On 9 November, momo, Taiwan’s leading online retailer, said at its earnings call that it is still watching the outside AI shopping agents that can place orders for people, and that it will create one dedicated to its own site. The same week, Jeff Ku, momo’s president and also vice-chairman of a trade association for non-store retailers, put the questions of authorisation and liability when AI places orders for customers to the Minister of Digital Affairs, Yi-Jing Lin, at a forum that the association held, and Lin said in reply that the ministry would study a risk framework. Our position is that AI picking products for customers, and even placing the orders, is now moving in Taiwan from both ends, the platforms and the government. Whether a customer asks the AI inside momo or ChatGPT or Gemini, an AI can only pick you using what it can read, such as your product information and what’s written about you elsewhere, and all of that needs to be corrected now and made consistent.
What happened
The part about AI shopping agents came up when Ku was asked about them at momo’s third-quarter earnings call, and it isn’t in the slides that momo published, so what follows is based on same-day reports by the business magazine Business Next and Yahoo Taiwan’s stocks section. Ku said that momo is still watching how outside AI shopping agents develop, and Business Next reported that momo’s concerns are transaction authorisation, liability in disputes, information security and regulatory definitions. Meanwhile, momo will use the first-party data that has built up on its platform, meaning the data that it collects directly on its own platform, to develop an AI shopping agent dedicated to its own site. Business Next quoted him as saying, in translation: “Using AI to offer a dedicated agent on our own site, to make shopping easier for consumers, is something we will do.” Momo didn’t say at the call when this agent on its own site will launch. But momo said back in April that it planned to roll out an “AI shopping assistant” within one to three years, and that it had already brought generative AI search into big sales events such as Double 11.
The forum was held by Taiwan’s trade association for non-store retailers (in Chinese, 中華民國無店面零售商業同業公會). Lin and a team from the ministry’s Administration for Digital Industries attended, along with more than 30 company representatives, Ku among them as the association’s vice-chairman, as Commercial Times, Mirror Media and Anue all covered from 12 November. Ku said at the forum that once AI can place orders for consumers, or even pay, and it buys the wrong product or goes beyond what it was authorised to do, the scope of authorisation, how liability is shared and how an AI agent’s identity is verified all need to be settled early. Lin replied that the ministry would study an “AI application risk framework for the e-commerce industry”, and he asked the association for practical input. The reports don’t say what the framework will cover or when it will be put forward.
Why it matters to you
What Ku raised at the forum is who is responsible when an AI order goes wrong, which for now is something that a platform operator has put to the government and the government says it will study, with no conclusion yet; what matters more to you is momo’s line about a “dedicated” agent on its own site, which means that momo wants customers to be able to ask AI right inside its own site. When a customer asks momo’s own AI inside momo, the products that get recommended are products on momo. Momo itself said in April that how good AI search is depends more on the quality and structure of the product data itself, and that “in the era of generative AI search (GEO), the semantic quality of product information will directly affect a platform’s exposure and recommendations”; it said that it was adding things like use cases and the problems a product solves to its product data. If a condition that the customer gives isn’t on your page, an AI will struggle to confirm that you meet it.
Customers might also go somewhere else to ask. Also on 9 November, Google launched a Chinese-language shopping feature for Gemini in Taiwan, which recommends a few products and lists prices, reviews and where to buy. AI outside the platforms isn’t tied to any one platform: your website, your product pages on each platform and articles that other people have written about you can all end up in its answers.
For example, suppose you make handmade glutinous rice balls filled with red bean in Wandan, Pingtung, with a shop on momo and a website of your own. Your website says clearly that the filling is made with vegetable oil, so vegetarians can eat them too, but your momo page only gives the flavours and the weight. With the winter solstice coming up, when many families in Taiwan eat rice balls like these, a customer asks AI: “I want to buy some handmade rice balls, and some of my family are vegetarian. Any recommendations?” Put that to Gemini today, and if it reads your website, it can say that yours are fine for vegetarians; once momo’s AI shopping agent launches and a customer asks the same thing inside momo, if all it can see is your page there, it can’t answer. The same box of rice balls may get an answer in one place and none in the other.
What to do about it
Start now, and get this done while momo hasn’t announced the launch of its AI shopping agent and the ministry’s framework is still being studied, especially since ChatGPT and Gemini are already answering customers today. What you need to do is write the things that customers ask about into your product pages on momo and other platforms, and into your website, and make every place say the same thing.
The gap can be in one of two places: the page on the platform doesn’t say it, or different places say different things. These are gaps that business owners can find hard to see for themselves, because you know your products too well. When you read your own pages, your mind fills in whatever isn’t written there, while all the AI has is the words on the page.
So these gaps are best looked at by someone who doesn’t know your products. One of geoweb.tw’s consultants can take you through a demo, and as momo hasn’t announced the launch of its AI shopping agent yet, the consultant will demonstrate live with ChatGPT and Gemini, asking the kind of question that the customer above would ask about what you sell, and whether AI’s version of you matches the one you know will be plain on the screen. The same session also brings up your website’s GEO audit score: geoweb.tw’s audit is free and looks across 12 dimensions at how much of your website AI can make sense of.
A demo shows you an answer or two, whereas managed GEO starts by taking stock: how you’re described in the places that AI can read, all pulled together and set against your website. We revise the content on your website; your product pages on the platforms are still changed by whoever handles your listings, and we put together what needs changing and hand it to you. After that, we regularly put the questions that customers ask to ChatGPT, Gemini and other leading AI engines, save the answers and check whether the way they describe you has caught up.
Further reading
- AI shopping agents drop brands whose product data doesn’t add up — how an AI agent comparing prices for someone treats a product that’s described differently on your website and on each platform
- Gemini launches Chinese-language shopping in Taiwan — on the same day as momo’s earnings call, Google let customers in Taiwan ask Gemini to recommend products and list where to buy them
- Big companies are laying the rails for AI agents to pay — the ordering and payment side: how far the rails being laid by the card networks and Google have been tested in Taiwan