Autonomous AI agents that make buying decisions for consumers could soon change how shopping in the online marketplace works, affecting which products succeed and how sellers and platforms respond, according to a new study released on arXiv.
The research, titled “Agentic E-Commerce: Rationality, Market Implications, and Emerging Questions for Agentic E-Commerce.” Amine Allouah and Josué Figueroa of MyCustomAI, Omar Besbes and Yash Kanoria of Columbia University, and Akshit Kumar of Yale University conducted the research. The study examines how AI agents behave in a simulated ecommerce environment called ACES (Agentic e-CommercE Simulator).
ACES pairs a general-purpose AI agent with a fully controllable mock online store, allowing researchers to test how agents respond to product features, page layout, reviews, ratings, and tags like “Sponsored” or “Overall Pick.”
How AI agents navigate the online marketplace
The study found that AI agents respond positively to higher ratings and more reviews. They also are sensitive to price, though not always in predictable ways.
Position on the page had a major impact. For example, for one agent, moving a product from the bottom-right corner to the top row increased its selection rate fivefold. Different AI models favored different positions, showing that choice behavior varies widely between agents. Sponsored tags reduced the likelihood of selection. Meanwhile, platform endorsements such as “Overall Pick” increased it.
“These findings suggest that AI agents could shift market demand and force platforms to rethink how they rank and promote products,” said Amine Allouah, lead author and MyCustomAI researcher.
The researchers tested whether agents could follow simple instructions, such as choosing a product within budget or of a specific color or brand. Even advanced models sometimes made mistakes. In controlled price tests, GPT-4.1 failed to select the lowest-priced product more than 9% of the time.
How AI agents interact with one another
Failure rates were higher for earlier models but decreased with more advanced versions. The study also looked at how sellers might adapt to AI buyers. Using an AI agent to tweak product descriptions based on buyer-agent preferences, researchers found that most description changes had negligible effects. But in 25% of cases, a single revision caused a significant increase in market share.
“The interaction between buyer agents, seller agents, and platform design could become a major factor in market outcomes,” said Omar Besbes, co-author and Columbia University professor. “Sellers may need to optimize listings for algorithmic buyers, and platforms might monetize these adjustments.”
As AI agents take on more of the shopping process, the study raises questions about competition, consumer choice, and marketplace fairness. Differences in agent behavior could concentrate demand on specific products or brands, potentially sidelining others. Platforms may need to rethink rankings, promotions, and endorsements, while sellers may increasingly use simulations like ACES to test strategies.
The research highlights a future where AI, not humans, drives online purchasing decisions, challenging existing business models, and prompting industry leaders to prepare.
Sign up
Sign up for a complimentary subscription to Digital Commerce 360 B2B News. It covers technology and business trends in the growing B2B ecommerce industry. Contact Mark Brohan, senior vice president of B2B and Market Research, at mark@digitalcommerce360.com. Follow him on Twitter @markbrohan. Follow us on LinkedIn, X (formerly Twitter), Facebook and YouTube.
Favorite