afterwards whether ChatGPT had understood you. On a short prompt, a misheard word is a small annoyance. On a five-minute ramble about a project plan or a messy debugging problem, it could mean searching a wall of text for the spot where the transcription went wrong. Seeing the words as they arrive lets you correct course mid-sentence, before small errors add up.
Removing the time limit may matter more for some people. Many users treat ChatGPT as a thinking partner, talking through ideas out loud before shaping them into a prompt. Ten minutes sounds generous until you’re in the middle of explaining something complicated and the recording stops. That limit no longer applies, though OpenAI hasn’t said whether a new upper limit has replaced it.
There’s also a small comfort setting for people who find a scrolling transcript distracting. Tapping or clicking the audio waveform while dictating hides or shows the live text, so you can speak without watching yourself type.
The details are thinner than the announcement suggests. OpenAI calls the new dictation faster and more reliable, but it hasn’t published latency figures, accuracy benchmarks or any data showing how big the improvement is. Until people test it on real accents, noisy rooms and technical vocabulary, those claims are the company’s word.
Privacy settings are worth checking too. OpenAI says audio from dictation can be kept out of model training through ChatGPT’s data controls. You can turn off the option to include audio recordings or stop sharing for model improvement altogether. Anyone dictating sensitive work material should look at those settings before talking at length.
The update also says something about where OpenAI is heading. Built-in dictation on phones and computers already works in any text box. By making its own version faster and uncapped, OpenAI is betting that people will speak to ChatGPT directly instead of relying on those built-in tools. The next question is whether long-form voice input becomes as normal as typing, or stays a feature mostly used by people who already prefer talking to typing.
