Blog••Tutorials

Can ChatGPT Transcribe Audio?

Sarah Johnson

Sarah Johnson

Writes about field sales, meeting notes and voice-first workflows at ParrotNotes. Every article is reviewed by the ParrotNotes product team before it goes live.

Can ChatGPT Transcribe Audio?

Priya, a real estate agent, recorded a 40-minute consultation with a couple who are buying their first home (both agreed to the recording). Back in her car, she opened ChatGPT on her iPhone, tapped the attach button and picked the file. She wanted the text before her next showing. (Priya is an illustration, not a real customer.)

Whether that works depends on one thing she didn't think to check: her plan.

So, can ChatGPT transcribe audio? Yes, on a paid plan. You can upload an audio file (up to 512 MB) and ask for a transcript, or record a meeting live with Record on the ChatGPT Mac app. On the Free plan, you can't upload audio at all as of November 2026. Free users can only dictate their own voice into a prompt.

You've probably seen people say ChatGPT "can't hear files." That was true for a long time, and it's out of date. This guide covers the six ways ChatGPT and OpenAI turn speech into text, which one works on your iPhone, Android, Mac or PC, how long a file each limit really allows, and when a phone transcription app is the simpler choice. Every limit here comes from OpenAI's own help pages, checked on November 8, 2026.

If you mostly record conversations on your phone and want the transcript and a summary without moving files around, ParrotNotes does that on iPhone and Android, with 100 free minutes a month.

Can ChatGPT transcribe audio? The short answer

ChatGPT has more than one way to turn speech into text, and they do different jobs. Here they are as of November 2026:

RouteWhat it doesWho can use it
Upload an audio fileTranscribes a recording you attach, then lets you ask about itPaid plans and workspaces only
RecordRecords and live-transcribes a meeting or voice note, then writes a summaryPlus, Pro, Business, Enterprise and Edu, Mac app only
Meetings pluginTakes meeting notes from your mic and system audio, no botPro and Business, Mac app only (beta)
Dictation (microphone icon)Turns what you say into editable text for your next messageAll users, any device
VoiceA spoken conversation with ChatGPT; a transcript is saved to the chatAll users, with tighter limits on Free
OpenAI APIDeveloper access to the gpt-transcribe model, billed per minuteAnyone with an API account

The first two are what most people mean by "transcribe audio." Dictation and Voice turn your speech into text, but they don't transcribe a recording you already have.

Which ChatGPT route works on your iPhone, Android, Mac or PC

This is where most guides go vague. OpenAI's pages are specific, so here is what they say, device by device and plan by plan:

RouteiPhone and Android appMac (ChatGPT desktop app)Windows app or browserFree plan
Upload an audio fileAttach button is there; OpenAI says audio availability "may vary by client version"Paid plansPaid plans, at chatgpt.com in any browserNo
RecordNoYes: Plus, Pro, Business, Enterprise, EduNo ("coming soon" for Windows)No
Meetings pluginNo ("coming soon")Yes: Pro and Business, betaNo ("coming soon")No
DictationYesYesYesYes
VoiceYesYes (desktop has its own limits)YesLimited
APIThrough your own code or a third-party appSameSameSeparate pay-as-you-go billing

Sources: OpenAI's Uploading files and audio to ChatGPT, ChatGPT Record, Meetings plugin, ChatGPT Voice and Voice Dictation help pages, all checked November 8, 2026.

Two things stand out. First, there's no way to record a meeting for transcription inside the ChatGPT phone app today; on a phone, you record somewhere else and upload the file. Second, the cheapest paid plan is Go at $8 a month, then Plus at $20 and Pro at $100 (ChatGPT pricing). OpenAI says audio uploads come with "paid ChatGPT subscriptions" but doesn't list them plan by plan, so check the attach menu works before you pay for a plan only for this.

How to transcribe audio files in ChatGPT, step by step

On a paid plan, the upload route takes about a minute to start:

  1. Open a new chat at chatgpt.com or in the ChatGPT app.
  2. Select + in the message box, then Add photos & files, and pick your recording.
  3. Check the format. ChatGPT accepts WAV, MP3, M4A, AAC, FLAC, OGG, PCM, and audio-only WebM or MP4. iPhone Voice Memos saves M4A, which works.
  4. Say exactly what you want. "Transcribe this" can come back as a tidy summary. Ask for the words instead (prompts below).
  5. Read it against the audio for anything that matters: names, numbers, dates and quotes.

A video won't work as is. OpenAI's page says "WebM and MP4 files identified as video are not supported by audio uploads." Pull the sound out first with a free video to audio converter, then upload the audio. If a file uploads but won't transcribe, OpenAI suggests converting it to WAV and trying again.

