When an AI Calls Your Shop to Ask About Stock

The shop phone rings, and the caller is not a person. The request is "do you have this item in stock," and the one asking is not the customer but an AI calling on the customer's behalf.

Google offers a feature that works exactly like this. When someone searches for a product "near me," an option called "Let Google Call" appears. Choosing it prompts Google to ask a few follow-up questions, then call nearby shops and gather the information. Once it has finished, it summarises what it found and sends it to the user by email or text.

For now, the feature is limited to selected categories in the United States — toys, health and beauty, electronics and similar. Shops in the Philippines are not yet receiving calls from an AI. Even so, designing how you receive them is slow work if you start only once they arrive. This case study reads the development from one angle: how to decide, in advance, what your phone answers.

Part 1: Reading the News, Then Asking What It Means For Us

Start with the facts.

Google has rolled out agentic features that include completing purchases and calling shops on a shopper's behalf. TechCrunch reported on 13 November 2025 that Google had expanded AI shopping with conversational search, agentic checkout, and an AI that calls stores for you.

The flow works like this. A user searches for a product "near me," and a "Let Google Call" option appears in the results. Selecting it prompts Google to ask follow-up questions about the search, then place calls to retailers and shops to gather information on the user's behalf. When the agent finishes the call, it summarises what it collected and delivers it by email or text message.

Here is what has been reported, organised:

The voice layer is handled by Google's Duplex, which checks nearby shops for information such as stock, price and current promotions.

The scope is narrow. The calling feature launched in selected US categories, with toys, health and beauty, and electronics among those named. It is designed for the kind of narrow shopping errand where you would ring a hardware shop to ask whether they have a particular part.

What is happening, then, is that "an AI calls a shop on a person's behalf" now exists as a shipped feature rather than an experiment.

Related: What Are AI Agents? A Five-Minute Guide for Philippine Business Owners, With Real Project Examples explains this in detail.

Part 2: Key Terms Explained (For Decision Makers)

A few terms make this easier to judge.

What "agentic" actually means A system that takes an instruction and then carries out multiple steps by itself. In this case, that means organising the questions, selecting shops, placing the call, listening to the answers, and summarising and sending the result — as one continuous action, without a person approving each stage.

Being on the receiving side AI is almost always discussed from the position of the user. This is the opposite case. Your company is not the one placing the call; it is the one answering. That is where the design work sits.

A phone call becoming a record Whatever your staff member says on the phone is summarised and delivered to the customer. One off-hand remark becomes your published product information. The way a call is retained is simply different from before.

Narrow launches widen This feature launched in selected US categories. That pattern is not unusual: it starts narrow, and once it works, it spreads. Whether it reaches the Philippines is unknown, and if it does, there is no guarantee of a long lead time.

Related: How AI Smart Search Helps Philippine Online Stores Improve Customer Experience explains this in detail.

Part 3: Applying This To Your Own Operation

For a Japanese-affiliated company operating in the Philippines, three things are worth deciding now.

The first is whether what you say on the phone is accurate.

Even with a human caller, the answer to "do you have this in stock" depends on the memory and judgement of whoever picked up. In most operations, the inventory system and the phone are not connected. "Probably, shall I check?" works with a person. It works poorly with a caller that records and summarises.

What you say on the phone can be reduced to four items:

This preparation is useful regardless of AI. Put the answers to your most common questions on a single sheet as plain fact. Decide those four points — opening hours, categories carried, how stock is confirmed, and whether price may be quoted — and everyone who picks up gives the same answer.

The second is who answers the phone.

In Philippine subsidiaries, calls often funnel to general affairs or reception. Asked about stock, they cannot answer on the spot. Until now, "our specialist will call you back" was a workable reply. When an AI is calling on someone's behalf, the party to call back is not the customer but the agent, and whether a callback even completes depends on how that system is built.

So you need to decide one of two things: widen the range of what can be answered immediately, or state clearly what cannot be answered. The outcome to avoid is a vague reply being summarised and delivered to a customer.

The third is tidying the information you publish externally.

Before the AI calls, it finds the shop through search. That means the opening hours, address, phone number and product range on your map listing and shop pages have to be correct first. If those are stale, either the call never comes, or it arrives on a wrong premise.

When I ran an SEO business in Japan in the 2000s, I spent an hour a day checking search rankings and a full day each month producing reports. I restructured the work, set aside two hours each morning for actual improvements, and cut the daily workload to a third. The ranking-check automation itself, however, lost accuracy every time a search engine changed its specification, and in the end I went back to checking by hand. The cause was that I had not built it to be repaired when outside rules changed. Assume the external system will change, and assign an owner and a frequency for keeping your own information current — it costs less in the long run.

