In the AI shopping wave, are consumers becoming "outsiders" in their own decisions?

AI as decision agent — momentum and players

It has been reported that Meta is testing an AI shopping feature to compete with OpenAI’s ChatGPT and Google’s Gemini. Last year OpenAI rolled out a new shopping-capable ChatGPT guided by a shopping‑trained GPT‑5 mini; this January Google’s Gemini struck a deal with Walmart to let users search, compare and buy within the Gemini app. In China, Alibaba’s Qianwen (千问) has aggressively pushed into commerce — reportedly spending ¥3 billion in Spring Festival subsidies and driving over 130 million uses in 11 days for tasks from ordering milk tea to booking flights. McKinsey has predicted that by 2030 trillions of dollars of purchases could be routed through AI agents. So who really picks the product: us or the algorithm?

Practical limits and platform politics

AI shopping promises speed and consolidation: one ask, one answer, one click. But practical constraints are material. AI struggles with highly subjective categories such as fashion, can hallucinate price or stock details, and depends on the scope of data it can access — a fatal weakness when platforms guard their catalogs. In China’s centralized ecosystems, for example, Qianwen cannot tap Meituan (美团) data and so offers only “ecosystem‑local” bests; in the U.S., Amazon has blocked scraping and eBay tightened terms to blunt AI crawlers. Geopolitics matters too: these data fights sit atop broader U.S.–China tech rivalry and tightening rules on cross‑border data flows, shaping who gets to build the neutral shopping layer consumers want.

Trust, safety and the economics of recommendation

Trust is another battleground. Industry insiders warn that recommendation models may hide commercial incentives and amplify promoted goods via generation‑aware optimization (so‑called GEO). Data security risks are real — it has been reported that OpenAI and analytics vendor Mixpanel faced litigation after a breach — and The Information has reported OpenAI abandoned plans for full instant checkout in ChatGPT, citing fraud, synchronization and behavioral hurdles. Will consumers accept opaque algorithmic choices plus the privacy tradeoffs? Many still will: surveys find a rising share prepared to buy inside AI interfaces, but commercial viability depends on resolving fraud, data access and disclosure.

The human cost: serendipity, ownership and regret

Beyond mechanics lie psychological costs. Research from Columbia and Yale suggests AI recommendations converge on typical products and narrow discovery; Waseda psychologist Yang Fan (杨帆) warns that outsourcing choice can degrade decision quality when people process more options than they receive. The act of searching and deciding builds psychological ownership; skip that labor and goods may feel like “AI’s” rather than “mine,” increasing regret and returns. AI shopping is a mirror: it reflects efficiency gains and a test for consumers’ agency. The coming years will decide whether smart agents augment our preferences — or subtly reassign the role of chooser to the platforms that feed them.