Google rolls out free Gemini task tools for calls, bills and shopping
The rollout reaches six countries. Google says purchase approvals and spending limits give users control, while acknowledging that some tasks will remain unfinished.
Loading page…
The rollout reaches six countries. Google says purchase approvals and spending limits give users control, while acknowledging that some tasks will remain unfinished.
Listen to this story
Gemini can now move from recommendations into calls, account review and purchase flows, with Google offering the new task tools at no charge across six countries. The rollout puts account access, spending limits and user approval controls at the center of delegated actions, but it is not yet a hands-off assistant for every transaction: some websites still require users to finish checkout, and advanced capabilities are starting in the U.S. Google says tasks use isolated virtual machines and safeguards against unsafe links and prompt injection.
Josh Woodward says Gemini booked a car-service appointment in under five minutes by calling two dealerships and using his calendar and saved information.
In a separate example, Gemini reviewed a utility portal and email statements; its explanation of a $389 bill is Woodward’s account, not an independent accuracy test.
Users can set spending caps, pre-approve routine actions or interrupt tasks; Google says Gemini asks for confirmation before critical actions such as buying.
Google is asking people to hand Gemini their errands—not just their questions. On November 17, the company began rolling out free tools for scheduling appointments, analyzing bills and shopping across six countries. The promise is finished work, but Google acknowledges that tasks can fail and some websites still require users to complete the final step.
The update covers the U.S., Japan, South Korea, Canada, Australia and India, using Google’s Gemini 4 Hydrogen model. Advanced features are starting in the U.S., so the announcement does not promise the same capabilities everywhere at once. Google also says Gemini now connects with more than 60 popular apps.
Google executive Josh Woodward describes an assistant that combines personal information with actions across services. In one example, he says Gemini used its knowledge of his car and home address to find nearby dealerships, called two simultaneously, and booked a service appointment around his calendar. He says the process took less than five minutes. Gemini identified itself as AI, and businesses could opt out.
Another example requires deeper account access. Woodward says he gave Gemini permission to enter his utility portal through a new Google Password Manager integration and review billing statements in email. It attributed a $389 electricity bill largely to a heat pump running during peak pricing. These are Woodward’s accounts of his own use, not independent tests of Gemini’s accuracy.
Shopping can involve either checkout assistance or a handoff. Woodward describes Gemini comparing USB-C chargers through Google’s Shopping Graph, asking about wattage and color, then helping him check out through a new Google Pay integration. In a grocery example, he texted a list and a photo of a cereal box; Gemini added the items, then sent him to Instacart to finish ordering.
While we’ve tested tasks extensively to minimize mistakes, we haven't eliminated all of them.
Josh Woodward, Google vice president, in the rollout announcement
Google says Gemini is designed to ask for confirmation or pre-approval before critical actions such as clicking Buy. Users can set a hard spending cap, and Gemini is designed to stay within it. Routine actions can receive an Always allow setting—for example, adding cereal to a grocery cart—rather than requiring approval each time. Users can also interrupt a task with further instructions.
For purchases, Gemini uses the Agent Payments Protocol. Google says it is working with Mastercard and Visa to secure supported cards through digital tokens, with American Express support coming soon. Card rewards and purchase protections still apply at checkout. A further integration with the Universal Commerce Protocol is planned, rather than part of the capabilities available now.
Google says each task runs in a hardware-isolated virtual machine dedicated to the user’s account through Google Cloud’s Agent Compute Engine. That separates its working environment from other users. Additional defenses include filters against prompt injection—attempts to redirect the assistant through content it encounters—and Google Safe Browsing checks that limit which links it can open.
Website compatibility remains a practical limit. Google says Gemini may navigate most of a site before returning control so the user can click the final button. The company is working with major websites and security providers on access for personal assistants, arguing that blocking bots by default needs a more nuanced approach.
Google’s instructions offer three ways to assign work as the rollout reaches users:
Connecting an account is not permanent: users can change task settings, disconnect Google or third-party apps, and delete chat history at any time.
Loading discussion...
Join the conversation
Explain which errands you would delegate—and which still need your approval.
Be the first to share a perspective or an experience.
Reader comments
Newest comments first. Replies stay oldest first.