Transcribe Audio & Video in ChatGPT — The Complete 2026 Guide

The 90-second TL;DR

If you're here because you searched for "ChatGPT transcribe audio" and want the answer in two sentences:

Yes, ChatGPT can transcribe. Use the Whipscribe Custom GPT (works on the free plan and every paid plan), or — if you're on Plus or Pro — add Whipscribe as an MCP Connector so it's available in every chat without switching to a specific GPT. Both paths run on the same backend; pick the one that matches how you use ChatGPT.

Why people are asking ChatGPT to transcribe in 2026

The shift over the last 18 months: ChatGPT became the place a lot of people start a task that touches a recording. They have a voicemail, a Zoom export, a podcast file, a lecture, an interview — and they want the next step (notes, action items, summary, post draft) without bouncing between tools.

Until last year, doing this meant uploading the audio to a transcription tool, copying the text, pasting it into ChatGPT, and asking for the artifact. Three apps, two tab switches, one transcript living somewhere outside the chat where you've stored everything else.

The Whipscribe integrations close that loop. Drop the file once. Ask the question. The transcript shows up in the same conversation as everything else you've worked on with ChatGPT, and a copy lives in your Whipscribe library at whipscribe.com/home for later.

Path 1 — The Whipscribe Custom GPT (everyone)

The Custom GPT is the right starting point for most people. Three reasons:

• It works on every ChatGPT plan, including the free tier. No upgrade required.
• It runs on web and the ChatGPT mobile apps, which is how voice memos most often arrive.
• Setup is one OAuth click. After that, you open the GPT and use it.

How it works in practice

Open the Whipscribe GPT, click Start Chat. The first time you ask it to transcribe something, ChatGPT prompts you to authorize. Sign in with your Whipscribe email — same one you use on whipscribe.com — and approve. After that, the GPT can:

• Transcribe a file you drop into the chat. Audio or video, up to a couple of hours per file in practice.
• Transcribe a URL to a file you host (your own podcast feed, your own meeting recording, your own video).
• Search across your previous transcripts by keyword and quote the matching turn back into the chat.
• Save items into Knowledge folders so a project's calls are searchable later.
• Run a saved Recipe or Workflow against a transcript — your own templated post-processing pipeline.

Path 2 — Whipscribe as an MCP Connector (Plus / Pro)

The MCP Connector is the second path. Configured once in Settings → Connectors, it makes Whipscribe tools available in every conversation you have on ChatGPT — not just inside the Whipscribe GPT.

Why people pick this over the Custom GPT:

• You already have a chat going on a topic and don't want to start a new one in a different GPT.
• You want ChatGPT to decide when transcription is the right tool. Ask "what was decided in this call?" with a file attached, and ChatGPT picks Whipscribe automatically.
• You're building a workflow that mixes other Connectors (Drive, Calendar) with audio.

The endpoint is https://whipscribe.com/mcp. Add it as a new MCP server in Settings → Connectors, authorize once, and you're done. The setup post above has screenshots.

What ChatGPT can actually do once a transcript exists

Transcription is the door, not the room. The reason this integration matters is what ChatGPT does on the other side of it. The most useful patterns we see:

Recipes — your own saved post-processors

If you do the same post-transcription task often (action-item extraction, show-notes pass, weekly recap), save it as a Recipe. From the GPT or the MCP Connector, you can then say "run my action-items recipe on this" and skip re-typing the prompt.

Recipes live in your Whipscribe account. They're shared between the GPT path, the MCP Connector path, and the web app at whipscribe.com/home. Build them once, use them everywhere.

What ChatGPT alone can't do (and why Whipscribe is the bridge)

ChatGPT's native voice features handle short, real-time speech in the chat — voice mode for spoken conversations, the microphone button on mobile for dictation. They're built for talking to ChatGPT, not for processing a 45-minute meeting recording you already have on disk.

Three things Whipscribe adds that the native voice features don't:

• Speaker diarization. A two- or three-person meeting comes back as Speaker 1 / Speaker 2 / Speaker 3 with timestamped turns, not as one undifferentiated text stream.
• Word-level timestamps. Every word in the transcript is timestamped, which is what makes "click the timestamp, hear the moment" workflows work — and what makes Shorts and subtitles possible downstream.
• Persistence. The transcript is saved to your Whipscribe library so you can search it later, reuse it in another chat, or pull a quote out months from now.

Practically: voice mode is the right tool for a 30-second question to ChatGPT. Whipscribe is the right tool for a 45-minute call you need to do something with.

A worked example — the 45-minute product call

Concrete, end-to-end, from inside ChatGPT:

1. Open the Whipscribe GPT.
2. Drag product-call.m4a into the message box.
3. Send: "Transcribe this with speaker labels. Then pull decisions, action items, and open questions as separate sections."
4. About 60-90 seconds for a 45-minute file. The transcript and structured summary come back inline.
5. Reply: "Save the transcript to my Knowledge folder named 'Product calls'." A folder is created if it doesn't exist; the transcript is filed.
6. Reply: "Now run my 'weekly recap' recipe across all transcripts in 'Product calls' from the last 7 days." A summary spanning the week's calls is produced.

Three messages, one workflow, no tab switching. The transcripts are also visible at whipscribe.com/home — the chat and the web app share state.

Privacy and account specifics

• OpenAI training is off on this GPT. User-uploaded audio in the Whipscribe Custom GPT is not used to train OpenAI models.
• Files are processed by Whipscribe, not OpenAI. The transcription itself runs on Whipscribe infrastructure.
• Default 7-day retention on raw audio. Transcripts saved to a Knowledge folder are kept indefinitely; raw audio not saved is purged after 7 days.
• One ChatGPT email = one Whipscribe email. Sign in with the same address both places so credits and library are shared.
• Free tier covers 30 minutes of transcription per day, no card required. Beyond that, pay-as-you-go is $1 per hour of audio at whipscribe.com/credits; credits never expire.

What this looks like on mobile

Mobile is where most voice memos live, so the GPT path's mobile-first design matters. On the ChatGPT iOS or Android app, opening the Whipscribe GPT works the same as on web. Tapping the paperclip in the message box pulls from your phone's Files / Voice Memos / Camera Roll. A 5-minute voice memo transcribes in roughly 30 seconds; the structured summary follows.

One concrete pattern: record a thought as a voice memo while walking, drop it into the Whipscribe GPT during your next coffee break, ask for the structured outline. The transcript and the outline both end up in your library and in the chat for later edits.

The one thing not to do

Don't paste a YouTube, Spotify, or other third-party platform URL into the chat and ask Whipscribe to transcribe it. The integration is built around your own content: files you have, URLs to media you host, recordings you made. Transcribing other people's hosted content is a different category of question with platform-specific terms attached, and we don't route around them. Bring your own audio.

Frequently asked

• Does it work on the ChatGPT free plan? Yes. The Custom GPT works on Free, Plus, Pro, Team, and Enterprise. The MCP Connector is Plus and Pro only.
• Do I need a Whipscribe account first? No — first authorize creates one for you using your ChatGPT email.
• Can I use my Whipscribe credits in both places? Yes. Credits are per-account, not per-surface.
• What languages are supported? Whipscribe runs Whisper-family models which cover ~99 languages with auto-detect. ChatGPT can then summarize, translate, or restructure in any of its supported languages.
• How long can a single recording be? A couple of hours per file in practice. Very long files are best split — ChatGPT itself prefers shorter contexts even when the transcription succeeds.