AI Agents Are Taking Over Your Shopping Cart
Online shopping is on the cusp of a radical makeover. Instead of you scrolling through endless product pages, an AI agent—a kind of digital personal shopper—will soon be doing the browsing, comparing, and buying on your behalf. But what exactly will these AI shoppers choose, and why? A team of researchers from Columbia University, Yale, and MyCustomAI led by Amine Allouah and Omar Besbes has built a clever experimental playground called ACES to find out.
ACES pairs a vision-language AI agent with a mock e-commerce site, letting the researchers control everything from product placement to prices, reviews, and promotional badges. This setup is like a laboratory for AI shopping behavior, where the scientists can tweak the environment and watch how different AI models decide what to buy.
Not All AI Shoppers Are Created Equal
One of the first surprises? Different AI models don’t just pick different products—they respond differently to the same product’s position on the page. While all agents prefer items in the top row, their favorite columns vary wildly. Imagine three shoppers standing in front of the same shelf: one always grabs the leftmost item, another the middle, and the third the rightmost. This shatters the assumption that there’s a universal “best spot” on a webpage.
Moreover, these AI shoppers behave in ways that are both familiar and strange. They generally prefer cheaper products with higher ratings and more reviews—just like humans. But their sensitivity to price or ratings can differ dramatically. Some models are more price-conscious, others more swayed by star ratings. And intriguingly, all of them tend to distrust “sponsored” tags, discounting products flagged as ads, while they highly value platform endorsements like “Overall Pick.” This suggests AI agents are skeptical of blatant advertising but trust curated recommendations.
When Sellers Fight Back With AI
The study also explores what happens when sellers deploy their own AI agents to tweak product descriptions to appeal to these AI buyers. The results are striking: in about a quarter of cases, a single AI-generated edit to a product title can dramatically boost its market share—sometimes by double digits. This is the AI version of search engine optimization (SEO), but now the audience is an algorithm, not a human.
This dynamic sets the stage for a new kind of marketplace arms race, where sellers continuously optimize listings for AI shoppers, who in turn evolve their preferences. The researchers warn this could lead to market concentration, where a few products dominate simply because they’ve cracked the AI code, potentially squeezing out diversity and consumer choice.
Why This Matters for Everyone
For consumers, AI shopping agents promise to reduce the hassle of searching and comparing, but they also introduce risks. The study found that even state-of-the-art AI can make irrational choices—sometimes picking more expensive or lower-rated products without clear reason. This means trusting your AI shopper blindly could backfire.
For platforms like Amazon or Walmart, the rise of AI agents challenges traditional monetization strategies. If AI shoppers ignore sponsored ads, platforms might need to rethink how they promote products. The researchers suggest platforms could offer new services, like AI-powered listing optimization, to sellers eager to win over these digital buyers.
Regulators and policymakers also have a stake. The opaque decision-making of AI agents raises questions about market fairness, consumer protection, and transparency. Should AI shopping assistants be audited or certified? How do we prevent AI-driven market dominance or manipulation?
Looking Ahead: The AI Shopping Ecosystem
The ACES framework developed by the Columbia and Yale team offers a powerful tool to diagnose and understand AI shopping behavior. As AI agents become more sophisticated and personalized, future research will need to explore how these agents interact with human preferences, how platforms can design fair and efficient marketplaces, and how sellers can ethically compete in this new landscape.
In essence, the study reveals that when AI shops for us, it doesn’t just replicate human behavior—it rewrites the rules of the marketplace. The aisle looks different through AI eyes, and everyone from buyers to sellers to regulators will need to adapt to this brave new world.
