Can ChatGPT Transcribe Audio? Yes, With These Catches

ChatGPT has become a handy transcription tool, but there are three different features people mix up: uploading an audio file, record mode on a Mac, and voice dictation. They do different jobs, they're on different plans, and none of them is word-perfect. Here's which to use and what to watch for.

Three ways ChatGPT handles audio

Feature	What it does	Where it works
Audio upload	Transcribes a recording you already have	Plus and work plans, not Free
Record mode	Records a meeting live, then transcribes and summarises it	Mac desktop app, on Plus, Pro and work plans
Dictation and voice	Turns what you say now into a prompt or a spoken chat	Most apps

How to transcribe an audio file with ChatGPT

1. Open a new chat.
2. Click the attach button and choose your audio file.
3. Type a clear request, such as: Transcribe this recording word for word, with a new paragraph for each speaker.
4. Send it and wait for the transcript.
5. Read it through against the recording before you use it.

Supported formats include MP3, WAV, M4A, AAC, OGG and FLAC. If the upload fails, the file may not have a readable audio track, which happens with some screen recordings and odd exports. Convert it to MP3 and try again.

The catch: it isn't word-perfect

OpenAI's own help page warns that transcripts may contain errors and that identifying who's speaking may be unreliable. In practice that means names, numbers, jargon and anyone talking over each other are where mistakes creep in. For a blog draft or meeting notes that's usually fine. For anything legal, medical or that you'll quote publicly, check every line against the recording.

Speaker labels need extra care. ChatGPT may guess who said what, and it can guess wrongly. If speakers matter, tell it their names and roles in your prompt, then check the labels.

Record mode on a Mac

Record mode is built for meetings and calls. In the ChatGPT desktop app for Mac, click the Record button in a chat, run your meeting, then stop. ChatGPT uploads the transcript and creates notes you can ask follow-up questions about. OpenAI says the audio is used only for transcription and then deleted. It's available on Plus, Pro and work plans, and only in the Mac app.

Tell people you're recording before you start. It's courteous, and in many places it's a legal requirement.

Dictation and voice

If you just want to speak instead of type, use dictation: tap the microphone, talk, and ChatGPT turns your words into a prompt you can edit before sending. Voice conversations are different again: you talk and ChatGPT talks back. OpenAI notes that voice transcripts aren't verbatim, so neither is the right tool for an accurate record.

Prompts that get better transcripts

• Word for word: 'Transcribe exactly, including filler words.'
• Clean read: 'Transcribe, but remove ums, repeats and false starts.'
• With speakers: 'The speakers are Sam (host) and Priya (guest). Label each paragraph.'
• With timestamps: 'Add a rough timestamp at the start of each section.'
• Then repurpose: 'Now turn this into show notes with five key points.'

When to use a dedicated transcription tool

ChatGPT is ideal when you want a transcript and something made from it, like notes, quotes or a summary, in the same place. For long recordings you transcribe every week, accurate speaker labels, subtitles or searchable archives, a dedicated transcription service is usually more reliable. My guide to AI video summarisers covers tools that work on recordings directly.

Keeping recordings private

Before uploading, think about what's in the recording. Client calls, staff conversations and anything with personal details deserve care. Check your ChatGPT data settings, and on a work plan follow your organisation's rules. If a recording includes people who haven't agreed to it being processed, don't upload it.

Turning a transcript into content

The real value of transcribing in ChatGPT is what you can do next in the same chat. Once the transcript is there, ask for a summary, a list of action points, a few quotable lines, a blog outline or social posts drawn from it. Check each piece against the transcript before you publish, because a summary is only as accurate as the words it's built on, and a misheard name can travel straight into a post if you don't.

Getting the best audio in the first place

• Record close to the microphone, in a quiet room.
• Ask people not to talk over each other on important points.
• Spell out names, brands and numbers at the start of the call.
• Split very long recordings into parts, so a problem in one doesn't spoil the lot.

My verdict

Yes, ChatGPT transcribes audio well enough for notes, drafts and repurposing, as long as you're on a plan that allows uploads and you check the result. Upload the file and ask for a word-for-word or clean transcript, or use record mode on a Mac for meetings. Don't trust names, numbers or speaker labels without checking. For more ways to use it at work, see using ChatGPT for business development, and my explainer on context windows for why very long recordings can trip it up.

Frequently asked questions

Can ChatGPT transcribe an MP3?

Yes. Attach the MP3 to a chat and ask for a transcript, on a plan that includes audio uploads.

Can ChatGPT transcribe audio on the Free plan?

No. Audio uploads aren't offered on the Free plan.

Is ChatGPT's transcription accurate?

Mostly, but OpenAI warns transcripts may contain errors and speaker labels may be unreliable.

What is ChatGPT record mode?

A Mac desktop app feature that records a meeting, transcribes it and creates notes.

What audio formats does ChatGPT accept?

Formats include MP3, WAV, M4A, AAC, OGG and FLAC.

Does ChatGPT keep my record mode audio?

OpenAI says the audio is used only for transcription and then deleted.