Can ChatGPT transcribe audio files?

Yes, on a paid plan. ChatGPT accepts audio files as attachments, and you ask it in the chat for the transcript. The Free plan does not include audio uploads ¹, so you need a paid subscription or a workspace plan. Audio files can be up to 512 MB ². The answer is a chat reply, not a transcript file, and a chat model may summarize the recording unless you ask for an exact copy.

For a full comparison of the three routes, Whisper, ChatGPT and Gemini, read the guide to transcribing with Whisper, ChatGPT or Gemini. For the same question about Google's model, read can Gemini transcribe audio.

Audio uploads need a paid plan, and the file limit is 512 MB

OpenAI says audio uploads in ChatGPT are available with paid subscriptions and workspaces, including Enterprise ¹. If you are on the Free plan, the attach option for audio is not part of your plan, so a transcript needs a different tool or a paid plan.

The limit is about file size, not length. OpenAI's help article gives 512 MB per audio file ². How many minutes fit depends on the format and bit rate of the file, so a long recording in a compressed format may fit where a shorter one in a larger format does not. Check the size of your file before you attach it.

The API is a different product with a much smaller limit. OpenAI's transcription API accepts files up to 25 MB, and for larger recordings it advises a compressed format or splitting the file ³. The chat limit is about 20.5 times that figure ⁴, but the API is the route to use in code.

Long uploads can time out, so split very long recordings

OpenAI says processing of long audio uploads is best-effort, and that very long recordings may time out ⁵. If a long file fails or stops early, cut it into shorter parts at sentence breaks, attach each part on its own, and join the text in order. Check the join, because a sentence cut in half is easy to miss.

Dictation in the chat box is a separate feature

Many people search for ChatGPT's transcription features and find dictation instead. Dictation is for speaking a prompt. OpenAI's Voice help article says to use ChatGPT Dictation when you want to record a prompt, review and edit its transcription, and then send it as text ⁶. It is a short-form input tool, not a way to turn a recording you already have into a transcript. For a file you already recorded, attach it as an audio upload.

A chat model may summarize, so ask for the speech word for word

A chat model is built to write, which is useful for a summary and risky for a transcript. A study of AI-written clinical notes reported a 1.47 percent hallucination rate, meaning sentences that the source conversation does not support ⁷. Its authors worked for the company that sells the system they tested. That study covered notes rather than transcripts, but it shows that a model asked to write can add content.

Use a prompt that asks for a copy and forbids edits:

Transcribe this audio word for word, in the language spoken. Do not summarize, translate, shorten, reorder or correct grammar. Mark unclear words as [inaudible]. Output only the transcript.

Keep the word-for-word rule for a verbatim transcription. If you want a cleaner text, say so in the prompt and name the filler words to remove.

Whisper is different. OpenAI says Whisper “doesn’t follow instructions like a general-purpose text model,” so a prompt is not a reliable way to control its output ⁸. If exact output matters most, a Whisper transcript is often easier to trust than a chat reply, and you can run it on your own computer. The steps are in how to use Whisper to transcribe audio.

What to do before you rely on a ChatGPT transcript

1. Confirm your plan. Audio uploads are not part of the Free plan.
2. Check the size. Keep the file under 512 MB, and split very long recordings.
3. Ask for a copy. Use the word-for-word prompt above, and ask for one reply per file.
4. Check three passages. Compare a passage near the start, one in the middle and one near the end with the recording.
5. Check names, numbers and the ending. A transcript that stops early or ends with a neat conclusion the speaker never gave is a summary.

If a passage fails, do not ask the chat to fix the summary. Re-run the copy on a shorter part, because a corrected summary is still a summary. For the full checklist, see the section on checking the output in the guide to transcribing with Whisper, ChatGPT or Gemini. If you need a transcript without an account or plan, audio to text transcribes a file in your browser.

Related questions

• How do you use Whisper to transcribe audio?
• Can Gemini transcribe audio?
• Can NotebookLM transcribe audio?
• How to transcribe audio to text, in the transcription overview.