Can ChatGPT Transcribe Audio to Text? And What About Gemini?

Yes, ChatGPT can transcribe audio to text, but only on paid plans and only from audio files: OpenAI's help centre says audio uploads are not available on the Free plan, and video files and YouTube links are not supported as audio sources. Google's Gemini can transcribe uploaded audio too, with a 10-minute total limit on free accounts and up to 3 hours on paid Google AI plans. For long recordings, interviews or anything you need with timestamps and speaker labels, a dedicated transcription tool is still the more dependable first step, and the chat assistant is better used afterwards, for summaries.

This guide covers what each assistant can actually do with audio, video and YouTube links as of October 2026, where the limits are, and a workflow that combines the strengths of both. AI products change quickly, so the help-centre pages linked below are the place to check current details.

Can ChatGPT transcribe audio to text?

There are three different ways ChatGPT handles speech, and they are often confused.

1. Uploading an audio file

OpenAI's article on uploading files and audio to ChatGPT says you can attach an audio recording to a conversation and ask ChatGPT to transcribe it, summarise it or answer questions about it. The key facts from that page, checked in October 2026:

• Plans: audio uploads are available with paid ChatGPT subscriptions and workspaces, including Enterprise. They are not available on the Free plan.
• Formats: WAV, MP3, OGG, FLAC, AAC, M4A, PCM, and audio-only WebM or MP4. The file must contain a valid audio stream.
• Size: audio files can be up to 512 MB.
• Length: there is no fixed duration limit, but OpenAI says longer recordings may be processed in sections, processing is best-effort, and very long recordings may time out.
• Quality: OpenAI states that transcripts may contain errors, speaker identification may be unreliable, and performance varies across languages.

2. ChatGPT Record

ChatGPT Record records a meeting, brainstorm or voice note live, transcribes it and generates notes as a canvas. According to OpenAI's help page, it is available on Plus, Pro, Business, Enterprise and Edu, and only in the macOS desktop app. Sessions are capped at 4 hours, the audio is deleted after transcription, and Record works best in English. It can label multiple speakers ("Speaker 1") and lets you rename them. Record is for capturing something happening now; it does not transcribe a file you already have.

3. Dictation

The microphone icon in the message box is voice dictation: your speech is transcribed into a message you can edit before sending. It is handy for talking instead of typing, but it is not a way to transcribe an existing recording, and you cannot play a file into it in any reliable way.

Can ChatGPT convert video to text?

Not directly. OpenAI's upload guide lists no video formats, and it says that WebM and MP4 files identified as video are not accepted as audio uploads. If you want text from an MP4 or MOV file, you have two options:

1. Extract the audio first with a converter or video editor, save it as MP3 or M4A, and upload that to ChatGPT (on a paid plan).
2. Use a transcription tool that accepts video, such as Audiotexta's video to text tool, which takes the video file as it is and returns a transcript with timestamps.

The second route also gives you subtitle files (SRT, VTT), which ChatGPT does not produce as a standard output.

Can ChatGPT transcribe a YouTube video?

Generally, no, and this is the question people ask most often. When you paste a YouTube link, ChatGPT cannot play the video or download its audio. With web search switched on it may be able to read text on the page, such as the title or description, and it may then summarise what it thinks the video is about. That summary can sound confident while being based on very little, so treat it with caution. At the time of writing, OpenAI's help centre has no article describing YouTube link support.

What works reliably is to get the transcript first and give ChatGPT the text:

• Copy it from YouTube's own transcript panel if the video has captions (see how to see the transcript on YouTube).
• Or paste the link into a tool that transcribes the audio, which also works on videos without captions. Our guide to turning a YouTube video into text compares the options.

Can Google (Gemini) transcribe audio to text?

