Can ChatGPT Transcribe Audio? What Works in 2026
Dzianis Kukareka, Founder · Updated

Yes, ChatGPT can transcribe audio, but only on a paid plan. As of November 2026, you can attach an audio file (MP3, WAV, M4A and several other formats, up to 512 MB) to a chat and ask for a transcript. The Free plan can't upload audio, video files are rejected by the audio uploader, and OpenAI itself warns that long recordings may time out and speaker identification may be unreliable.
That makes ChatGPT a good option for a short voice memo you want to discuss right away, and a weak one for a two-hour interview you need to quote accurately. Below: the four ways ChatGPT handles speech, the exact limits, a prompt that keeps the output literal, and how to tell when you need a dedicated transcription service instead.
Can ChatGPT transcribe audio files? The four routes
"ChatGPT transcription" can mean four different features. They have different limits, so it helps to know which one you're actually using.
| Route | Who gets it | Input | Main limits |
|---|---|---|---|
| Audio file upload in a chat | Paid plans only | WAV, MP3, OGG, audio-only WebM, PCM, FLAC, AAC, M4A, audio-only MP4 | 512 MB per file; long files are best-effort and may time out |
| Dictation (microphone icon) | Built into the message box | Your voice, live | Meant for dictating a prompt, not for transcribing a recording |
| Record mode | Plus, Pro, Business, Enterprise, Edu | Live microphone and system audio | macOS desktop app only; 4 hours per session |
| OpenAI API (developers) | Pay per use | mp3, mp4, mpeg, mpga, m4a, wav, webm | 25 MB per request; you write the code |
The upload limits come from OpenAI's help article on uploading files and audio to ChatGPT, Record mode details from the ChatGPT Record page, and the API limits from the file transcription guide. OpenAI updates these pages often, so check them if a limit matters to you.
How to transcribe an audio file with ChatGPT
Check your plan. Audio uploads are listed for paid subscriptions and workspaces. On the Free plan the option isn't available at the moment. See ChatGPT pricing for current plans.
Make sure the file is audio, not video. The uploader accepts audio-only MP4 and WebM but rejects files identified as video. If you have a screen recording or a camera clip, extract the audio first.
Attach the file. Use the + button in the message box and pick the recording.
Ask for exactly what you want. "Transcribe this" is vague. Use the prompt in the next section.
Review the result against the audio. OpenAI's own note on the upload page is that transcripts may contain errors. Spot-check names, numbers and any line you plan to quote.
If the upload works but the transcript never appears, OpenAI's troubleshooting advice is to convert the recording to WAV and try again.
A prompt that keeps the transcript literal
A chat model is built to be helpful, which sometimes means tidying up. Fillers vanish, broken sentences get repaired, and a rambling answer turns into a neat paragraph. For notes that's fine. For research, journalism or legal work it isn't. Spell out the rules:
Transcribe the attached recording word for word in its original language. Do not summarize, translate, correct grammar or remove repetitions. Start a new paragraph each time the speaker changes and label speakers as Speaker 1, Speaker 2 and so on. Mark anything you can't make out as [inaudible]. If the recording is too long to finish in one reply, stop at a sentence boundary and tell me where you stopped.
The last instruction matters most. If processing stops partway, you want to know, not get a transcript that quietly ends early.
Can ChatGPT transcribe audio from a video or YouTube?
Not directly from a video file. The audio uploader rejects MP4 and WebM files that contain video, so you need to strip the audio first. Our free MP4 to MP3 converter runs in your browser and doesn't upload the file anywhere. Then attach the MP3.
A YouTube link isn't an audio file, so the upload route doesn't apply. Most public videos already have captions you can copy. Our guide on getting the transcript of any YouTube video covers the free ways to do it.
Can ChatGPT transcribe audio for free?
Not from a file. The Free plan can't upload audio. The microphone icon in the message box (dictation) turns your speech into text, but it's designed for writing messages, not for transcribing a recording you already have. OpenAI's dictation FAQ describes it as recording a clip that is transcribed into an editable message.
If you want free transcription of an existing file, look at tools built for that. Our roundup of free AI transcription tools and their real limits compares the options, including running OpenAI's open-source Whisper model yourself.
Where ChatGPT transcription falls short
Long recordings
OpenAI says longer recordings "may be processed in smaller sections" when Data Analysis is available, that processing is best-effort, and that very long recordings may time out. In practice that means the longer the file, the more you should check that the transcript covers the whole recording, from the first minute to the last. Look at the final lines and compare them with the end of the audio.
Made-up text
Speech models can produce fluent sentences nobody said. A 2024 study of OpenAI's Whisper API, Careless Whisper, found that about 1% of transcriptions contained entire hallucinated phrases or sentences, and the problem was worse for speakers with long pauses. ChatGPT runs newer models than the one in that study, but the lesson still holds. Silence, music and crosstalk are where invented text tends to appear, and a polished-looking transcript gives you no warning.
