As of September 23, ChatGPT Voice can work with the plugins and connected apps available to a user’s account. Under the right conditions, a spoken question can now become a task: the assistant may draw on a calendar or initiate a supported action. For users, the meaningful change is less about a more natural voice than access to tools that previously lived mostly in text chat.
Key takeaways
- OpenAI is bringing available plugins and connected apps to Live voice on the web, iOS, and Android.
- In ChatGPT Work, users can also start documents, presentations, and spreadsheets by voice; an unfinished task can continue in text.
- Existing connections, permissions, usage limits, and workspace rules still apply.
- When an action requires approval, users review it on screen rather than saying yes aloud.
What actually changes in everyday use
Until now, a voice conversation was primarily a way to ask questions and hear answers. It can now use apps that are already set up and permitted for the account. Consider a straightforward sequence: ask about an appointment, retrieve relevant calendar information, then draft a response. Whether the assistant can also change an event or send a message depends on the app, the permissions granted, and the actions available. The new voice interface does not, by itself, turn on write access.
OpenAI explicitly lists the web, iOS, and Android for the new Live capability. Written responses remain visible in the chat during the conversation. That matters when names, dates, or results need checking before a further action: users do not have to rely only on what they heard. Someone who has not connected an app may first need to connect it and sign in. Connecting a service also does not confer permissions inside that service.
This shift matters more to a workflow than another model name. OpenAI’s recent model changes affect cost and performance; voice access to apps instead changes how a specific task is carried out. It may reduce the back-and-forth between dictation, an app, and the chat. The announcement alone cannot tell us how reliably the system will handle more complex sequences.
Work broadens the workflow, but does not grant new rights
For ChatGPT Work, OpenAI also describes a work experience on the web and mobile devices. Users can ask by voice to create documents, presentations, and spreadsheets, use connected apps, or work in a browser. If the voice conversation ends, an unfinished task can continue in text. That is particularly useful for work that starts with a spoken instruction but needs to produce a file or a traceable record.
Availability varies by plan. Free and Go users can use Voice in Chat with the plugins their plans support. Voice in Work requires access to both Voice and Work. Company accounts also have workspace settings to consider. A spoken instruction cannot bypass a blocked app or missing access rights. For teams, this is a crucial boundary: a new way to issue instructions does not silently expand the data an employee is allowed to reach.
Email and calendars illustrate the distinction. OpenAI’s Outlook documentation describes reading and writing as separate capabilities. Available actions depend on account rights, Microsoft permissions, and ChatGPT workspace settings. Someone who only wants to ask about today’s schedule needs a different set of permissions from someone who wants to change meetings or send messages. Before connecting an app, the useful question is therefore not simply whether it appears in the directory, but which specific actions are enabled for that account.
How to try it without losing track of the result
A first test can start with a read-only task: in Live voice, ask about the next appointment or request information from an app that is already connected, then inspect the result in chat. Check which source the assistant used and whether the time, recipient, or context is right. Once that simple chain works, it becomes easier to judge an action that changes data elsewhere. OpenAI says the options available can depend on the plan, region, app version, and, where relevant, workspace rules.
The separation is especially clear for approvals. If an action needs confirmation, Voice on the web and mobile devices asks the user to review it on screen. According to OpenAI, spoken approval is not supported. That may be less convenient on the go than entirely hands-free control, but it also keeps a consequential action from depending only on a possibly misunderstood word. For messages, bookings, or calendar changes, the visible review step is part of the value, not merely friction.
The next test is a verifiable workflow
The update makes voice more useful for people who combine information from several apps or want to begin work away from a keyboard. The question now is whether the assistant chooses the right app, shows intermediate results clearly, and pauses at the right point before making changes. A good voice feature saves steps; a good task feature also reveals what data it read and what action it actually completed. That is the standard users should apply when they try the new mode for themselves.