In an Age of AI Shopping Assistants, Conventional Wisdom Prevails: 'Buyer Beware'

Artificial intelligence is steadily creeping into every aspect of our daily lives, including how we shop, Amazon, Alibaba's Taobao, TikTok and other online retail sites all offer their own AI shopping assistants, and early numbers suggest the agents are pretty adept at peddling goods.

According to Amazon's own data, shoppers who click a "sponsored prompt" inside Alexa for Shopping end up buying something 48 percent more often and spend 21 percent more on average than those who do not. Roughly half of Walmart's app users have interacted with its AI assistant Sparky, and according to JD.com, its AI customer-service agent Jingxiaozhi 5.0 offered advice 160 million times during the 2025 Singles Day shopping festival alone, reducing the need for human follow-up by over 28 percent.

So, these assistants clearly help brands sell more products. But do they really help consumers make more informed choices?

To answer that question, follow the money. AI assistants require substantial computing power to run, and that power comes at a cost. Who pays for it? Consumers? Brands? The platforms themselves?

In most cases, it is the brands. That is a rational business model for the platforms because consumers are highly unlikely to be willing to pay extra for an AI shopping assistant. Thus, brands pay to have their products to be featured, giving the platform advertising income, while every product sold generates an additional commission.

As a result, click-through and conversion rates have become the two most important measures of an AI shopping assistant's performance. What's more, influence can also help brands sell more indirectly. A new marketing discipline known as "generative engine optimization" has emerged to help brands appear in AI-generated recommendations.

When an AI assistant recommends only four to six products per query, failing to make that shortlist is close to being invisible. In China, industry reports value this nascent market at more than 4 billion yuan (US$600 million).

What do consumers gain in the process? Traditionally, online shoppers have relied on search boxes or recommendation feeds. Such advice is typically one-way and non-interactional. Whether you find something you like depends entirely on platform algorithms. Search requires keywords, usually no more than 10 words, and shoppers often have to try several variations before getting useful results.

An AI shopping assistant, on the other hand, offers a fundamentally different entry point – a conversation. Consumers can describe what they want in long, detailed sentences, giving the platform a far clearer picture of their needs and then allowing it to search vast product catalogs for the closest possible matches.

This raises an obvious question: Are products recommended truly the best choices or is there an advertising bias built in? For independent AI platforms, the picture looks different. OpenAI describes product results in ChatGPT Shopping as "organic and unsponsored," ranked by relevance rather than ad spending, while Perplexity says advertisers cannot buy their way into its recommendations. OpenAI began testing clearly labeled advertising in ChatGPT earlier this year – a reminder that such independence is a business choice, not a permanent guarantee.

AI assistants built into e-commerce platforms do carry sponsored recommendations, making many consumers leery about trusting AI assistants with purchase decisions. A survey released by the Shanghai Consumer Council in June, based on 4,308 responses collected through the council's WeChat account, found that while 85 percent of respondents said they had tried AI-powered product selection features, only 16 percent said they believed the recommendations accurately matched their needs.

Notably, a survey of more than 10,000 consumers by China's National Business Daily ahead of this year's mid-year shopping festival found that over 70 percent of respondents believe AI's greatest value lies in optimizing supply chains and making goods cheaper, while only 16 percent wanted AI to handle shopping tasks like price comparison and product selection. In the same survey, 48 percent said AI's role should be to save time and effort, not to take away the fun of shopping or eliminate the consumer's final say.

Consumers need a new kind of skill – call it AI literacy. Consumers should always ask who is paying. If an assistant or shopping tool discloses a commission or sponsorship, that information should factor into how the recommendation is judged. And any prices quoted by an AI assistant are best treated as unverified until checked against a retailer's own page.

One way to make AI shopping assistants more trustworthy is to build verification directly into them. Shanghai-based Dewu, for instance, uses an AI-powered authentication system to verify products before sale, and its overall rate of returned goods of around 5 percent stands far below the 35 percent typical of livestream e-commerce. When every AI judgment is traceable, consumers respond with trust.

In China, a large number of shoppers continue to turn to social media – Xiaohongshu (RedNote), for example – to read other consumers' first-hand experiences before making a decision. That habit may be another pointer to the future. If social media data can be effectively combined with AI shopping assistants, reviews and recommendations from real people might increase consumer trust. Until then, the wise consumer is the one who keeps asking an AI assistant a very human question: whose interests are you really working for?

(The author is an assistant research fellow at the Institute of Information Science, Shanghai Academy of Social Sciences.)