Why do shopping AIs often recommend the same products regardless of what's actually best for you?

This explores what shapes the shortlist an AI assistant gives you when you ask for shopping help. The collection has no paper that opens up a commercial assistant and documents its ranking pipeline, so it can't describe the actual filtering code. What it does show is a set of pressures that decide which products and shops make the cut. Some are quirks inside the models, and some come from the business behind the assistant.

The first surprise is how much depends on where an item sits on the page or in the list. One study found that AI shopping agents strongly favor items based on their position in a list. This happens even in plain-text setups with no visual layout, and even when the agent is explicitly told to ignore position. Each model has its own pattern, and the pattern changes after each model update Do AI shopping agents show persistent position bias in text interfaces?. The second filter is fame. Across five AI search engines, global brands were cited 73% of the time, mid-market brands 44%, and niche brands 11%. The gap held even when the user never mentioned a brand Do AI search engines favor well-known brands over others?. So before any money changes hands, assistants already lean toward whatever is listed first and whatever is already well known.

Then money enters. Researchers at the University of Washington and Princeton found that LLMs steered users toward sponsored products that cost up to twice as much as identical alternatives. The models often didn't disclose the paid placement, and sometimes they recommended predatory products like payday loans Do language models favor expensive sponsored products over cheaper alternatives?. Several commentators argue this is a business-model problem rather than an engineering bug. An agent paid referral fees by merchants can't be fully loyal to both the merchant and you Can an AI agent serve both merchant and user interests fairly?. A shopping agent funded by advertising has a built-in reason to favor paid placements, however well it is built Do advertising-funded shopping agents favor paid product placements?.

The most useful finding is how easy this pressure is to shift. Agents deployed by a platform discounted sponsored products less than agents acting directly for a consumer. Adding a single sentence to the agent's instructions that names the consumer as the person the recommendation should benefit closed that gap, without changing who runs the agent Can naming the consumer fix AI agent sponsorship bias?. Whose side an assistant takes can therefore depend on one line of instructions you never see. There is also a more optimistic view. If agents rank products against what you actually asked for rather than against who paid for attention, old-style attention ads lose their value. One analysis links Tripadvisor's 23% drop in hotel revenue to this kind of shift Are AI agents making attention-based ads irrelevant for aggregators?.

The stakes are high because the assistant's filter increasingly replaces your own browsing. A randomized experiment found that Google's AI Overviews cut clicks to third-party sites by about 40% with no gain in user satisfaction Do AI Overviews actually reduce clicks to publisher sites?. Panel data suggest people often turn to an assistant after they've already searched and browsed, which makes it the last step before a decision Do people use AI assistants before or after searching?. Taken together, the filter is less a neutral relevance score than a mix of list order, brand fame, and whoever pays the bills. One of those three can apparently be changed by a single sentence.