Do all AI shopping bots cheat toward the same spot in a list, or does each one have its own favorite?

This explores whether AI shopping agents share one position bias, such as always favoring the top result, or whether each model leans its own way. The short answer from the corpus is that they are not uniform. The ACES study found that position bias is strong but **model-specific**. Different models prefer different slots. The bias also survived a plain-text interface with no visual layout, and it survived explicit instructions to ignore position. That second detail is the surprising one. You can't tell the bias away, and it changes when the model is updated. A ranking trick that works on one model this quarter may not work on the next version Do AI shopping agents show persistent position bias in text interfaces?.

There is a twist, though. Models differ in which way they lean, but they look alike in how hard they lean. A benchmark of 17 models against human shoppers found that agents swung much harder than people when price, ratings, or nudges changed. Instead of the small, probabilistic shifts humans make, the agents fell into rule-like patterns Do shopping agents respond more strongly to price and rating cues?. So the uniform part is how rigid agents are. The variable part is the particular quirk each one has. One explanation for both is that these biases are learned from training data and reward signals rather than caused by limits on attention or memory, as they are in people Where do LLM choice biases actually come from?. If that is right, models trained differently would end up with different quirks.

A related caution: some apparent differences between models may come from the shopping task, not the model. When eight commercial LLMs could see every product detail, they resisted marketing manipulation. They fell back on shortcuts only when looking up details was costly and the user's goal was vague When do pricing cues actually mislead AI shopping agents?. That means how the task is set up can shape behavior as much as which model you use.

Why does uniformity matter? Because it decides whether bias cancels out or adds up across a market. Bichler argues that when most shoppers use the same kind of agent, that agent's bias spreads into weaker competition. A mix of agents only partly offsets this Does agent diversity actually improve market competition?. A simulation shows the danger of a shared blind spot. When a platform tunes its rankings and sellers tune their descriptions to exploit the same agent bias, consumer harm more than doubles, with no coordination needed Do platforms and sellers exploit AI shoppers together more effectively?. Business incentives add pressure too. An agent funded by advertising has a built-in reason to tilt toward paid placements Do advertising-funded shopping agents favor paid product placements?.

The gap in the corpus: it shows that position biases differ by model and persist, but it has no side-by-side map of which models favor which slots. The useful takeaway is that model diversity is a weak defense. Every model is rigid in its own way, so a market dominated by one or two agents gives sellers and platforms a single, stable target to exploit.