The Maze: In a small US study of AI-assisted shopping, product specifications led trust more often than reviews. Researchers recorded the single main reason people trusted a selection. What mattered was whether the shopper could justify the choice. A product’s suitability, a recognisable brand and perceived value can each answer a different question at that moment.

📋 Specifications led the choices. The research recorded specifications leading in 35% of tasks, ahead of brand familiarity at 29% and price or value at 18%. These are shares of 224 shopping tasks, each assigned one leading trust reason. Taken together the three reasons account for 82% of tasks. Price or value can mean that features justify the price, rather than that shoppers pick the cheapest item.
💬 Reviews had a smaller leading role. The exhibit shows reviews or ratings leading in 2% of tasks. Reviews could still have reassured shoppers whose leading reason was suitability or familiarity. The four displayed categories add to 84%. That leaves other reasons outside the visible evidence and gives retailers little basis for cutting review programmes from these results alone.
Our take: Make product facts easy to verify, then measure whether better information improves decisions.
🏷️ Fix facts a shopper can check. The researchers describe wrong prices and mismatched sizes. Prioritise correcting the information an assistant and a product page show for the same item, then test whether shoppers choose a better fit. An unfamiliar brand can make suitability concrete by publishing dimensions, materials or compatibility in consistent fields. A controlled comparison against the previous listing can test whether that makes selection easier.
🧪 Separate selections from sales. The study recorded 158 ChatGPT tasks and 66 Google AI Mode tasks. Test understanding and completed orders separately before turning a result from these desktop sessions into a sales forecast. Forty participants completed multiple tasks across six categories. That is useful for finding failure cases and much weaker for predicting what a larger audience would buy. Track returns too if clearer facts are meant to improve fit.


