Can ChatGPT Transcribe Audio?
Short answer: sometimes, for short clips, and not in a form you can rely on for a long recording. ChatGPT is very good at working with a transcript and much less dependable at making one. Here is what it does, where it stops, and how to get the text first.
What ChatGPT does with sound
ChatGPT handles speech in several separate ways, and they are easy to confuse. In voice mode you talk and it answers; that is a conversation, and it is not meant to hand you a transcript of a file. The microphone button in the message box is dictation: it turns what you say right now into the text of your message. Some plans also include a record feature in the desktop app that captures a meeting as it happens and writes notes from it.
An existing file is a different matter. Whether ChatGPT accepts an MP3 or an MP4 as an attachment, and what it does with it, depends on your plan, the model you have selected and the app you use. In some setups the file is refused. In others ChatGPT tries to process it with its data analysis tool, which is built for spreadsheets and code, not for speech recognition, and you get an apology or a rough guess. OpenAI changes these capabilities and their limits often, so anything you read about them, this page included, should be checked against OpenAI's own help pages.
Where it falls short for real recordings
Even when an upload goes through, three things get in the way of a one-hour meeting or lecture. The first is size: attachments have a size limit, and a long recording or a video file reaches it quickly. The second is length of text: an hour of speech is roughly nine thousand words, and a model asked to reproduce that much tends to shorten, paraphrase or stop partway, which is the opposite of what a transcript is for.
The third is structure. A working transcript has a timestamp on each segment so you can go back and listen, and you often need it as a subtitle file. A chat answer is a block of prose. It is also tied to one conversation: next week, in a new chat, the recording is gone and you upload it again.
A YouTube link has the same problem from another direction. ChatGPT can read a web page, but it does not download the video and listen to it. When it summarises a video from a link, it relies on text it can find about that video, which may or may not include captions.
What about the OpenAI API?
OpenAI does offer speech-to-text models to developers through its API, and they are good. That is a programming interface, not a feature of the ChatGPT app: you need an API key, code or a command-line tool, and you pay per minute of audio. The API also has a per-file size limit, so long recordings must be split before they are sent and stitched together afterwards. For a developer that is an afternoon of work. For someone who wants the text of a lecture, it is the wrong tool.
ChatGPT alone and with a transcription tool
The ChatGPT column describes typical behaviour. It varies by plan and changes over time.
Two ways to do it properly
Let a speech model make the transcript, and let ChatGPT do what it is good at with the text.
Upload the file, then bring the text to ChatGPT
Drop the recording into the box on the Transcrify home page. Audio and video files are accepted as they are, long recordings included, and the language is detected automatically. The transcript opens in your browser with a timestamp on every segment.
From there you can copy the text into ChatGPT and ask for whatever you need: a summary, an email, a translation. This is the quickest route for a single file.
Transcribe a file freeConnect Transcrify to ChatGPT
Transcrify has an MCP server that ChatGPT can add as a connector. You sign in once, and after that ChatGPT can search every transcript in your account, read a recording by page or by time range, and send a YouTube or public file link to Transcrify for transcription.
This is the better route when you have more than a handful of recordings, because nothing is pasted by hand and the transcripts are available in every new conversation. One limit to know: the connector does not take a file you attach to the chat, so recordings from your computer are still uploaded on the Transcrify site.
Set up the ChatGPT connectorUse Claude or Cursor instead? The same server works there: Transcrify for Claude, Transcrify for Cursor, or the full MCP server reference.
Common questions
Can ChatGPT transcribe an audio file?
Sometimes. It depends on your ChatGPT plan, the selected model and the app. Short clips may work; long recordings are often refused, shortened or paraphrased, and the answer has no reliable timestamps. OpenAI changes these limits, so check its help pages for the current state. A dedicated transcription tool is the dependable route.
Can ChatGPT transcribe a YouTube video from a link?
Not by listening to it. ChatGPT does not download and process the audio of a video; it works from text it can find. With the Transcrify connector, ChatGPT sends the link to Transcrify, which fetches the audio and transcribes it, and ChatGPT then reads the result.
Is voice mode the same as transcription?
No. Voice mode is a spoken conversation with ChatGPT, and dictation turns your live speech into a message. Neither is designed to produce a transcript of a recording you already have.
Is the Transcrify route free?
You can transcribe recordings free on the Transcrify home page without an account, within the free allowance shown there, and a free account can connect to ChatGPT. Subtitle export, summaries and translation are part of the paid plans.
Does the connector let me attach audio in ChatGPT?
No. The connector cannot receive a file attached to a ChatGPT conversation. Upload the file on transcrify.com, or give ChatGPT a YouTube or public file link. Once the recording is transcribed, ChatGPT can search and read it.
Get the transcript first, then ask ChatGPT
Upload a recording or paste a YouTube link on the home page. No account is needed to try it.