Prompts that get a transcript, not a summary

Copy these and change the details in brackets:

  • Word for word: "Transcribe this recording verbatim. Don't summarize, shorten or correct anything. Start a new paragraph when the speaker changes."
  • With speakers: "Transcribe this conversation between [Priya] and [two clients]. Label each line with who is speaking. If you can't tell, write Speaker 1, Speaker 2."
  • Clean read: "Now remove filler words (um, uh, you know) and fix punctuation. Don't change the meaning or add anything."
  • Check yourself: "List every name, number and date in the transcript, with roughly where it appears, so I can check them against the audio."

Then ask your follow-ups in the same chat: "What did they say about their budget?" or "Write a follow-up email from this."

How big a file can ChatGPT transcribe?

OpenAI gives the limit in megabytes, not minutes. Uploads in ChatGPT can be up to 512 MB. The developer API takes up to 25 MB per file. How much talking fits depends on the format, so here's the arithmetic (1 MB = 1,000,000 bytes):

Format and qualitySize per minuteFits in 25 MB (API)Fits in 512 MB (ChatGPT upload)
WAV, CD quality stereo (1,411 kbps)10.6 MBabout 2 minutesabout 48 minutes
WAV, mono speech at 16 kHz (256 kbps)1.9 MBabout 13 minutesabout 4.4 hours
MP3 at 128 kbps0.96 MBabout 26 minutesabout 8.9 hours
M4A (AAC) at 64 kbps0.48 MBabout 52 minutesabout 17.8 hours
MP3 at 32 kbps, mono0.24 MBabout 1.7 hoursabout 35 hours

These are worked numbers from the bitrate, not tests. Your file's real size depends on the app that made it, so check the file size before you upload.

The 512 MB figure is a size cap, not a promise. OpenAI says long recordings "may be processed in smaller sections," that processing is "best-effort," and that "very long recordings may time out." In practice, a two-hour WAV is a bigger risk than a two-hour M4A, even though both fit. If a file is too big, a free audio compressor can shrink it to a lower bitrate first.

Record mode and Meetings on a Mac

If you use a Mac, ChatGPT can record for you instead of waiting for a file.

Record lives in the ChatGPT desktop app for macOS. Click the Record button at the bottom of a chat, allow microphone (and, for calls, system audio) access, and talk. ChatGPT transcribes as you go, and a timer shows how long you've been recording.

Click Stop, then Send, and you get a transcript plus a summary saved as a canvas that you can turn into an email or a plan. Sessions stop automatically at 4 hours. OpenAI says Record can tell speakers apart and uses labels like "Speaker 1" when it can't find a name, which you can rename.

Meetings is the newer plugin in the Mac app, in beta for Pro and Business plans. You install it from Plugins, click Take notes during an online call or an in-person meeting, and it writes a summary with suggested action items. There's no bot joining the call. OpenAI says iOS, Android and Windows support is "coming soon," and that the consent reminder only shows on your screen: it doesn't tell anyone else in the meeting.

Neither one runs on an iPhone or Android phone today.

Voice mode and dictation are not file transcription

People often try these two first, and both disappoint for a recording you already have.

Dictation is the microphone icon in the message box. You speak, ChatGPT turns your speech into text, and you can edit it before sending. It's handy for a long prompt on your phone. It doesn't take a file, and it isn't built for a meeting with other people in the room.

Voice is a back-and-forth spoken conversation with ChatGPT. A transcript of the conversation is added to the chat afterwards, but OpenAI says plainly that "Voice transcripts are not verbatim records and may not exactly match what was said." Holding your phone up to a speaker and pressing Voice isn't a reliable way to transcribe anything.

Using OpenAI's transcription API

Developers can skip the chat window and send audio to OpenAI's transcription API directly. This is where "ChatGPT Whisper" searches usually lead, and it's changed.

OpenAI now recommends a model called gpt-transcribe. The older models, including whisper-1, gpt-4o-transcribe and the speaker-labelling gpt-4o-transcribe-diarize, are deprecated and shut down on February 26, 2027, according to OpenAI's speech-to-text guide. Files can be up to 25 MB, in mp3, mp4, mpeg, mpga, m4a, wav or webm.

The price is low. OpenAI's API pricing page lists gpt-transcribe at $0.0045 a minute as of November 2026.

Here's a worked example. Dev, a freelance podcast editor, has 30 hours of old interviews to turn into text. That's 1,800 minutes × $0.0045 = $8.10 through the API, against $20 for a month of ChatGPT Plus. The catch is the 25 MB cap: at 128 kbps MP3, each file has to be cut into roughly 26-minute pieces first, and someone has to write the script that sends them. (Dev is an illustration, not a real customer.)

So the API suits people who can code, or apps built on top of it. If that's not you, the chat upload or a transcription app is simpler.

Can ChatGPT transcribe audio accurately? The limits to know

ChatGPT can transcribe audio to text well enough for many jobs, but go in knowing where it falls short.

