Igor Pogany distinguishes ordinary voice dictation from interactive voice, where ChatGPT also replies aloud and keeps a visible text conversation. The redesigned experience accepts images in the web interface and works in the mobile app, although live camera and screen-sharing features may still require the older voice mode rather than the new model.
The desktop app is the strongest implementation because the floating conversation can sit beside other work and inspect the current screen when asked. Pogany demonstrates it explaining the Codex interface to a new user and breaking a Pomodoro timer project into a concrete first prompt, turning voice into contextual guidance instead of a separate chat window that must be continually managed.
Voice also reaches ChatGPT Work and connected services. In the demonstration, it reads a Slack request, creates a scheduled release-monitor task, checks a connected calendar and proposes an available meeting slot. These examples show how conversational input can trigger structured work rather than merely produce spoken answers.
The limitations are mostly about context and control. New voice chats do not reliably inherit project files, which makes persistent rules and recurring commitments harder to apply without broad account-level instructions. Pogany also finds Claude's improved voice useful for normal chat and dictation but not yet available across its coworking, coding and desktop surfaces, leaving ChatGPT ahead on integrated voice-driven workflows.
Watch on YouTube
