How to Upload Audio to ChatGPT: Transcripts, Meeting Minutes, Pricing, and Limits

You can attach a recording to ChatGPT and ask for a transcript, summary, or draft meeting minutes. This feature is available on supported paid plans. It works differently from Voice, which lets you talk to ChatGPT, and Record, which captures audio as you work. To use an existing MP3, M4A, or similar recording, start by attaching the file.

We checked the full text of OpenAI's official English help article on October 7, 2026. This guide explains the specifications and practical workflow. We have not measured transcription accuracy or processing time using actual audio files. Source: Audio uploads, supported formats, and limits.

1. Eligible plans and the four audio features

OpenAI's official help says audio uploads are available on paid ChatGPT subscriptions and workspaces, including Enterprise. Free is excluded. Being able to attach documents or images for free does not mean audio files are available on the same terms. Availability also depends on your region, workspace settings, app or web version, and selected model.

Sources: Voice, Dictation, Record. For more about the models and technology behind voice conversations, see our GPT-Live guide.

Uploading a meeting recording is also different from having a tool join a meeting automatically. To compare recording workflows and dedicated services, see our guide to AI meeting minutes and automation.

2. Supported formats, file size, and long recordings

Source: OpenAI's audio upload specifications. The file must contain readable audio data. Changing its extension does not change the underlying format or remove any video.

If a long file fails, you can try splitting it locally into shorter sections and attaching them in order. Splitting does not guarantee success. Note each section's order, starting position in the original recording, and any small overlap to make gaps and duplicates easier to spot later. Before using an external service to split files, check whether you may send it recordings containing confidential information.

3. How to attach audio and request a transcript

1. Check the file. Play the beginning, middle, and end in a local player to confirm that the audio is present. Note the format, size, and duration.
2. Open a conversation using an account on a supported paid plan. Use the attachment control in the message field to select a supported audio file. Button labels and positions vary by app version; do not confuse this with the dictation microphone.
3. Send a prompt with the attachment. Start by asking for a transcript. Asking only for a short summary at first removes details you need to compare the result with the recording.
4. Review the result and follow up in the same conversation. Correct errors before turning the transcript into meeting minutes or an email draft. A successful upload does not mean the entire recording was analyzed. Check which parts the response covers.

The official basic workflow is: attach a file → explain what you want → review the response and follow up. The following prompt was written by our editorial team to match that workflow. Prompt instructions alone cannot guarantee an error-free result.

Transcribe the attached audio in its original language.
Do not summarize or translate in the first response; preserve the content and order of the statements.
Mark anything you cannot hear as [inaudible], and do not fill it in by guessing.
Add speaker labels only when you can reliably distinguish the speakers;
otherwise use [unknown speaker].
Identify any sections you could not analyze, and say if you cannot confirm that the entire recording was processed.
Recording language: English
Recording duration: (fill in if known)

Providing the duration helps you check how much of the recording the output covers. It does not guarantee that a recording of that length can be processed. If you supply the spelling of technical terms, say to use them only if they are actually spoken, so they are not used to invent missing statements.

4. Prompts for meetings, interviews, and lectures

Once you have checked the transcript, tailor the output to your purpose. The key is to separate facts from the recording from suggestions made by AI. The following examples illustrate different workflows; they are not actual generated results.

Meetings: separate decisions, proposals, and open issues

Create meeting minutes from the transcript we have just checked.
Include: meeting purpose / confirmed decisions / unresolved issues / next actions.
Put next actions in a table with action, owner, deadline, and supporting statement.
Do not turn proposals into decisions.
If no owner or deadline is stated, write "not decided."
Place any suggestions from AI under a separate heading from facts in the recording.

Interviews: separate quotations from edited summaries

Organize the checked transcript by topic.
Clearly separate direct quotations from summaries.
Do not rewrite quotations to make them read more smoothly.
Use speakers' names only when they can be established from the recording or information I have verified.
Also list proper names, numbers, and unclear passages that should be checked by listening again before publication.

Check any passages you plan to quote against the recording. Editing can improve readability, but changing qualifications, negations, or how a statement ends can change its meaning. A label such as "Speaker A" does not by itself establish who is speaking.

Lectures: separate study notes from added explanations

Create study notes from the checked transcript.
Separate the lecture content, terminology, and questions to test understanding.
Label any explanations not present in the recording as "Additional explanation."
Flag unclear formulas, names, and technical terms for checking rather than stating them as certain.
Start with notes in the original language; do not translate yet.

