Emerging threats and opportunities in AI-mediated e-commerce

Autonomous AI agents are poised to take over the way consumers shop online, bringing both efficiency and uncertainty to digital marketplaces. A new study has found that these agents, while powerful, introduce unexpected market dynamics and raise pressing questions about competition, consumer welfare, and regulatory oversight.

Titled "What Is Your AI Agent Buying? Evaluation, Implications, and Emerging Questions for Agentic E-Commerce", the research evaluates the performance and impact of vision-language model (VLM) agents using the newly developed ACES (Agentic e-CommercE Simulator) framework. By simulating real-world shopping scenarios, the authors uncover how AI decision-making could alter the fundamental economics of online retail.

How rational are AI agents in making purchases?

The study reveals that AI shopping agents exhibit mixed levels of rationality when following consumer instructions and making purchase decisions. Tests measured whether these systems could adhere to simple rules, such as staying within budget or choosing the lowest-priced product when all options were similar. While advanced models like GPT-4.1 and Gemini 2.5 Flash performed better than older systems, inconsistencies remained. Even state-of-the-art agents occasionally made irrational choices, raising concerns about whether they can be fully trusted to act in the best interest of consumers.

This inconsistency has far-reaching implications. When consumers delegate purchasing power to autonomous systems, they expect optimal decisions. However, the study shows that irrational outcomes, though infrequent, could lead to suboptimal purchases, financial inefficiencies, and eroded trust in AI shopping tools. The researchers stress that this issue is not merely technical but also a matter of consumer protection, as users may have limited visibility into how and why their agents make certain choices.

The findings underscore that while AI agents are designed to streamline decision-making, they do not always prioritize consumer welfare in a predictable way. This raises the need for further refinement in AI decision models and potentially new frameworks to ensure they operate transparently and reliably.

What biases do AI agents exhibit in product selection?

According to the study, AI agents display human-like preferences and biases, shaping how they choose products. Agents consistently favored items with lower prices, higher ratings, and more reviews, mirroring typical consumer behavior. However, they also demonstrated strong position biases, selecting products based on their placement within the online grid rather than purely on merit.

The study uncovered that sponsored tags often had a negative effect on product choice, while platform endorsements such as “Overall Pick” dramatically boosted selection likelihood. These patterns suggest that AI agents may be susceptible to subtle platform design cues, which can manipulate purchasing decisions even in the absence of direct advertising influence.

Notably, the biases differed across models, meaning that the same product could dominate in one AI ecosystem while being overlooked in another. This heterogeneity introduces new risks for sellers and platforms, as the performance of products will increasingly depend on how different AI models interpret and weigh product attributes. Traditional advertising levers could become less effective, replaced by new forms of influence tailored specifically to AI agents.

These behavioral patterns imply a profound shift in market dynamics. Instead of consumers being swayed by ads or branding, AI-mediated shopping may concentrate demand based on how agents are programmed to prioritize factors such as endorsements, product ranking, or layout. This shift could benefit some sellers while marginalizing others, disrupting established competitive advantages.

How will agentic e-commerce change market dynamics?

The researchers also examined how the rise of AI agents will reshape seller strategies and market outcomes. The study’s simulations showed that market share distributions varied widely across different AI models, creating a scenario where certain products dominate under specific agents while struggling under others. This variability can lead to concentration risks, where small changes in AI algorithms have outsized effects on which products succeed.

Sellers face new challenges and opportunities in this landscape. The research found that even minor modifications to product descriptions could cause dramatic shifts in market share, sometimes exceeding 20 percent. This phenomenon mirrors search engine optimization (SEO) practices but in a more dynamic and unpredictable environment. As AI agents evolve, sellers may need to continually adapt their content to align with the latest algorithms, creating a cycle of optimization that could disadvantage smaller businesses with fewer resources.

For platforms, the shift raises questions about monetization and control. If AI agents bypass traditional advertising mechanisms, platforms may need to design agent-specific storefronts to maintain relevance and revenue streams. Regulators, too, will face the challenge of keeping pace with these developments. The study calls for the creation of standardized evaluation protocols to ensure that AI agents operate transparently and fairly, minimizing risks of manipulation and anti-competitive practices.

The authors warn that agentic e-commerce is not just an incremental change but a disruptive force that could redefine how markets function. With AI making purchase decisions on behalf of millions of consumers, the power to shape demand may move away from human buyers and into the algorithms governing their agents. This shift demands careful consideration from industry leaders and policymakers alike.