Can ChatGPT transcribe videos? Yes, with limits. You can attach a video, including on the Free plan where your app supports it, but OpenAI warns that it may not analyze the whole video and can misinterpret the audio. The more direct route, uploading an audio-only file for transcription, needs a paid plan. Either way, check the words against the recording before you publish, and check the timing too if you need subtitles. This guide covers each route (video attachments, audio uploads, YouTube transcripts, ChatGPT Record, Whisper and transcription editors) and when each one fits.
Based on OpenAI’s help and developer pages as of November 12, 2026. We did not run a transcription accuracy test.
Can ChatGPT transcribe videos directly?
Yes, if your account and upload method support video attachments. But a successful upload doesn’t mean every spoken word was transcribed: OpenAI cautions that video analysis can skip parts of the video and misunderstand the audio. OpenAI’s video attachment guidance.
Pick the route from what you have:
| What you have | Start here | Check before using it |
|---|---|---|
| A saved video | Supported video attachment | Full speech coverage |
| An audio-only copy | Paid ChatGPT audio upload | Wording and omissions |
| A captioned YouTube video | Existing transcript text | Match to the recording |
| A meeting starting now | ChatGPT Record | Availability and speakers |
| Media for local processing | Open-source Whisper | Setup and output |
| A video needing subtitles | A transcription editor | Words and cue timing |
A saved MP4 video
Route: attach the video, if your ChatGPT app and account offer video uploads; availability varies by platform, upload method and account. Check: ask for a transcript, not a summary, and compare it with the source. A description of the scene doesn’t tell you the speech was fully transcribed.
An MP4 with video in it is not the same as an audio-only MP4. OpenAI accepts audio-only MP4 for audio uploads but rejects files it identifies as video on that route. See the audio format list.
An extracted audio file
Route: if video uploads aren’t available or the transcript comes back incomplete, make an audio-only copy with a tool you already have. Paid accounts can upload formats including MP3, M4A and WAV. Audio upload documentation. Check: ask for a transcription and review it against the recording.
An existing YouTube transcript
Route: for a YouTube video with captions, click Show transcript in the video description, then paste that text into ChatGPT for cleanup or questions. Check: compare key passages with the video. Pasting only the link doesn’t mean ChatGPT heard the soundtrack.
Our YouTube transcript guide covers finding and copying the existing text.
A new meeting recording
Route: ChatGPT Record records and transcribes a new session; it doesn’t transcribe a saved video. Classic Record is available in the macOS app on Plus, Pro, Business, Enterprise and Edu. The same page describes a separate Meetings beta with its own availability. Check: review the transcript and the speaker labels.
A local Whisper workflow
Route: the open-source Whisper package runs speech recognition on your own computer from the command line, in many languages, and needs ffmpeg installed. It is separate from the hosted whisper-1 API model. Check: review the transcript; running locally doesn’t make it more accurate.
An editable transcript and subtitles
Route: bring the video into a transcription editor such as ChatCut when you need to review the speech against playback and deliver captions. In ChatCut, clicking a transcript word jumps to that moment, and you can export SRT or TXT once transcription finishes. Transcript and captions guide. Check: the source track, the words and the caption timing.
Get a transcript with ChatGPT
Prepare and attach your source
Pick a route from the list above. If you make an audio copy, play it first to confirm it contains all the dialogue, including the ending. Keep the original video for review.
Attach the file and ask for a transcript explicitly; a summary leaves out the detail a transcript needs. Try this request with an attached audio file:
The prompt sets your requirements; you still have to check that ChatGPT met them. If the attachment is rejected, see the video upload guide and the audio upload failures below.
Check coverage before cleaning text
Start with the opening, a passage from the middle and the final spoken sentence. A transcript that ends neatly can still stop before the recording does. Then review the whole recording before publishing, with particular attention to names, figures, speaker changes and overlapping speech.
Keep uncertainty visible. If you cannot hear a word clearly, retain an uncertainty marker rather than accepting a plausible replacement. OpenAI warns that transcripts can contain errors and speaker identification can be unreliable.
Preserve the original during cleanup
Save the first transcript separately before requesting a readable version. Decide whether you want exact wording or an edited article; those are different outputs. For punctuation and paragraph cleanup only, paste the transcript into this request:
Compare the cleaned version with the original. A useful edit should make the text easier to follow without quietly changing the evidence.
Understand ChatGPT’s transcription limits
Free and paid upload access
Free accounts can attach videos, but not audio files: audio uploads are limited to paid subscriptions and workspaces, and availability also depends on workspace settings, region, app version and model. Upload availability.