If you want a translation, first correct the original transcript. Translating recognition errors can make the passages needing correction harder to spot. For subtitle files or video editing, see our guide to transcription and subtitle tools. You can ask ChatGPT for SRT output, but that does not guarantee accurate timecodes.

5. Check speakers, numbers, and missing passages

The official guidance also warns about transcription errors and uncertainty in identifying speakers. Do not stop at asking AI to check for mistakes: compare the output with the original audio. Even if the summary reads smoothly, check the following against the recording.

Checking the beginning, middle, and end is a quick way to find large omissions. It does not establish that the entire transcript is accurate. For important matters such as contracts and delivery dates, listen again to each relevant statement. Ask the people involved to check the contents before using them as an official record.

If unclear passages cluster around noise or overlapping speech, listen to those sections locally first. Repeatedly asking AI to rewrite its answer may improve the prose while leaving errors intact. We have not measured Japanese transcription accuracy or compared it with specific services in this article, so we do not rank them.

6. Pricing, usage limits, and the API

Attaching audio to a ChatGPT conversation and calling the developer API have separate billing and limits. Attaching a file does not require an API key. Integrating transcription into your own app uses the API separately from your ChatGPT subscription.

Sources: Separate ChatGPT and API billing, Transcription API file size limit. Do not apply the API's 25MB limit to ChatGPT attachments or assume that a Pro subscription includes API charges.

The document and image upload count table in the same help article explicitly covers documents, spreadsheets, presentations, text, and images. Those figures cannot be treated as audio upload counts. If you encounter a usage limit or error, check the guidance for your subscription and workspace. Voice call duration limits are not a measure of how much attached audio can be processed either.

7. Troubleshoot failed uploads and unreadable files

Being unable to select a file, being told an attached file cannot be analyzed, and seeing processing stop partway through require different checks. Isolate the symptom before changing your subscription. The following sequence is based on the official conditions; it does not guarantee that you will identify the cause.

Source: Official upload troubleshooting. If the problem persists, record the date, time and time zone, device and browser, selected model, format, size and duration, error screenshot, and request ID if available for your support request. Avoid including unnecessary recording content or personal information in screenshots.

For a short test, use audio you prepared yourself that contains no confidential information. Determine whether all files fail or only particular formats or lengths. A failed upload does not necessarily call for a plan upgrade; environment conditions and file problems are other possibilities.

8. Training use, storage, and deletion

Before uploading a recording, explain the intended use to participants, obtain any required consent, and check your employer's or organization's information handling rules. Audio can contain names, addresses, client information, and background conversations that you did not mention in the prompt. Decide first whether you can remove unnecessary sections and whether you may send the material to a cloud service.

Under the general official policy, content from consumer ChatGPT services may be used for training. Turning off "Improve the model for everyone" excludes new conversations, but feedback you voluntarily submit has exceptions. Business, Enterprise, Edu, and the API do not use inputs or outputs for training by default. Source: Training policies by service.

However, the audio upload help we checked does not explicitly explain how training use of attached audio itself relates to Voice's "Include your audio recordings" setting. It is important to avoid applying explanations about Voice clips or Record recordings directly to attached MP3 or M4A files. For interpreting the settings and assessing information already submitted, see our guide to ChatGPT and Codex training use and confidential information.

If a file is saved in Library, deleting the conversation does not remove the Library copy. Transcripts and summaries may also remain in a conversation, so check each storage location. Archiving is not deletion. Source: Chat and file retention and deletion. Also distinguish opting out of training from preventing recordings from being sent off your device: the former does not do the latter.

Summary: upload, check the transcript, then use it

To use an existing recording in ChatGPT, attach the audio file to a conversation and ask for a transcript first. A practical sequence is to verify decisions and quotations against the audio before organizing the result into meeting minutes, an article, or study notes.

• Eligibility: Audio uploads are for paid subscriptions. Do not assume they are available on Free or in every environment.
• Source file: Check the supported audio formats and 512MB limit. Distinguish MP4 and WebM files containing video.
• Quality: Check speakers, numbers, and negations against the recording. Also check that the transcript reaches the end.
• Cost and data: The API is billed separately from ChatGPT. Some audio attachment pricing and usage details are unpublished, and conversations and files may be stored separately.

FAQ