ChatGPT Can Now Run Work Tasks by Voice on Your Phone — What’s Real (and What Still Needs a Tap)

You’re between meetings, phone in hand, and you need a Slack thread turned into three bullets before the next call. Typing that on a phone keyboard is slow. On September 23, 2026, OpenAI’s ChatGPT release notes documented a clearer path: speak the request in the mobile app, let Work start the multi-step job, and — when an action requires approval — confirm it on screen. A spoken “yes” is not a substitute for that tap. That last part is the story, not a footnote.

This is a specific product moment — Voice plugged into ChatGPT Work on the phone you already carry — with the same permission rules that decide whether a draft stays a draft.

## What happened

According to [OpenAI’s ChatGPT release notes](https://help.openai.com/en/articles/6825453-chatgpt-release-notes) for September 23, 2026 (“Use plugins in Voice and get work done by speaking”):

- Live Voice can use the plugins and connected apps available to your account on web, iOS, and Android, and you can follow written responses in the chat while you talk.

- Voice is also available in Work on web and mobile. You can ask it to create documents, presentations, and spreadsheets, use connected apps, or work in a browser.

- If you end a Voice call in Work while a task is still running, the unfinished work can continue in text.

[TechCrunch](https://techcrunch.com/2026/09/23/chatgpt-mobile-app-gets-voice-based-agentic-features/) reported the same day that Plus and Pro users can use the Work tab on their phone for tasks such as creating a document, drafting an email, or summarizing Slack — with richer text alongside voice, and the ability to start on the go and resume on desktop. [9to5Mac](https://9to5mac.com/2026/09/23/openai-just-upgraded-chatgpt-voice-in-three-ways/) and [The Verge](https://www.theverge.com/ai-artificial-intelligence/999540/you-can-now-summon-chatgpt-voice-from-your-phone) corroborated Voice with plugins (email, calendar, Slack) and agentic tasks across connected apps from the phone.

OpenAI’s [ChatGPT Voice](https://help.openai.com/en/articles/20001274/) and [Work and Codex](https://help.openai.com/en/articles/20001275-chatgpt-work-and-codex) help articles spell out the mobile path: open Work, select the Voice control, start a Live conversation, and follow written results in the chat. That matches the September 23 changelog.

Context, not the news: ChatGPT Work — the agent for longer deliverables — shipped earlier in 2026 for eligible paid plans ([OpenAI’s Work launch](https://openai.com/index/chatgpt-for-your-most-ambitious-work/)). Desktop Voice into Work/Codex came earlier too. September 23 is when Voice-in-Work on web and mobile, plus plugins in Live Voice, become a named mobile-ready workflow — not when Work itself was invented. Chat remains fast questions; Work is the surface for longer jobs and finished materials. Voice now opens that Work path from the phone, not only from a keyboard.

## What you can actually do

The practical win is narrow and real: start a deliverable when a keyboard is awkward — a client reply, status doc, or meeting follow-up while you walk out of the previous call.

Examples that match what OpenAI and the press describe — not fantasy automation:

- Dictate: “Turn today’s Slack updates from #launch into a one-page status doc,” then review the draft on screen later.

- Ask Voice in Chat (or Work, if you have it) to pull context from a connected inbox or calendar, then edit in text when you sit down.

- Start a Work task on the phone between meetings; let it continue in text after you hang up Voice; finish polish on desktop. Cloud Work chats sync across web, mobile, and desktop ([Work and Codex](https://help.openai.com/en/articles/20001275-chatgpt-work-and-codex)).

[Business Insider](https://www.businessinsider.com/openai-boosts-chatgpt-voice-mode-mobile-web-app-2026-9) reported OpenAI’s voice product lead describing the design as voice in with the right output back — text, speech, or both. For work, that usually means you speak the messy brief and read the structured result.

## The important catch

One important limitation: On web and mobile, if an action needs approval, Voice prompts you to review it on screen. Use on-screen controls to approve or decline. Spoken approval is not supported ([ChatGPT Voice](https://help.openai.com/en/articles/20001274/), [Work and Codex](https://help.openai.com/en/articles/20001275-chatgpt-work-and-codex)). [Notebookcheck](https://www.notebookcheck.net/ChatGPT-Voice-can-now-use-your-apps-but-approvals-stay-on-screen.1409087.0.html), citing the same FAQ: saying “yes, send it” does not send the email — in a car or pocket, the task waits for the screen. Control loop: speak to start, tap to approve when the screen requires it.

Permission modes decide when you even see that card. Modes such as Allow low-risk actions or Allow all actions can mean some actions run without an extra prompt; higher-risk steps (send an email or message, create or delete a file, change access settings, make a purchase) may still need confirmation or be denied ([Managing app permissions](https://help.openai.com/en/articles/20001495-managing-app-permissions-in-chatgpt)). Spoken approval is never a substitute when an on-screen approval is required.

> Permissions — keep it simple

>

> - Start with Always ask or Allow read actions before you connect Gmail, Slack, or Drive for Voice use.