  • Accuracy isn't guaranteed. OpenAI's Record page says "ChatGPT may make mistakes, including in its transcriptions." Check names, numbers and quotes against the recording before you send or publish anything.
  • It can summarize when you asked for a transcript. A chat model likes to tidy up. Use the verbatim prompt above and spot-check a minute of the audio.
  • Speaker labels vary. Record mode labels speakers. For an uploaded file, OpenAI's page doesn't promise speaker labels, so ask for them and check.
  • Languages. OpenAI says Record "works best in English today," and that upload transcription "may vary across languages."
  • Long files. Big uploads can time out, and Record stops at 4 hours.
  • Privacy. On Free, Plus and Pro, transcripts and canvases may be used to train models if Improve the model for everyone is on in Settings > Data controls. OpenAI says Record deletes the audio after transcription. Business, Enterprise and Edu content is excluded from training by default.
  • Consent. Recording other people has legal rules that change by state and country. OpenAI's own Record page tells you to get the right consents first. Our guide to one-party consent states explains the US rules in general terms; it isn't legal advice, and you should never record anyone secretly.

When a phone transcription app is the better tool

ChatGPT is a good choice when you already pay for it, the file is on your computer, and you want to ask questions about it straight away. It's a weaker fit in three common cases.

You're on the Free plan. Free users can't upload audio. The best free transcription apps compares free options by job, and the free audio to text converter handles a short file (up to 10 minutes, with a free account).

You record on your phone. On an iPhone or Android phone, ChatGPT can't record a meeting for you. You'd record in another app, find the file and upload it. The built-in phone apps can often do the job themselves: see how to transcribe a voice memo on iPhone and Android.

You record the same kind of conversation every week. Marcus, an outside sales rep, visits six stores a day. On Mondays he used to record a store visit, upload it to ChatGPT, paste the transcript into a summary prompt, then copy the result into his CRM, every time. (Marcus is an illustration, not a real customer.) A recording app that transcribes and summarizes on its own cuts out the middle steps.

That's the job ParrotNotes is built for. It records in-person conversations and meetings on iPhone and Android, keeps going with the screen locked, and gives you a transcript and an AI summary when you stop.

The Free plan includes 100 minutes of recording a month (up to 30 minutes per recording), an AI summary on every recording and 3 audio or video imports a month. Pro ($19.99 a month, or $14.99 a month billed annually) raises that to 3,000 minutes and up to 3 hours per recording, and adds speaker identification. Pro also has a connector for ChatGPT, so you can still ask ChatGPT about your notes.

Prefer Word or Docs? Microsoft 365 subscribers can also transcribe audio in Word.

The bottom line on ChatGPT transcription

Can ChatGPT transcribe audio? Yes, with conditions:

  1. Paid plan: upload a file up to 512 MB on a paid plan, and ask for a verbatim transcript.
  2. Mac: use Record (Plus and up) or the Meetings plugin (Pro and Business) to record and transcribe live.
  3. Free plan: dictation only, no audio files.
  4. Developers: gpt-transcribe at $0.0045 a minute, 25 MB per file; Whisper shuts down in February 2027.
  5. Always: check names and numbers against the audio, and get consent before you record anyone.

If your recordings start on your phone, try the free plan of ParrotNotes on your next meeting and compare the transcript with what ChatGPT gives you.

Frequently Asked Questions

Can ChatGPT transcribe audio for free?

Not from a file. As of November 2026, OpenAI says audio uploads are not available on the Free plan. Free users can use dictation (the microphone icon) to turn their own speech into a prompt, and Voice with limits, but neither transcribes an existing recording. For a free file transcript, use a free transcription app or web converter instead.

Can ChatGPT transcribe an MP3 or a voice memo?

Yes, on a paid plan. ChatGPT accepts MP3, M4A (the iPhone Voice Memos format), WAV, AAC, FLAC, OGG, PCM and audio-only WebM or MP4 files up to 512 MB. Attach the file with + and Add photos & files, then ask for a verbatim transcript.

Can ChatGPT transcribe audio from a video?

Not directly. OpenAI's help page says WebM and MP4 files identified as video are not supported by audio uploads. Extract the audio track first with a video to audio converter, then upload the MP3 or WAV.

Is ChatGPT transcription accurate?

It's usable for clear recordings, but OpenAI warns that ChatGPT may make mistakes, including in its transcriptions, and that results vary by language. Check names, numbers, dates and quotes against the audio before you rely on them.

Does ChatGPT label different speakers?

Record mode on the Mac app does: it separates speakers and uses labels like Speaker 1 when it can't find a name, and you can rename them. For uploaded files, OpenAI doesn't promise speaker labels, so ask for them in your prompt and check the result.

Is ChatGPT's Record mode on iPhone or Android?

No. As of November 2026, OpenAI says Record is only available in the ChatGPT desktop app for macOS, on Plus, Pro, Business, Enterprise and Edu plans. The Meetings plugin is also Mac only, with iOS, Android and Windows listed as coming soon.

Does ChatGPT still use Whisper for transcription?

OpenAI's API now recommends the gpt-transcribe model. The whisper-1 model and the older gpt-4o transcription models are deprecated and shut down on February 26, 2027.