Yes. Google's help page on uploading and analysing files in Gemini Apps covers audio, video and documents. As of October 2026 it says:

• You must be signed in. Up to 10 files can be added to one prompt.
• Audio: total audio length up to 10 minutes, or up to 3 hours with Google AI Pro or Google AI Ultra.
• Video: each video can be up to 2 GB, with a total length of up to 5 minutes, or 1 hour on the paid plans.
• Other files can be up to 100 MB each.

So Gemini, unlike ChatGPT, accepts video files, and it also accepts audio on a free account, within that 10-minute total. Google's documentation does not promise a verbatim transcript format, timestamps or speaker labels, so ask for them explicitly in your prompt and check the result.

Gemini also has an advantage with YouTube: Google's article on YouTube content in Gemini Apps describes asking Gemini to find a video and summarise it, using public information from YouTube. That is useful for a summary, but it is not the same as a full, checkable transcript.

For Google Workspace users, Gemini in Google Drive can summarise and answer questions about videos stored in Drive on eligible Workspace and Google AI plans, according to Google's Workspace announcements. Google Docs itself offers voice typing for live dictation, not transcription of uploaded files. Plans and availability vary by edition and language, so check with your administrator.

ChatGPT vs Gemini vs a dedicated transcription tool

Limits checked on the vendors' help pages, October 2026.

The assistants are clearly better at conversation: asking follow-up questions, rewriting a recap as an email, explaining a passage. The dedicated tools are better at the transcript itself.

Practical limits to know before you upload

Length and size

Long recordings are where general chat assistants struggle most. ChatGPT may split long audio into sections and can time out; Gemini caps the total length per prompt. A one-hour lecture fits on paid plans of both, but a full-day workshop does not.

Missing timestamps and speakers

A transcript you want to quote, subtitle or analyse needs timestamps you can click and speaker labels you can trust. Chat assistants are not built around these. If you ask for timestamps anyway, check a few against the recording before relying on them.

Gaps and invented text

Every speech-to-text system can mishear words. A specific risk with AI models is text that sounds plausible but was never said, especially in silent stretches, music or crosstalk, and when a chat model "tidies up" a long transcript it may shorten or paraphrase instead of writing it out word for word. Ask for a verbatim transcript, and spot-check passages that matter. Clean audio helps every system: one speaker per microphone, little background noise, and no music under speech.

Privacy and training settings

What happens to your recording depends on the product and your settings:

• ChatGPT: on consumer plans, OpenAI may use content to improve its models if "Improve the model for everyone" is enabled; you can switch it off in Data Controls, and there is a separate toggle for audio recordings. OpenAI says it does not train on business offerings such as ChatGPT Enterprise by default.
• Gemini: with Keep Activity on, Google's privacy hub says some conversations may be reviewed by people to improve its services; you can turn Keep Activity off, but several Gemini features, including the YouTube integration and Drive uploads, require it.

For interviews, patient conversations or client calls, check these settings first, get consent from everyone recorded, and read our guide to GDPR and transcription if you work in the EU.

The workflow that works: transcribe first, then ask the assistant

Chat assistants are excellent at working with text and less dependable as transcription engines. Splitting the job plays to both strengths:

1. Transcribe with a speech-to-text tool. Use Audiotexta in the web app (upload a file, paste a link or record from the microphone), or run Whisper locally if you prefer to keep everything on your machine. Our comparison of Whisper and cloud speech APIs explains the trade-offs.
2. Review the transcript. Fix names and jargon, rename speakers, and check any passage you will quote against the audio.
3. Export it as text. TXT (free) or Markdown (Pro) is easiest to paste into a chat.
4. Paste it into ChatGPT or Gemini and ask for what you need: "Summarise this interview in five bullet points", "List every decision and who owns it", "Draft a follow-up email". Because the assistant now works from complete text, you can check every claim in its answer against a timestamp.

Audiotexta can also produce the summary step itself on the Pro plan (summary, chapters, key points, decisions and action items), with no use of your content for model training. Use whichever combination fits; the important part is starting from a full, verifiable transcript.