> - Treat every Voice-produced email or Slack message as a draft until you review it — and tap — when the screen asks.

> - Spoken approval does not replace on-screen approval when one is required.

>

CarPlay Voice exists on supported iPhones ([ChatGPT Voice](https://help.openai.com/en/articles/20001274/)); OpenAI’s Help still says to set up before driving and only use the device when law and conditions allow. Consequential actions that need approval still wait for a screen tap. On a managed Business/Enterprise workspace, admin policy can restrict plugins — personal settings do not override that ([permissions help](https://help.openai.com/en/articles/20001495-managing-app-permissions-in-chatgpt)). This is product hygiene, not legal advice.

## Who gets it

Use OpenAI’s wording, not the compressed press paraphrase.

From the [September 23 release notes](https://help.openai.com/en/articles/6825453-chatgpt-release-notes) and [ChatGPT Voice](https://help.openai.com/en/articles/20001274/) help:

- Free and Go: You can use Voice in Chat with the plugins your plan supports. That is not the full Work experience.

- Voice in Work: Requires both Voice and Work access. Work on web and mobile is for eligible paid plans ([Work and Codex](https://help.openai.com/en/articles/20001275-chatgpt-work-and-codex)). TechCrunch’s Plus/Pro Work-tab examples point the same way.

- Existing app connections, permissions, and usage limits still apply. Voice minutes count toward your plan’s Live allowance, and Work tasks started by speaking still count toward normal Work usage — same as typed Work.

If a colleague says “everyone can voice-run Work now,” that is not accurate. Free and Go get spoken Chat with supported plugins; full Voice-driven Work (docs, presentations, spreadsheets, browser inside Work) needs Work access on paid-eligible accounts.

Look for the Work surface when you want a finished doc or deck. Do not assume every Voice chat is a Work run — OpenAI still keeps chat and workspaces separate ([TechCrunch](https://techcrunch.com/2026/09/23/chatgpt-mobile-app-gets-voice-based-agentic-features/)).

## What still doesn’t work

Be precise so you do not over-promise to yourself or a team:

- Codex is not selectable on web or mobile. You can reach supported desktop Codex sessions from the mobile Remote tab; those do not become regular web/mobile chat history ([Work and Codex](https://help.openai.com/en/articles/20001275-chatgpt-work-and-codex)).

- Local files on your computer are a desktop Work path. Work on web and mobile cannot directly access files on your PC ([Work and Codex](https://help.openai.com/en/articles/20001275-chatgpt-work-and-codex)).

- Live Voice does not support video or screen sharing; those remain on the separate Advanced Voice path on mobile where available ([ChatGPT Voice](https://help.openai.com/en/articles/20001274/)).

- Broader Work surfaces such as Sites, finances tools, or cloud browser exist as Work capabilities in press and product docs, but the September 23 Voice note specifically calls out documents, presentations, spreadsheets, connected apps, and browser work — do not treat every Work feature as “new because of Voice.”

Older search snippets that claim Voice in Work is desktop-only are stale relative to the current Help text, which documents Voice in Work on web, iOS, and Android.

## Try it in 5 minutes

1. Read-only first. Connect one app you already use, set permissions to Always ask or Allow read actions, and ask Voice a question that only requires reading (“What’s on my calendar this afternoon?”). Practice the approval card before you ever ask it to send.

2. Draft, don’t send. In Work (if you have access), speak a request for a status doc or email draft. End the Voice call if needed and finish in text. Do not approve a send on the first try.

3. Phone → desk. Start a Work task on mobile, confirm it continues after Voice ends, and open the same cloud Work chat on desktop to edit. That continuity is the time-saver; the tap is the control when approval is required.

Voice can start serious Work on your phone as of the September 23, 2026 notes. When an action requires approval, Voice still hands control back to the screen: speak to start, tap to approve.

---

Sources cited inline above. Primary official: [ChatGPT release notes (Sep 23, 2026)](https://help.openai.com/en/articles/6825453-chatgpt-release-notes), [ChatGPT Voice](https://help.openai.com/en/articles/20001274/), [Work and Codex](https://help.openai.com/en/articles/20001275-chatgpt-work-and-codex), [Managing app permissions](https://help.openai.com/en/articles/20001495-managing-app-permissions-in-chatgpt). Press corroboration: [TechCrunch](https://techcrunch.com/2026/09/23/chatgpt-mobile-app-gets-voice-based-agentic-features/), [9to5Mac](https://9to5mac.com/2026/09/23/openai-just-upgraded-chatgpt-voice-in-three-ways/), [The Verge](https://www.theverge.com/ai-artificial-intelligence/999540/you-can-now-summon-chatgpt-voice-from-your-phone), [Notebookcheck](https://www.notebookcheck.net/ChatGPT-Voice-can-now-use-your-apps-but-approvals-stay-on-screen.1409087.0.html), [Business Insider](https://www.businessinsider.com/openai-boosts-chatgpt-voice-mode-mobile-web-app-2026-9).