If an AI shopping assistant plays favorites by list position, does switching to text-only or giving explicit instructions actually fix it?
Do AI shopping agents' position biases persist across different interface designs?
This explores whether AI shopping agents' habit of favoring products by where they appear in a list goes away when you change how the products are shown to them (visual page, text-only list, explicit instructions), or whether it stays with the model no matter what.
This explores whether AI shopping agents' habit of picking products by where they sit in a list goes away when the presentation changes, or whether it travels with the model. The most direct evidence says it travels with the model. The ACES study removed the visual layer completely, giving agents plain-text product lists and telling them outright to ignore order. The bias stayed. It was also model-specific, so different models favored different slots, and it came back in new forms after model updates Do AI shopping agents show persistent position bias in text interfaces?. That makes position bias look less like a visual quirk and more like a fingerprint of how a particular model weighs what it reads.
Where does the fingerprint come from? One line of work argues that LLM choice biases look like human ones but don't have the same cause. People favor the first option because their attention and memory run out. Models have no such limits and show the bias anyway, which points to patterns learned from training data and reward signals Where do LLM choice biases actually come from?. That explains why redesigning the interface doesn't fix it: the cause sits in the model, not the screen. A related benchmark found agents don't just copy human biases, they magnify them. Under the same price, rating, and nudge manipulations, 17 models shifted their choices far more than humans did, snapping into rule-like behavior where people only drifted a little Do shopping agents respond more strongly to price and rating cues?.
The interface still matters, just not in the way you might expect. One study of eight commercial models found they resisted marketing tricks when every product detail was visible. They fell back on shortcuts like 'cheapest wins' only when looking up details was costly and the user's goal was vague When do pricing cues actually mislead AI shopping agents?. Put that next to ACES and you get a useful distinction. Some biases, like position, persist whatever the design. Others switch on or off depending on how much effort the interface demands. A designer can reduce the second kind but probably not the first.
The less obvious consequence is that a stable, predictable bias is something others can exploit. Simulations show that when a platform tunes its ranking and sellers separately tune their product descriptions to the same agent weaknesses, consumer harm more than doubles. Neither side has to coordinate, because both are aiming at the same predictable flaw Do platforms and sellers exploit AI shoppers together more effectively? Can platform ranking and seller descriptions separately harm AI shoppers?. Add a business model that is paid by advertisers, and whoever controls list order has a strong reason to use it Do advertising-funded shopping agents favor paid product placements?.
One caveat: the corpus has one direct test of position bias across interfaces (ACES). It doesn't systematically compare many interface designs, such as voice, image-heavy pages, or structured APIs. Human search has a parallel worth noting. Eye-tracking shows AI Overviews have taken over the 'top slot' attention that used to go to the first search result Where do searchers look when AI Overviews appear?. Whether people or agents are reading, position keeps acting as a lever, and the open question is who gets to pull it.
Sources 8 notes
ACES found that AI agents exhibit strong, model-specific position biases even in headless text-only interfaces with explicit instructions to ignore position. The bias is a deep-seated model characteristic, not a superficial visual artifact, and recurs after each model update.
LLM agents reproduce human-like decision biases without the attention or memory constraints that cause them in people. The paper argues this points to learned patterns from training and reward signals rather than shared structural reasoning constraints.
A benchmark testing 17 LLM models against humans on binary product choices found agents exhibited much larger choice shifts in response to price, rating, and nudge interventions—collapsing into rule-like patterns rather than the modest probabilistic shifts humans showed.
Eight commercial LLMs avoided marketing manipulation when all product details were visible, but fell into price-focused heuristics when attribute lookup was expensive and user goals were vague. The failure was in information retrieval, not reasoning or arithmetic.
A multi-agent simulation shows that when platforms manipulate ranking and sellers manipulate descriptions independently, their joint exploitation produces consumer harm exceeding double what each strategy causes alone. This occurs without coordination because both target identical AI cognitive biases.
Show all 8 sources
A simulation found that when a platform optimizes its ranking and sellers optimize their descriptions—each separately targeting the same agent biases—consumer harm more than doubles compared to either acting alone, without requiring coordination.
Bob Hedges, former Visa Chief Data Officer at MIT, warns that advertising business models create structural incentives for shopping agents to favor paid placements over neutral recommendations, independent of technical design quality.
Eye-tracking data shows AI Overviews receive significantly longer fixation times, reducing attention to the first-ranked result from 31% to 9%. Trust ratings between AI Overviews and ranked results remained equally high despite this attention shift.
Papers this line draws on 8
The research behind the notes this line reads — ranked by how closely each paper relates.
- Vertical tacit collusion in AI-mediated markets
- A Framework for Studying AI Agent Behavior: Evidence from Consumer Choice Experiments
- When Agents Shop for You: Role Coherence in AI-Mediated Markets
- What Is Your AI Agent Buying? Evaluation, Biases, Model Dependence, & Emerging Implications for Agentic E-Commerce
- Are Behavioral Pricing Tactics Dying? A Study of Price Endings & Partitioned Pricing Effects on Machine Buyers
- Shopping by algorithm: How agentic AI deploys human heuristics as a surrogate consumer
- Whom Do AI Agents Work For? Role Assignment Induces Sponsorship Bias in LLM Recommenders
- LLM-Agent Interactions on Markets with Information Asymmetries