On 19 November, Walmart, the US retail giant, released its third-quarter results, and on the earnings call its chief executive, John Furner, said that usage of Sparky, the AI shopping assistant in Walmart’s own app, had grown 129% over last year in the US, and that Sparky baskets had expanded by 40%. A little over two weeks earlier, the general manager of PX Mart (全聯), Taiwan’s leading supermarket chain, was asked by reporters about AI agent shopping and said that adoption of such applications in Taiwan’s retail sector “won’t come that quickly”, and that PX Mart would first watch for mature success stories and full security safeguards. Our position is that on the success-story half, there are now numbers: at Walmart, customers’ use of AI to help them pick things out has more than doubled in a year. In Taiwan, Yahoo Shopping has launched an AI assistant of its own and momo says it will build one; once customers in Taiwan get used to letting AI choose on these platforms, what AI reads on your product pages is the main thing it goes on when deciding whether to pick you, so you need to move now.

What happened

Walmart launched Sparky in June last year: customers tap the “Ask Sparky” button in the Walmart app and can ask it questions directly, getting it to find products, compare options and sum up reviews, and then make recommendations. The two figures at the top come from what Furner said about Sparky on the call, and you can find them in the transcript; the 129% also appears on the earnings page of Walmart’s own website.

Both figures were published by Walmart itself. Neither the earnings page nor the call said how many people use Sparky or how much revenue it brings in, or what the 40% is being compared with.

PX Mart’s comment comes from a report by Taiwan’s Central News Agency on 2 November. The general manager was at a press event that day, and when he was interviewed afterwards, he talked about PX Mart’s plans for retail technology. Asked about the AI agent shopping service that Walmart had recently launched, he said that adoption of such applications in Taiwan’s retail sector “won’t come that quickly”, and that PX Mart would first watch for mature success stories and full security safeguards before assessing whether to adopt them.

Why it matters to you

In the same passage, Furner mentioned that Sparky had recently launched in Mexico, in Canada and at Sam’s Club in the US. Taiwan isn’t on that list, but platforms in Taiwan are making moves on the same kind of AI assistant too. On 27 October, Yahoo Shopping Taiwan launched a conversational shopping assistant called AI懂買, while at its earnings call on 9 November, momo mentioned that it would develop an AI shopping agent of its own for its site, though it didn’t say at the call when that will launch. When customers ask a platform’s AI on that platform, what it recommends are the products that are listed there.

As for the 40%, Furner’s explanation is that baskets grow because customers use Sparky to find more of the things they want and need. That is Walmart’s own reading; the figure doesn’t prove that Sparky makes customers buy more, and it could just as well be that customers who already buy more use Sparky more often. But for anyone selling things, what matters is the same either way: at Walmart, customers are using AI to help them choose more and more, and the CEO named agentic commerce as one of the three strategic areas that the company advanced this quarter.

Say you make socks in Shetou, in Changhua County, and have started your own brand of performance socks, with shops on both momo and Yahoo Shopping Taiwan and a website besides. One of your products is a hiking sock with a thickened sole and a reinforced heel, made for walking mountain trails all day, but its product page on the platforms lists only the fibre content, the sizes and “pack of 5 pairs”. A customer who is going to hike to Jiaming Lake with friends next month opens the platform’s AI assistant and asks: “It’s my first three-day, two-night hike in the mountains. Apart from hiking boots, what else do I need to buy?”

This customer never mentions socks at all. If the AI puts a pair of hiking socks on the list that it suggests, it has to choose them from a huge number of socks on the platform, and the main thing it has to go on when judging which pair is right for three straight days on mountain trails is what’s written on each page. Your page lists only the fibre content and the sizes, so the AI has little to go on when choosing you, and however much care went into your socks, the spot is unlikely to go to you.

What to do about it

Get moving now. Walmart’s usage has more than doubled in a year, Yahoo Shopping Taiwan’s AI懂買 has been live for almost a month, and customers are already asking AI on the platforms to help them choose. What you need to add for each product is the occasions where it comes in handy and what it does that ordinary versions don’t; put that on your product pages on momo, Yahoo and other platforms, and put the same wording on your website, so that whether customers ask the AI on a platform or ask ChatGPT, Gemini or other leading AI engines, they get the same picture of you.

You can’t work out which occasions to write about by searching for your product name. The most obvious way for business owners to check their products is to search by product name or category, but the customer heading to Jiaming Lake never typed the word “socks”. The questions that make AI think of socks might be about hiking or running a marathon, or about a job where you’re on your feet all day; which occasions AI puts you in and which it leaves you out of only come out when you ask in terms of the customer’s occasion.

How AI answers for these occasions is something that a demo will show you. Arrange one with our consultants, and the consultant will take what you sell, think of an occasion when a customer would need it, like the Jiaming Lake trip above, and put the whole thing to ChatGPT and Gemini; whether you’re on the lists that they come up with is right there to see. In that session, the consultant will also go through your website’s GEO audit score with you. GEO is about making sure that when AI picks things out for customers, it can make use of your information; the GEO audit on geoweb.tw is free, scores your website across 12 dimensions and shows how much AI can read from it.

Managed GEO is where we turn this into regular, routine work: we rewrite the product pages on your website ourselves; for your product pages on the platforms, we work out which occasions need adding and how to word them, then hand that to whoever handles your listings to change; and every so often, we put questions about all kinds of occasions to ChatGPT, Gemini and other leading AI engines, keep the answers one by one and see whether the number of occasions that put you on AI’s lists is growing.


Further reading