ChatGPT audio uploads let you drop a recording into a chat and get a transcript, a summary or action items back. OpenAI added the feature on October 6, and it’s limited to paid plans for now. If you’ve got a meeting, lecture or interview sitting in a voice memo, this is the quickest way to turn it into notes.
It isn’t perfect, though. Transcripts can contain errors, and what happens to your file afterward is worth a minute of your attention.
Here are the quick facts:
- Added: October 6, 2026, per OpenAI’s release notes.
- Available on paid ChatGPT subscriptions and workspaces. Free isn’t included “at this time.”
- You attach the file like a PDF and say what you want back.
- Output includes transcripts, summaries, structured notes and follow-up email drafts.
Now for the details that decide whether it fits your workflow.
How ChatGPT audio uploads work
You start a new chat, attach a supported audio file, and describe the result you want. According to OpenAI’s release notes, examples include turning a meeting, interview or lecture into structured notes or a follow-up email draft.
You can also ask questions about the recording afterward, such as “what did we decide about the budget?” The same chat keeps the context, so follow-ups are quick.
OpenAI says availability may vary by workspace settings, region, client version and model. If the option is missing, update the app or try the web version first.
Supported formats and file limits
OpenAI’s own entry doesn’t list formats or sizes. A third-party FindSkill guide, which cites OpenAI help pages, gives the numbers below, so treat them as reported rather than confirmed by us.
| Item | What the guide reports |
|---|---|
| Formats | WAV, MP3, OGG, FLAC, AAC, M4A, PCM, plus audio-only WebM and MP4 |
| Max file size | 512 MB per file |
| Max length | Not stated; very long files may time out |
| Not supported | WebM or MP4 files that are tagged as video |
The 25 MB cap people sometimes quote belongs to OpenAI’s developer speech-to-text API, not the ChatGPT app, per the same guide. If you have a video call recording, export the audio-only track first.
Get better notes from a recording
A vague prompt gives vague notes. Tell ChatGPT what the recording is and who the notes are for.
- Name the output: summary, decisions, action items or a draft email.
- Ask for owners and deadlines when people made commitments.
- Ask it to flag anything it wasn’t sure it heard correctly.
- Request quotes only if you’ll verify them against the audio.
That last habit matters. Spot-check names, numbers and dates against the recording before you send anything. The guide also warns that speaker labels can be unreliable, so don’t lean on them for anything consequential. For live voice chats rather than files, see our post on the ChatGPT voice mode update.
Uploads versus Record and the Meetings plugin
Uploads are for files you already have. ChatGPT has two other routes for capturing audio live, and they work differently.
| Option | Best for | Reported limits |
|---|---|---|
| Audio uploads | Existing recordings, voice memos, call exports | Paid plans; 512 MB per file |
| Record | Capturing a session live | macOS desktop app only; 4-hour sessions |
| Meetings plugin | Live meeting capture | macOS only; 4-hour cap; beta for Pro and Business |
These details come from the FindSkill guide and may change as the features mature. The guide also notes that the upload help page makes no promise that audio is deleted after processing, unlike Record and the Meetings plugin.
In practice, that means uploads suit recordings you made elsewhere, such as a phone voice memo. If you’re on a Mac and want live capture, the other two are the ones to look at.
What happens to your audio file
This is the part most people skip. Retention depends on where the file is stored and on any workspace retention policy, according to the FindSkill guide.
It adds that files saved in Library can be kept separately from the chat, and deleting a chat doesn’t delete a saved file. On consumer plans, whether content helps train models depends on your data settings, so check “Improve the model for everyone.” OpenAI says it doesn’t use ChatGPT Enterprise or API content for training.
If the recording is sensitive, delete the saved file yourself afterward. And if you’re weighing other transcription options, our rundown of Microsoft’s MAI-Transcribe-2 covers a competing model.
Frequently Asked Questions
Are ChatGPT audio uploads available on the free plan?
Not at this time. OpenAI says audio uploads are available with paid subscriptions and workspaces.
How big can an audio file be?
OpenAI’s entry doesn’t say. A third-party guide reports 512 MB per file and no stated length cap, so very long recordings may time out.
Is the transcript accurate?
OpenAI warns that transcripts may contain errors and that performance varies across languages. Check key names, numbers and dates against the original.
Does deleting the chat delete my audio file?
Not necessarily. Per the FindSkill guide, a file saved in Library can outlive the chat, so delete it separately if it’s sensitive.
Can ChatGPT tell who is speaking?
Sometimes, but not reliably. Treat speaker labels as a hint, especially for anything that affects people or money.
My take: it’s worth using for your own lectures, brainstorms and routine meetings, where a small mistake costs little. For legal, medical or HR recordings, keep a human in the loop and think hard before uploading at all. Fast notes are great, but only if you’ve checked them.
Leave a Reply