Google Trials Gemini’s ‘Call for Me’—AI Makes Real Calls on Your Behalf

Preface

Google is testing a new capability that lets its Gemini AI place phone calls to businesses on behalf of users. This experiment, named "Call for Me," expands on years of AI-driven calling features by giving Gemini permission to share approved personal details and perform a wider range of tasks during live calls. The initial rollout is limited — available to Pixel 11 owners in the U.S. who subscribe to Gemini and use the beta Phone app — because Google wants to refine the system in real-world settings before a broader release. The feature aims to blend autonomy with user control, allowing people to monitor and interrupt calls in progress.

Lazy bag

Gemini’s "Call for Me" lets the AI place calls using your phone number, navigate menus, and perform tasks like reservations or checking stock. Users can view a live transcript, follow the interaction, and take over anytime. Initially experimental and limited to specific devices and subscribers, it builds on Google’s prior AI calling tools while introducing broader capabilities and tighter user control.

Main Body

Google is advancing its voice-assistant experiments by introducing a new Gemini feature called "Call for Me." This function enables the AI to place phone calls to businesses on behalf of a user, using the user’s own phone number and acting with approved personal details. The aim is to let the assistant handle more complex, real-world tasks that previously required direct human intervention.

The initial deployment is deliberately narrow: it’s available only to Pixel 11 owners in the United States who subscribe to Gemini and install the beta version of Google’s Phone app on Android. Google framed the launch as an experiment, acknowledging that human conversations are nuanced and that incremental testing will help the company refine the system. By keeping the rollout small, Google can monitor outcomes, collect feedback, and improve the assistant’s conversational judgment before offering it more broadly.

Historically, Google has offered several AI-assisted calling capabilities. Early demonstrations—most notably at Google I/O—showcased an assistant making a salon appointment with lifelike speech patterns. More recent features let Pixel users avoid hold time with "Hold for Me," ask the assistant to seek information with "Ask for Me," or display IVR menu options with "Direct My Call." These tools have been helpful, but they were limited in what they could authorize or disclose on a user’s behalf.

"Call for Me" represents an evolution: it can share personal information that the user explicitly approves and then execute a wider set of actions. For example, the AI can call a store to check whether an item is in stock, make a restaurant reservation for a specific time, reschedule appointments to another date, or place items on hold. It accomplishes these tasks by introducing itself, navigating automated phone menus, waiting on hold when needed, and carrying on the conversation with the person at the other end.

Crucially, the feature is designed with user oversight in mind. Call activity is visible in real time: users can follow along via a live transcript that displays the conversation as it unfolds. If users prefer to take control at any moment—whether to correct details, answer a question, or conclude the call—they can interrupt and assume the call directly. Calls are dialed from the user’s own number, preserving continuity and ensuring recipients see a familiar caller ID.

From a technical perspective, implementing such a feature requires addressing multiple challenges: accurate speech recognition in varied acoustic environments, appropriate handling of sensitive personal data, robust navigation of IVR systems, and the ability to respond to unpredictable conversational turns. Google’s phased approach helps mitigate these issues by letting engineers observe real interactions, gather error cases, and tune the assistant’s behavior before expanding availability.

Privacy and consent are central concerns. Google emphasizes that the assistant will only share personal details that users have approved beforehand. This limits the scope of information disclosed and gives users control over what the assistant can communicate. Still, real-world deployments will require ongoing attention to consent flows, data retention policies, and transparent controls so users understand what the assistant is permitted to do on their behalf.

There are clear user benefits: delegating routine calls can save time and reduce friction, particularly for tasks like appointment scheduling, product checks, or hold-time navigation. For businesses, interacting with an AI agent could change how customer service lines operate—agents may need processes that account for automated callers and their behavior. Google’s transcript feature could also help users retain records of conversations, which is useful for confirmations or dispute resolution.

Yet there are open questions about social norms and the design of polite, clear AI speech. Earlier demonstrations attempted to make AI speech more natural by adding human-like hesitations; whether that approach remains appropriate is debatable. Critics have raised concerns about AI agents posing as humans without disclosure; Google’s public messaging around this feature stresses transparency and user control, but real-world practice will determine how consistently those principles are followed.

In short, "Call for Me" is a measured step toward more capable, user-supervised AI agents that can interact with businesses over the phone. By using the user’s number, allowing live monitoring and takeover, and limiting the rollout to a controlled group, Google is balancing innovation with caution. The experiment will likely reveal important lessons about conversational accuracy, consent mechanics, and the practical value of delegating voice tasks to AI—insights that will shape future iterations and potential broader deployment.

Key Insights Table