Updated October 16, 2026: This article revisits the free mobile voice rollout reported in November 2023. The update is based on OpenAI’s documentation, with no hands-on audio testing. Historical launch details and current guidance are separated below.
Cover image: The original artwork is retained as an archival illustration. Its “Groundbreaking Innovation” headline is promotional wording from the older article, not a measured performance claim.
On November 21, 2023, OpenAI’s ChatGPT release notes announced that voice conversations were available to free users. The instructions directed people to the phone app and its headphones icon. The news concerned broader access to an existing feature.
Two milestones in 2023
OpenAI had introduced voice and image capabilities on September 25, 2023, with a two-week rollout planned for Plus and Enterprise. Voice was initially an opt-in feature on iOS and Android. That earlier announcement and November’s free-access update are separate milestones.
The September system used Whisper to transcribe speech and a text-to-speech model to generate spoken replies. OpenAI worked with professional voice actors on its five voices. These are historical implementation details, rather than a description of every current Voice option.
The November notice concerned subscription access. It did not promise unlimited use forever or define which future Voice capabilities would be available without payment.
The distinction matters when reading older tutorials. A launch article answers what was announced then. A current support page is the better place to check today’s controls and eligibility. Our September 2023 voice and image retrospective covers the earlier event in more detail.
Voice today
The current ChatGPT Voice guide lists Live, Advanced, and Standard options. Standard transcribes speech before replying; Live can listen and speak simultaneously. Availability varies by plan, workspace, region, and app version. Voice in Chat supports web, iOS, and Android; select the Voice icon to begin. Microphone permission may be needed.
Free Live access is limited, with no fixed allowance promised in the guide. Paid allowances differ. Live does not support video or screen sharing; eligible mobile subscribers use Advanced for those capabilities. A familiar feature name does not establish identical access across versions.
A simple practice exercise
This fictional example illustrates a way to structure a conversation; it is not an account of a test. Imagine planning a casual board-game evening for four adults. You want a short opening explanation rather than a long set of instructions. A starting prompt could be:
“Help me prepare a one-minute welcome for a fictional board-game evening. There are four players, two games to choose from, and 45 minutes available. Ask one question at a time before drafting the welcome. Do not invent rules for either game.”
After supplying the game names and checking their printed rules yourself, a follow-up could be: “Give me two opening sentences, one relaxed and one more structured. Keep the rest as a checklist I can read.”
To judge whether the exchange helped, compare the result with your original requirements. Does it mention four players? Does the proposed timing fit 45 minutes? Has it invented a scoring rule you never supplied? Keep the useful phrasing and correct any unsupported detail before using it.
This keeps the exercise easy to evaluate. You are judging a small piece of writing against facts you already know, without treating an engaging delivery as proof that the content is right.
Check privacy first
The Voice guide’s data controls section says audio and video clips are not used for training unless you choose to share them. Transcripts and other files can be treated differently, depending on your plan and settings.
OpenAI’s Data Controls guide explains that turning off “Improve the model for everyone” excludes new conversations from model training while keeping them in history. For personal accounts, submitting feedback can still make the associated conversation eligible for training. Business, Enterprise, Edu, and Healthcare workspace content is not used for training by default.
Before a practice session, decide what information the task actually needs. The fictional evening above needs no guests’ addresses, contact details, or private conversations. Substitute made-up details when those will serve the exercise just as well. For work material, follow your organization’s rules before speaking it into any online service.
Keep the claim in context
A useful personal check is to separate three questions: Did the interaction feel comfortable? Did the answer address the request? Were its factual details correct? Those questions require different evidence. This documentation review makes no claim about response speed, accent handling, or how closely a voice resembles a person. Use the historical record to understand the rollout, and evaluate any present-day result on its own merits.