Google's Gemini Can Now Make Phone Calls for Pixel 11 Owners

Google has started rolling out Call For Me, a feature that lets its Gemini AI chatbot make and conduct entire phone calls on a user's behalf. The preliminary rollout begins today, and it marks a significant expansion of Google's phone-based AI assistance: the company's Hold for Me feature, available for several generations of Pixel smartphones, could only wait on hold during a call. Call For Me goes further — the AI speaks to a person on the other end of the line, navigates directory menus, and waits on hold, all without the user picking up the phone.

The feature is available on a narrow set of devices and users. According to Google's blog post, Call For Me arrives as an early preview for paid Gemini subscribers in the United States who own a recent Pixel 11 smartphone and who are enrolled in the Phone by Google Public Beta. Eligible users can trigger a call from within the Gemini app by asking the chatbot to make the call.

What the AI can actually do

Google says Call For Me can handle moderately complex tasks, not just simple scripted requests. The examples the company gives are practical, everyday errands: checking whether a hardware store has an item in stock, moving a haircut appointment to another day, or reserving patio seating at a restaurant. Each of these involves a real conversation with a human on the other end, including answering basic questions and confirming details.

The Gemini app shows a live transcript of the call as it happens, giving the user visibility into what the AI is saying and hearing. Google frames this as a control mechanism rather than just a convenience. "You remain in control at all times with a live transcript view and the ability to take over the conversation at any point," the company says in its blog post. That means a user watching the transcript can step in mid-call and continue the conversation in their own voice if the AI stalls or the task goes off script.

An incremental step with broader stakes

The release matters because it moves AI assistants from passive tools — summarizing, searching, drafting text — into live, real-time voice interactions with third parties who may not know they are talking to a machine. Google has been building toward this for years. Hold for Me, which the company has shipped on several generations of Pixel phones, transcribed hold music and call center queues for the user. Call For Me absorbs the entire interaction: dialing, speaking, waiting, and completing the task.

The gating of the feature also signals how Google is managing the risk. Limiting Call For Me to paid Gemini subscribers, a single device generation, and members of a public beta in the US keeps the initial audience small and self-selected. The tasks themselves are bounded — appointments, reservations, stock checks — which reduces the chance of the AI mishandling a high-stakes conversation.

The live transcript and manual takeover option address the most obvious failure modes. A user who watches the call unfold can correct the AI before an error compounds, whether that error is a wrong reservation time or a misunderstood request on the other end of the line.

What comes next

For now, Call For Me remains a preview with no announced timeline for wider availability. Google has not said when the feature will reach other Pixel models, other countries, or users outside the paid Gemini subscriber base. The company's examples — retail stock checks, appointment changes, restaurant reservations — suggest the initial scope is deliberately confined to routine consumer errands rather than sensitive calls like medical appointments or financial matters.

If the early preview goes well, Call For Me points toward a future where delegating a phone call to an AI becomes as routine as asking a navigation app for directions. The open questions are the ones Google's blog post does not answer: how the people on the other end of these calls respond to an AI interlocutor, and whether the live-transcript-and-takeover model holds up as the tasks grow more complex. Today's rollout to a small beta population is the first real-world test.