Guide

ChatGPT audio uploads: turn a recording into notes, with prompts that work

Published

A smartphone lying face-down on a wooden desk with white earbuds, an open notebook, a pen and a cup of tea in afternoon light

Short answerChatGPT can now take an audio file on paid plans and turn it into a transcript, a summary, action items or a follow-up email. OpenAI added it on 6 October 2026. It isn't on the Free plan, transcripts can contain errors, and you should get everyone's consent before recording them.

OpenAI's release notes list audio uploads for ChatGPT on 6 October 2026: you attach a recording and ask for a transcript, a summary, or answers about what was said. It's for paid plans and workspaces. A guide at FindSkill quotes OpenAI as saying audio uploads aren't available on the Free plan "at this time". Availability can also vary by workspace settings, region, app version and model.

Below is what to know before you rely on it, then prompts for the four jobs people will use it for most.

The practical details

These come from the FindSkill write-up of OpenAI's help page. Check OpenAI's own page before you depend on a limit.

  • Formats: WAV, MP3, OGG, FLAC, AAC, M4A, PCM, and audio-only WebM or MP4.
  • Video files: a video saved as MP4 or WebM isn't supported. Export just the audio track first.
  • Size: up to 512 MB per file in the ChatGPT app. The 25 MB figure you'll see quoted online belongs to OpenAI's developer API, not the app.
  • Length: no maximum is given. Long recordings may be processed in sections and could time out.
  • Rough sizes: a 45-minute MP3 at 128 kbps is about 43 MB, and the same length as CD-quality WAV is about 476 MB. Choose MP3 or M4A for long recordings.

Be honest about accuracy

OpenAI itself warns that transcripts may contain errors and that speaker identification may be unreliable. Performance can also vary between languages and accents. So the best habit is a quick test: upload a two-minute clip you know well and compare the result with what you remember. If names, numbers or dates are wrong in the test, they'll be wrong in the long version too.

Two rules that cost nothing:

  1. Tell the AI who is speaking. Putting the speakers' names in the prompt helps more than trusting automatic labels.
  2. Verify anything that matters. Dates, amounts and promises should be checked against the recording before they go into an email.

Prompt 1: a meeting into minutes and actions

prompt — 2 blanks
This is a recording of a meeting between [NAMES AND ROLES]. Produce:
1. A 3-sentence summary.
2. Decisions made, and who made each one.
3. Open questions, each with an owner.
4. Action items as a table: owner, task, deadline.

Only include commitments that were actually said. If a name, number or date is unclear, write [CHECK] next to it instead of guessing. If you can't tell who said something, say so.

The last paragraph is the important one. It stops the AI inventing a deadline nobody agreed to. For a tidy format, our meeting minutes prompt is a good template, and you can feed the output to the weekly status report prompt.

Prompt 2: a lecture into study notes

prompt — 2 blanks
This is a recording of a lecture on [SUBJECT]. Write study notes in plain language: the main idea in two sentences, the key terms with one-line definitions, and the five points most likely to be asked in an exam. List anything the lecturer said they would cover later or that I should look up. Mark any term you're unsure you heard correctly with [CHECK].

Then use the notes to flashcards prompt or the one-page revision sheet prompt to turn the notes into revision material. Recording a lecture usually needs the lecturer's permission, so ask first.

Prompt 3: a voice note into an email

Many of us think better out loud. Record a messy two-minute voice note, then:

prompt — 2 blanks
This is me thinking out loud about [TOPIC]. Turn it into a clear email to [PERSON] in my voice: polite, direct, under 150 words. Keep my main point first, remove the false starts, and don't add any facts I didn't say. Give me one subject line.

Prompt 4: an interview into a quote sheet

prompt — 1 blank
This is an interview with [NAME, ROLE]. Pull out the 8 strongest quotes, word for word, with a timestamp if you can work it out. Group them by theme. Then summarise their position in 3 bullets. Do not paraphrase inside the quote marks.

If you plan to publish a quote, play that part of the recording again first. Treat the transcript as a working draft.

A recording of a person is personal data. A few sensible rules:

  • Get a clear yes before you record. Recording laws differ by place, and some US states require every participant to agree.
  • In the UK and EU, voices that identify someone fall under data-protection rules, so tell people you're recording and that AI will help write the notes.
  • Know where the file lives. Deleting a chat does not necessarily delete a file saved in your Library, so remove that separately.
  • Keep the very sensitive stuff out. For medical, legal or HR conversations, check your organisation's rules first.

No paid plan?

Dictate to ChatGPT with voice input, or run the recording through a free transcription tool, then paste the text into any chatbot with the prompts above. The prompts work the same on text.

Where to go from here

For summary habits that apply to any input, see our ChatGPT productivity prompts and the ChatGPT Voice in Work guide. You'll find more ready-made prompts in the ChatGPT prompts collection, and the prompt builder can adapt these to your own recordings.

Quick checklist

  • ChatGPT audio uploads were added on 6 October 2026, for paid plans only.
  • Supported: WAV, MP3, OGG, FLAC, AAC, M4A and audio-only WebM or MP4, up to 512 MB.
  • Test with a short clip first, and verify names, numbers and dates.
  • Put the speakers' names in your prompt.
  • Tell the AI to mark anything unclear instead of guessing.
  • Get consent before recording anyone.

Prompts to try

Browse all 1611 prompts →

Keep reading
All guides →