On 27 October, Yahoo Shopping Taiwan launched a conversational shopping assistant called AI懂買 (the name means roughly “AI that knows what to buy”). Besides typing keywords to search for products, customers can now simply chat and ask the AI to pick for them. On 11 November, in its sales report for Double 11, the 11 November shopping festival, Yahoo Shopping Taiwan said that more than 80% of the assistant’s users had asked by spelling out what they needed and their situation in a full sentence; Yahoo also said that during Double 11 the assistant had pushed the add-to-cart rate up nearly threefold. As far as we can find, this is the first time that an e-commerce platform in Taiwan has published figures on how customers actually talk to a shopping AI. Our position is that to answer a question like this, AI has to find things like the situation, the use and who a product suits in the product information, and product pages that only list specifications and keywords will struggle to get picked when people ask this way, so now is the time to act.
What happened
The assistant sits inside Yahoo Shopping Center (Yahoo購物中心) and works on both the mobile website and the app. When Yahoo Shopping Taiwan previewed it on its official blog on 19 October, it described it like this: the assistant draws on the platform’s “structured product database”, meaning product information that’s organised field by field. Once a customer describes what they need and how they’ll use it, the AI goes back and forth with them for a few rounds, narrows things down to products that fit, then matches them against product specifications and current offers and makes recommendations.
The Double 11 sales report of 11 November (Yahoo News Taiwan republished it too) uses the term “situational, medium-to-long sentences”, which in plain terms means asking not with a few keywords but with a longer sentence that puts the need and the situation together. According to the report, more than half of the conversations were about two categories: clothing, shoes and bags, and consumer electronics and home appliances.
All of these figures come from Yahoo Shopping Taiwan’s own press release, and no third party has checked them; the release also doesn’t say whose add-to-cart rate the “nearly threefold” refers to, what period it is compared against, or what it rose from and to.
Why it matters to you
With more than 80% of the assistant’s users asking in longer sentences, what customers give the AI is often a single sentence with the need and the situation together, which means more than one condition. Since the assistant picks from the product information on the platform, the AI has to be able to find each condition that a customer sets somewhere in that information, at the very least, before it can check your products against it.
Suppose you run your own canvas bag brand, with a website of your own, and your bags are also listed on Yahoo Shopping Center. For one of your tote bags, the product page gives the material, the dimensions, the colours and the price, so none of the specifications are missing. A customer opens the assistant and types: “My niece has just started university this year, and I want to get her a bag that she can carry her laptop to class in. Under NT$1,500, and something that goes with most outfits. Any recommendations?”
That sentence has at least four conditions: a gift for a girl who has just started university, room for a laptop, under NT$1,500, and easy to wear with most outfits. The only one that your page matches is the price; whether a laptop fits means working it out from the dimensions; and who the bag is for and what it goes with aren’t mentioned on the page at all. The bag was made with students going to class and commuting in mind, yet all the AI can read is a list of specifications.
momo, for one, is already adding situations alongside specifications in its own product data. According to a report by the news site CNEWS on 1 April, momo said at the 2026 iKala Connection Day event that the platform was sorting out its product data, “beyond the existing specification parameters, systematically adding application scenarios, the problems a product can solve and the benefits of using it” (in translation), so that AI can understand products correctly.
What to do about it
It’s time to act. The assistant has been live for more than three weeks, customers are already asking this way, and for any condition that your product information doesn’t cover, the assistant has nothing to check against. What you need to do is write down who each product suits, when and where it’s used and what it solves for the customer, on your product pages on platforms such as Yahoo and momo and on your own website as well. Your website needs it too because when a customer puts the same sentence to ChatGPT, Gemini or other leading AI engines, what they read online may well be your website.
A common way to get situations wrong is to write them as another set of keywords. Add “perfect as a gift or for yourself” or “goes with anything, from the office to a date” to every product, and with ten bags carrying the same line, it does nothing to help the AI pick one for that customer. What helps is something that only this bag can do and that customers actually use as a condition, such as the size of laptop that it fits or whether it has a zipped inner pocket. Which things count as conditions has to be worked out from the kind of full sentences that customers type, and how much detail each product needs can only be decided one product at a time, so with a large range, it’s a long-term job.
Yahoo has published how customers ask, and you can see how ChatGPT and Gemini answer about you in a demo. Arrange one with our consultants, and a consultant will take one of your products, first ask ChatGPT and Gemini about it with a few keywords the way people used to search, then ask again with a full sentence like that customer’s, and whether you’re still among the recommendations once the question is a full sentence becomes clear when you compare the two. The demo also brings up your website’s GEO audit score: the GEO audit is geoweb.tw’s free website check, which looks across 12 dimensions at whether AI can make sense of your website.
To fill this in for every product, what we offer is managed GEO: we rewrite the product descriptions on your website, and for your product pages on platforms such as Yahoo and momo, we work out which situations need adding to each one and hand that to you, for whoever handles your listings to update them. After that, we regularly put the kind of full sentences that customers type to ChatGPT, Gemini and other leading AI engines, save the answers and check whether your products are starting to show up in the right situations.
Further reading
- From keywords to intent: AI now picks businesses by the customer’s checklist — once customers spell out all their conditions in one go, how AI picks businesses by those conditions, and how the ones that don’t spell out the answers get skipped
- Taiwan’s momo plans an AI shopping agent for its own site — published the day before: at its earnings call, momo said it will build its own AI shopping agent for its site, without giving a launch date there
- Gemini launches Chinese-language shopping in Taiwan — outside the platforms: since 9 November, Google has let customers in Taiwan pick products by chatting with Gemini, with a spec comparison table attached