ChatGPT Meetings Plugin Captures Calls and Deletes Audio After Use

OpenAI entered the meeting-capture market at DevDay 2026 with a Meetings plugin for ChatGPT: it records online or in-person meetings, generates personalized summaries and action items that land in ChatGPT Space, and connects to calendars so users can ask for follow-up work on what was discussed. The privacy posture is explicit — audio is deleted after processing. The beta is live on macOS for Pro and Business plans, Enterprise is in limited testing, and Windows, iOS, and Android are slated to follow.

This piece covers how the plugin works and where its output goes, how audio deletion positions it against incumbent notetakers, what personalized summaries mean in a shared-workspace world, and the compliance questions to settle before rolling it out. It is written for team leads and IT owners weighing meeting-AI tools.

The Meetings plugin: what launched

The capture surface spans both meeting modes — online calls and in-person conversations — which matters because the physical meeting has remained the gap in most transcription workflows. Output is not a raw transcript dump: the plugin produces summaries and action items, personalized to the user, and files them into Space, the shared team workspace OpenAI launched the same day.

The calendar connection turns capture into workflow: with meetings on the calendar linked, users can ask ChatGPT for follow-up work grounded in what was actually said — drafting the email a discussion implied, tracking the commitment someone made, preparing for the next occurrence.

Audio deletion as a design position

Deleting audio after processing is the line OpenAI chose to differentiate on. Incumbent meeting bots have normalized persistent recordings — searchable archives of everything said — which is precisely what makes legal and HR departments wary of them. OpenAI's model keeps the derived artifacts — summaries, action items — and discards the raw recording, trading archival replay for a smaller retention surface.

That trade will suit organizations that want meeting intelligence without building a warehouse of recorded conversation, and frustrate users who rely on going back to the tape. It also simplifies part of the consent conversation, though only part: capture still happens, and in-person capture in particular needs clear policy regardless of what is deleted afterward.

Where the output landing in Space matters

Filing meeting output into Space connects the plugin to the rest of DevDay's collaboration stack. A meeting's action items land where teams, dots, and shared tasks already operate — so the assignee's always-on agent can pick work up from the meeting record, and a team's space accumulates decisions alongside the documents and tasks they affect. The personalized-summary mechanic fits the same design: each participant gets the meeting through the lens of what they own.

That integration is the competitive argument against standalone notetakers: their transcripts terminate in their own apps, while OpenAI's summaries land in a workspace with agents attached to act on them.

Rollout, platforms and who gets it

Availability is deliberately staged: macOS beta for Pro and Business today, Enterprise in limited testing, Windows and mobile later. The macOS-first order tracks the ChatGPT desktop app's role as the hub for DevDay's other launches — dots creation and code review are desktop-anchored too. Teams on Windows-heavy fleets are waiting on the follow-up platforms before this can be standard kit.

For Pro and Business users on Macs, the pilot cost is near zero: enable the beta, run it on internal meetings first, and evaluate summary quality and action-item precision against whatever notetaker the team uses today.

What to verify before wide rollout

Consent and jurisdiction come first: recording rules vary by region, in-person capture raises them more sharply than call capture, and 'audio deleted after processing' should be verified against retention policy language — what exactly is retained (summaries, action items) and for how long. Enterprise buyers should also confirm how the plugin's artifacts interact with workspace data controls and the training-data defaults OpenAI applies to Business and Enterprise content.

Watch the platform expansion cadence and whether Enterprise testing surfaces admin controls — per-meeting-type policies, participant notification, export restrictions — that the consumer beta lacks. The meeting-notes market has been a feature race; OpenAI just made it an architecture race, and the incumbents' response will define the category's next year.