ChatGPT Voice Adds Connected-Service Actions and Voice-Driven Work Creation
OpenAI is positioning ChatGPT Voice as a way to act on connected services and create or revise work through spoken requests. Its launch description says Voice can use plugins such as email, calendar and Slack, and that ChatGPT Work users can make documents, decks, sites and spreadsheets on web and mobile. The examples include checking for duplicate charges, contacting a marina and building a pottery storefront; the source presents these as demonstrations, not evidence that every task is completed end to end.
Voice’s first proposition is action across connected services
OpenAI presents ChatGPT Voice as more than a way to ask questions: it can be used to carry out tasks through connected services. The stated plugin examples are email, calendar and Slack. In the on-screen scenarios, a user asks Voice to put a post-office trip on the calendar, check finances for duplicate charges, and contact the Fairfield Marina about a sailboat rental.
The financial example links a check to a requested follow-up. Voice reports two payments for Sound and Stereo; the user asks it to cancel the duplicate and email a refund request, and Voice agrees. The marina exchange extends the pattern across time: Voice says it will send a rental inquiry, later reports that a boat is available at 4 p.m., and receives the instruction to book it. The source does not show the booking being completed.
These scenes make the intended role legible: a spoken request can involve retrieving information, initiating an action, and returning to a thread after a reply. They are examples of the product’s stated scope, not evidence in the source about how every connected service behaves or what confirmation a task requires.
Other requests show the information-retrieval side of that same proposition. Voice answers a question about the salon appointment and reports west winds at the marina of around eight knots, with gusts up to twelve. The user jokes that Voice does not know how to sail; Voice says it does. The exchange keeps the product’s practical claims in a deliberately everyday register: the assistant is asked for a timely detail while the user is on the move.
OpenAI says the feature is rolling out globally in the latest version of the app. That availability claim accompanies a broader description of plugins, including email, calendar and Slack; the scenes themselves focus on calendar, finances, and marina email.
The second proposition is creating work and revising it by voice
OpenAI also describes Voice as a way to produce or change work, not only to operate existing services. The concrete example is a request for a website to sell pottery. Voice says it will start building; a laptop then displays a storefront called “Lavelle’s Pottery,” with ceramic products and the line “Colorfully crafted, wonderfully wonky ceramics meant to brighten the everyday.” The user asks for a simple checkout page, and Voice says it is adding one.
The example matters because the request does not end at generating a first version: the user can ask for a revision in ordinary speech. The source shows the storefront and the request for checkout, but does not show a completed checkout flow or a purchase. Later, the user asks whether the pottery site has received orders; Voice says there are none. The exchange therefore presents creation, revision, and a follow-up check without implying that the site has made sales.
OpenAI places that example within a wider ChatGPT Work offer on web and mobile. It says users can create documents, decks, sites and spreadsheets, or tackle complex tasks in the browser, by talking. The pottery site is the only creation task shown; the other formats and broader browser work are stated product capabilities, not additional examples in the source.
Together, the two propositions describe a voice interface for both acting through connected services and making or revising work. The distinction is useful: the calendar and marina requests concern actions and information tied to other services, while the pottery request concerns producing a new artifact. OpenAI’s launch description encompasses both, while the specific scenes show only a portion of the stated range.