There are also conditions specific to the Philippines: phone conversations mix English and Tagalog, and connection quality varies by location and time of day. Any system that depends on voice is directly exposed to both.

I helped a Japanese-run restaurant near Little Tokyo in Makati, with about 80 seats, adopt AI. We rebuilt an outdated website in WordPress, targeted small keywords with current SEO and GEO practice, and used AI in social media marketing centred on Facebook. We added an AI chatbot to the site so simple enquiries were handled automatically, and set the menu up so it could be rewritten easily from a PC. What made the difference was letting the shop's own staff update the menu. A system that requires an external party to change anything stops being used the moment things get busy. The same applies to phone answers: unless the people on the floor can correct it themselves, the old information keeps being given out.

Related: Responsive Design and AI Search: What a Philippine Company Website Actually Needs Now explains this in detail.

Part 4: Common Failure Patterns

Mistake 1: Concluding it is irrelevant because it is US-only It is true that the feature is currently limited to selected US categories. But everything this article recommends is useful whether or not the AI arrives. Putting your phone answers on one sheet and keeping your map listing current both pay off today. Judging that the feature itself is irrelevant is fine; stopping the preparation is not necessary.

Mistake 2: Designing a way to refuse AI calls first Starting from refusal turns the discussion technical and it stalls without a conclusion. Decide first the part that does not change whether the caller is human or machine — namely, what you answer. Whether to refuse can be decided when it actually arrives, against whatever system exists then.

Mistake 3: Making the phone manual longer The common reaction is to thicken the script book. The more pages it has, the less it is read on the floor. Keep it to one sheet. The four items — hours, products carried, how stock is confirmed, and whether price may be quoted — are enough.

Mistake 4: Not assigning an owner for updates Map listings and shop pages go stale if left alone. Changed hours, a new phone number, an added product line — all of it sits there unupdated. Assign one person to spend five minutes on it once a month.

Mistake 5: Deciding without asking the floor Local staff are the ones actually answering the phone, and they are the ones who know what gets asked and where it gets difficult. A script written by Japanese expatriate staff alone will not match the questions that actually come. Start by having the floor note down the questions they receive for one week.

Mistake 6: Deciding nothing about call records If an AI summarises and delivers what was said, then the call has a lasting form. You need your own side to be able to establish what was answered. Full call recording is not required, but if you cannot check afterwards what was said, you have nothing to point to when a misunderstanding arises.

Three Practical Tips

Tip 1: Put the four common questions on one sheet Opening hours, product categories carried, how stock is confirmed, and whether price may be quoted. Put those four on a single sheet next to the phone. Whether the caller is human or AI, the answers will match. It takes less than an hour to produce.

Tip 2: Log incoming questions for one week only Ask the floor to write down the questions they receive on the phone for one week. You will learn what is actually being asked rather than what you assumed. That single week is the only reliable material for writing a script.

Tip 3: Set a date for checking your published information Check three places: map listings, shop pages, and pinned information on social media. Set one day a month to confirm the opening hours and phone number in all three. It takes five minutes. Regardless of whether an AI ever calls, this check determines whether customers can find you at all.

Bonus: How to Make Use of PH AI Works

At the stage of working out how far your phone and chat enquiries can be organised, an outside view sometimes helps. PH AI Works offers a 30-minute free consultation for Japanese-affiliated companies in the Philippines.

We will work through it with you against your actual operation: identifying the questions you are most often asked, building the one-sheet that keeps answers consistent, setting up a routine for checking map listings and shop pages, and deciding the scope if you are considering an AI chatbot. It is fine to talk at a stage where you have not decided to adopt anything.

Sources & References

• TechCrunch, "Google expands AI shopping with conversational search, agentic checkout, and an AI that calls stores for you" (13 November 2025; announcement details): https://techcrunch.com/2025/11/13/google-expands-ai-shopping-with-conversational-search-agentic-checkout-and-an-ai-that-calls-stores-for-you
• Sourcing Journal (WWD), "Google Leverages AI Voice Agents, Buy For Me Function to Bring Agentic Commerce to Consumers" (how Let Google Call works and its target categories): https://wwd.com/sourcing-journal/industry-news/google-artificial-intelligence-agentic-commerce-buy-for-me-price-tracking-1238859307/