File size versus recording length
Audio uploads can be up to 512 MB, and very long recordings may time out. OpenAI gives no minute limit for uploaded audio, so a file under the size cap may still not finish. Audio processing limits.
Record sessions have a separate cap of 240 minutes, which doesn’t apply to uploaded files. Record limits.
Requested timestamps versus measured timing
OpenAI’s upload documentation doesn’t promise accurate timestamps. Treat any timecodes you ask for as unverified until you match them to playback: a line marked 00:01:23 may not start there.
The developer Audio API has model-specific timing examples, but those are a separate workflow. Check the current file-transcription guide before choosing a model for timestamped output.
Speaker labels versus speaker identities
OpenAI warns that speaker identification in uploaded files may be unreliable. Record supports multiple speakers and lets you rename them, but neither tells you who an unknown speaker is.
Use neutral labels such as Speaker 1 while you review, and add names only from information you can verify.
Accuracy and missing sections
A fluent transcript can still get a name wrong or skip a passage, and video analysis can miss parts of the source. OpenAI documents these limitations but gives no error rate.
Before you publish, check the complete recording. Spot checks find problems but can’t confirm that every section is there.
Languages and translated output
Transcribing the original speech, translating the text, making timed translated subtitles and dubbing new audio are four separate jobs. OpenAI notes that transcription quality varies by language.
Once you have checked the source transcript, try a text-only translation request:
Have a fluent speaker review the meaning and readability. In ChatCut, translating captions needs an existing caption track and creates a separate translated caption source you can show, style and export; it doesn’t dub the audio. See the caption translation documentation.
Subtitle formatting versus synchronized cues
Text laid out as SRT or VTT isn’t necessarily aligned to the media. Check each cue’s start, end and reading time in playback. OpenAI’s upload documentation doesn’t describe a synchronized subtitle export.
ChatCut exports SRT and TXT once the active caption track finishes transcribing. Use SRT for timed cues and TXT for plain text. Subtitle export.
Upload failures and audio decoding
Check that your account has audio uploads, which model is selected and whether the file has a valid, decodable audio stream. MP4 and WebM files identified as video are rejected on the audio route. If the audio uploads but won’t transcribe, OpenAI suggests converting it to WAV and trying again. Upload troubleshooting.
Record’s microphone and system-audio permissions only affect live recording; they won’t fix a saved file’s format problem. Record setup.
Choose Whisper or a transcription editor
Choose by the output you need and the setup you can manage. ChatCut publishes this guide and is one of the options; we compared documented workflows, not speed or accuracy.
| Route | Useful starting point | Main check |
|---|---|---|
| ChatGPT audio upload | Transcript and follow-up questions | Access and complete wording |
| Open-source Whisper | Local speech recognition | Setup, compute and review |
| OpenAI Audio API | Developer-managed transcription | Current model and limits |
| ChatCut | Transcript review and caption delivery | Source track, words and timing |
For local processing, Whisper’s code and weights are MIT-licensed, so the software is free, but you supply the installation and the compute.
For automation, see the Audio API guide. It recommends gpt-transcribe for new file transcription, lists a 25 MB file limit, and marks several older models, including hosted whisper-1, as deprecated. ChatGPT’s upload rules don’t carry over to the API, so check the limits of the model you choose.
For subtitles, use a workflow where you can listen, correct the text and check timing in one place. Here is how that works in ChatCut.
Transcribe and edit in ChatCut
- Import the video. Open ChatCut Desktop and drag the file into the media library. Keep the original recording for comparison.
- Select the speech source. When transcription finishes, open Transcript. If there are several speech tracks, choose the one you want.
- Review the words. Click a word to jump to it. Use Edit text to correct a transcription error; deleting transcript words cuts that part of the video from the track.
- Check the speakers. Double-click a speaker label to rename it, or use its menu to reassign words or merge a speaker.
- Create readable captions. Show captions and choose a style once the wording is right.
- Export the text. Open Export → Subtitles and choose SubRip (.srt) or plain text (.txt). Export is available once the active caption track finishes transcribing. Open the exported file and check it against the video. The transcript and captions guide documents these controls.
Everything in these six steps is on the Free plan: the core editor includes manual timeline editing, uploads, transcription and standard caption editing, which never use credits, and exporting subtitles doesn’t use credits either. Credits come in only if you ask the built-in ChatCut Agent to do the work or generate new media. See Free and Pro.