Speakers and timestamps
The upload page states that speaker identification may be unreliable. Record mode does label speakers, using generic labels like Speaker 1 that you can rename. The help pages don't say anything about timestamps or subtitle export for uploaded files. OpenAI documents word-level timestamps only in the API, and only for the older whisper-1 model. Treat any timecodes in a chat answer as unverified until you check them against the audio. That matters if you want to turn the transcript into captions.
Languages
OpenAI notes that transcription performance varies across languages, and the Record page says it works best in English. For interviews in other languages, or recordings that switch languages, test a short clip first.
Privacy
On consumer plans, whether your content is used to improve models depends on your data settings. The Record page also says transcripts and canvases may be used for training when "Improve the model for everyone" is on, while the audio itself is deleted after transcription. Business, Enterprise and Edu content is excluded from training by default. If you transcribe interviews with research participants or clients, check your settings before you upload.
ChatGPT vs a dedicated transcription service
| You need | ChatGPT is fine | A transcription service is better |
|---|---|---|
| A quick text version of a short voice memo | Yes | Not needed |
| To ask questions about a recording | Yes, that's its strength | Useful only if you want a summary next to the transcript |
| A full transcript of a 1–3 hour recording | Risky: best-effort, may time out | Yes |
| Reliable speaker labels | Unreliable for uploads | Yes, with an editor to fix them |
| Timestamps you can click to replay audio | Not documented for uploads | Yes |
| SRT or VTT subtitles synced to video | No documented export | Yes |
| Video files | Extract audio first | Usually uploaded as is |
The difference isn't the speech model so much as everything around it. A transcription tool keeps the text tied to the audio. Each word has a time, you can click a line and hear it, fix a misheard name, and export the result in the format you need. A chat window gives you a block of text and asks you to trust it.
If you need that kind of transcript, BlaBlaScribe's audio-to-text tool takes audio and video files up to 8 GB, including MP3, WAV, M4A, MP4 and MOV. It adds speaker labels and word-level timestamps, and exports SRT, VTT, TXT or DOCX. Accuracy is up to 99.8% on clear audio, and you get 30 free minutes once when you sign up. Short files are transcribed in full, and for long files the free minutes cover the first 10 minutes. Paid plans start at $14/month, with minute packs from $5 (see pricing). A sensible workflow is to transcribe there, then paste the text into ChatGPT if you want to discuss it.
If you're comparing all the options, our guide on how to transcribe audio to text covers every method side by side. If you're tempted to skip ChatGPT and use OpenAI's model directly, see how to use OpenAI Whisper and Whisper vs paid transcription tools.
Is ChatGPT transcription not working? Common fixes
No option to attach audio: you're probably on the Free plan, or your workspace admin has limited uploads.
Upload rejected: the file is a video, or the format isn't supported. Extract the audio or convert it to WAV or MP3.
Uploads, but no transcript: convert to WAV and upload again, as OpenAI recommends.
Transcript stops early: split the recording into shorter parts and transcribe them one at a time.
Record button missing: Record mode only exists in the macOS desktop app on eligible plans. Update the app and check that ChatGPT has Microphone and Screen & System Audio Recording permissions.
"Upload limit reached": OpenAI caps uploads at 80 files every 3 hours, and failed attempts can count toward the limit.
Frequently asked questions
Can ChatGPT transcribe audio to text?
Yes, on paid plans. You attach an audio file of up to 512 MB to a chat and ask for a transcript. OpenAI says the transcript may contain errors and speaker identification may be unreliable, so check it against the recording.
Can ChatGPT transcribe audio for free?
Not from a file. As of November 2026, audio uploads aren't available on the Free plan. Dictation through the microphone icon turns a spoken prompt into text, but it isn't designed for existing recordings.
How long can an audio file be for ChatGPT?
The limit is set by size, not duration: 512 MB per audio file. OpenAI says long recordings are processed on a best-effort basis and very long ones may time out. Record mode in the Mac app caps a session at 4 hours.
Can ChatGPT transcribe a video?
Not directly. The audio uploader rejects MP4 and WebM files identified as video, so extract the audio track first and upload that as MP3 or WAV.
Can ChatGPT transcribe audio in different languages?
Yes, but OpenAI notes that performance varies across languages, and Record mode works best in English. Test a short sample before you transcribe a long non-English recording.
Is ChatGPT accurate enough for interview transcripts?
It can be a useful first draft, but it doesn't keep the text linked to the audio, and speaker labels on uploads are unreliable. For quotes you'll publish or research data you'll code, use a transcription tool with timestamps and an editor, then check the passages you rely on.
Turn your recordings into text you can use
Upload audio or video, paste a YouTube link, or record in the browser — and get a speaker-labeled transcript in minutes, in 120+ languages.
Try BlaBlaScribe for freeNo credit card required · Stored in the EU · Never used for AI training





