// AI NEWS · FEATURE

OpenAI Brings Voice-Based Agents to the ChatGPT Mobile App: Plugins in Live Voice and Voice in ChatGPT Work

You can now talk to ChatGPT on your phone and have it act — invoking connected apps mid-conversation and kicking off agentic Work tasks like drafting a doc or summarizing Slack, all by voice.

2026-09-23 · by Moe Ameen

What happened

On September 23, 2026, OpenAI upgraded ChatGPT Voice in three connected ways, and the net effect is that the mobile app's voice mode stopped being just a talking chatbot and became a way to trigger actions hands-free. The company enabled plugins inside the Live voice experience across web, iOS, and Android, so you can invoke a connected app — email, calendar, Slack, and others your plan supports — while you're mid-conversation. It also brought Voice into ChatGPT Work on web and mobile, meaning you can speak a request and have OpenAI's cloud-connected agentic system carry it out.

Practically, that lets Plus and Pro subscribers open the Work tab on a phone and, by voice, create a document, draft an email, summarize Slack messages, build a spreadsheet or presentation, or run tasks in the cloud browser — the agent pulls from connected services to do the work. Free and Go users get the lighter half: plugins and connected apps in voice conversations, without the full Work task surface. OpenAI also said voice conversations now show richer text output, that you can switch between text and voice easily, and that you can start something on the go and resume it on desktop.

A few boundaries are worth stating plainly. Actions that need confirmation still surface an on-screen approval — voice approval isn't supported — and existing app connections, permissions, and usage limits carry over. A voice-started Work task consumes the same Work usage as a typed one, follows your plan's voice limits, and if the task runs past the end of the voice call it simply continues in text mode. Under the hood, ChatGPT Voice can now be powered by OpenAI's GPT-6 models — Astra, Sol, and Luna — rather than a single voice model. Specifics of tiering and limits change quickly; confirm what your plan includes in OpenAI's own release notes before relying on it.

Why it matters for creators

  • Voice becomes the fastest capture method. The friction of typing an idea on a phone disappears — you can talk through a script, an outline, or a brief while walking or driving and have it drafted for you.
  • The "agentic" part is real but office-shaped: the tasks OpenAI highlights are documents, emails, spreadsheets, Slack summaries, and browser errands — deliverables and admin, not finished posts for your audience.
  • Free and Go users only get plugins in voice, not the full Work task surface, so the headline agentic workflows are gated to paid tiers — worth knowing before you plan around them.
  • On-screen approval for confirmation actions means it isn't fully hands-free for anything consequential — you still look at the phone to approve, which limits true eyes-off use.
  • A drafted doc or a summarized thread is an input, not a channel. Turning that raw output into clips, carousels, a blog, a newsletter, and native posts across platforms is the work that still comes after the voice call.

How to act on this with Kompozy

The useful way to read this launch is that voice just became the best way to *get an idea out of your head* on a phone — and left completely untouched the job of turning that idea into published content. ChatGPT's voice agent will happily draft you a document, summarize a Slack channel, or build a spreadsheet, but those are office deliverables. None of it becomes a captioned Short, a brand-exact carousel, a blog post, or a scheduled week of posts across your platforms. That gap is exactly what [Kompozy](/) closes.

Here's the clean handoff: talk through your idea to ChatGPT on the go and let it draft the script, outline, or long-form notes. Paste that text into Kompozy as a source, and it generates the spread the voice agent can't — Persona Shorts and avatar video, brand-exact Carousel Posts, Photo Posts and Quote Graphics, a Blog Article and an Email Newsletter, and native Text Posts — all held to one voice by a Persona Brief. Then [Autopilot](/glossary/autopilot) and a per-post review pipeline schedule and publish the batch across the eight social platforms plus blog and email from a single queue. ChatGPT's mobile voice agent is a superb dictation-and-drafting front end; Kompozy is the generation-and-distribution engine that ships what you dictated. For OpenAI's broader agentic push, see [ChatGPT Work's launch](/news/openai-chatgpt-work-launch) and the [GPT-Live voice models](/news/openai-gpt-live-voice-search); for the desktop version of this same voice upgrade, see [ChatGPT Voice on the desktop app](/news/chatgpt-voice-desktop).

Quick takeaways

  • On September 23, 2026, OpenAI added plugin support to ChatGPT's Live voice on web, iOS, and Android and brought Voice into ChatGPT Work on web and mobile.
  • Plus and Pro users can run Work tasks by voice — draft docs and emails, summarize Slack, build spreadsheets and presentations, use the cloud browser; Free and Go users get plugins and connected apps in voice only.
  • Confirmation actions require on-screen approval (no voice approval), existing app permissions and usage limits apply, and voice-started Work tasks continue in text mode after the call ends.
  • ChatGPT Voice can now be powered by OpenAI's GPT-6 models (Astra, Sol, Luna). The agentic tasks are office deliverables, not published content.

Frequently asked questions

What can ChatGPT's mobile voice agent actually do now?

As of September 23, 2026, you can invoke connected apps (like email, calendar, and Slack) during a spoken conversation on iOS and Android, and — on Plus and Pro plans — start ChatGPT Work tasks by voice from a phone: creating a document, drafting an email, summarizing Slack messages, building a spreadsheet or presentation, or running tasks in the cloud browser. Free and Go users get plugins and connected apps in voice, but not the full Work task surface. Actions that need confirmation still show an on-screen approval.

Is it fully hands-free?

Not entirely. You can speak requests and have the agent act, but anything requiring confirmation surfaces an on-screen approval — OpenAI has said voice approval isn't supported for those. So for consequential actions you still look at and tap the phone. And if a Work task runs past the end of your voice call, it continues in text mode rather than staying voice-driven.

Does ChatGPT voice publish content to my social platforms?

No. The voice agent produces office-style deliverables — documents, emails, spreadsheets, summaries — and can act inside connected apps, but it does not generate finished multi-format content or publish to social platforms. To turn a voice-drafted idea into captioned Shorts, carousels, a blog, a newsletter, and scheduled posts across platforms, you bring the text into a content engine like Kompozy.

Related news

← All AI news · Get started →