Google's AI now calls stores to check stock: how to prepare

Plenty of stores have grown used to AI answering their customers. Google has flipped the roles: now the shopper has an AI that makes the calls. When the phone rings and the voice is a machine acting for a real buyer, how a store keeps its information decides whether it gets picked.

What Google shipped

On 13 November 2025 Google announced a set of AI shopping features on its official blog. The most notable is Let Google Call: when searching for a product "near me", the user taps a button and the AI calls stores on their behalf. The system runs on Duplex, the human-voice calling technology Google has used for restaurant bookings since 2018, now upgraded with Gemini models. The AI asks about stock, price and promotions, clearly identifies itself as an automated caller, and stores can opt out of receiving these calls. Results from the calls are summarised and sent to the user by text or email. It launched in the US with three categories: toys, health and beauty, and electronics.

The second feature is agentic checkout: the user picks an item in an exact size and colour, sets a budget, and the AI watches the price. When it drops within budget, Google notifies the user, confirms shipping and payment details, and completes the order with Google Pay. The first participating merchants are Wayfair, Chewy, Quince and selected Shopify stores. Everything stands on the Shopping Graph, a database of more than 50 billion product listings with around 2 billion refreshed every hour.

Why this matters even though the announcement is not new

For two years the market ran one way: businesses installed AI to answer customer calls and messages. The other direction - shoppers having an AI that makes calls for them - barely existed at consumer scale. Let Google Call is the first time a major platform has shipped that capability to ordinary users, and observers mark summer 2026, as the US rollout widens, as the point where that balance starts to flip.

The easiest thing to miss: a call from the AI is a call from a customer with real intent. The user already searched for the product and already tapped the button asking for calls. If the person answering the phone treats the automated voice as spam and hangs up, the store just removed itself from the list of results the AI sends back to the buyer - and nobody will know an order was lost.

The stock-check call happens exactly as before; only the caller changes, from the customer to the customer's AI assistant. Photo: Matthew Henry, StockSnap, CC0 licence.

Not in Vietnam yet, but the mechanism arrived first

As of August 2026, both Let Google Call and agentic checkout run only in the US. Put next to earlier milestones, though, the direction is straight: Deloitte projects AI agents will shop on behalf of Asian consumers, and Salesforce has already put storefronts inside ChatGPT through Agentforce Commerce. Three pieces of the same picture: purchase decisions are shifting from the person to the person's assistant.

And the core mechanism did not wait for the calling feature to reach Vietnam. Vietnamese customers already ask ChatGPT, Gemini and AI Overviews where to buy; those AIs answer using the store's public data - its Google profile, website, opening hours, reviews. When the party choosing the store is a machine, it does not see the nice signboard or the friendly owner. It reads data. A store whose data is wrong, missing, or split across three places with three different numbers loses in a round it never knew existed.

AI picks stores by reading stock and price data, not by looking at the signboard. Photo: Jakub Kapusnak, StockSnap, CC0 licence.

The Chạm AI view: 4 moves a Vietnamese store can make now

One, audit the Google Business Profile. Opening hours, phone number, product categories, address - correct and complete. It is the first source every AI reads, and the place an automated caller gets the number to dial. The full list of 8 tasks is in our guide to optimising Google Business Profile in the AI era.

Two, bring stock and prices into one source of truth. The stockroom says 12, the spreadsheet says 8, the fanpage says sold out - to a human customer that is annoying; to an AI it is grounds for instant elimination. How to build the single source is covered in data as the foundation for AI agents.

Three, brief whoever answers the phone. One sentence is enough: "A call that introduces itself as an automated assistant is a real customer by proxy - answer stock, price and promotions exactly as if the customer asked directly, and do not hang up." Zero cost, keeps the order.

Four, teach the website to talk to machines. Price and availability schema lets AI read the information without calling; service pages that answer questions directly get cited. Two guides: schema markup for AI and optimising service pages for AI recommendations.

Sources: Google, "New agentic checkout and AI shopping tools" - official blog, published 13 November 2025 · TECHi, updated 18 March 2026 (details on the AI identifying itself and the store opt-out) · Market commentary from August 2026 on the pace of the US rollout. The features are not available in Vietnam; the recommendations are Chạm AI's reading for the local market.

Frequently asked questions