OpenAI brings Record mode to ChatGPT on macOS

If you are tired of running a third-party transcription app alongside your meeting notes, OpenAI has some good news.

OpenAI is rolling out its new Record mode in the ChatGPT macOS desktop app to Pro, Enterprise, and Edu subscribers at no extra charge. The rollout follows a soft launch for Team users earlier this month.

Two hours of audio, live transcripts, and instant code

Mac users on eligible paid plans can kick off a session by hitting the new Record button right at the bottom of any chat. Sessions can run for up to two hours, which should comfortably cover even the most dragged-out brainstorms.

As participants speak, the app generates a live transcript with support for multiple speakers. OpenAI notes on its support page that it currently works best in English, while accuracy in other languages is improving but may vary.

Once you stop recording, the transcript lands in your chat history, and ChatGPT spins up a structured recap inside a private canvas workspace. You can edit the recap yourself or have ChatGPT refine it. Beyond standard bullet points, the AI extracts action items and can even generate ready-to-run code snippets straight from the conversation.

Audio gets wiped, but watch the training toggle

Recording sensitive conversations always raises privacy flags, and OpenAI took steps to address that upfront.

OpenAI states that raw audio recordings are not used for training and are deleted automatically once the transcript is generated.

However, the text transcripts are treated differently. Pro and Plus subscribers still need to dive into their settings and turn off the Improve the model for everyone toggle if they want to ensure their meeting text stays out of future training data.

OpenAI also reminds users to get explicit consent from everyone in the room before hitting record, making clear that compliance with local recording laws falls squarely on the user.

Alongside the desktop update, OpenAI quietly opened up its image generation feature to all WhatsApp users, expanding the reach of its 1-800-ChatGPT experiment.

Why it matters

We think this is a classic workflow takeover by OpenAI. Instead of stopping at voice-to-text, it bridges the awkward gap between raw audio and actual execution by dropping action items and code right into a canvas workspace. For developers and product teams already living on a Mac, ditching standalone meeting bots is almost a no-brainer.

This puts immense pressure on single-purpose AI note-taking startups. When the foundation model provider offers two-hour recording, live speaker separation, and direct code synthesis for free inside its desktop app, paying for a separate transcription wrapper makes little sense.

That said, the rollout has clear boundaries. Leaving Windows and mobile users out for now limits company-wide adoption, non-English accuracy remains a work in progress, and the requirement to manually opt out of training on transcripts means security-conscious teams still have homework to do.