%201.webp)
Baymard has published research showing that US online shoppers prefer AI for product research but want human help when orders go wrong. The study also found that AI-generated product images reduce buyer confidence for most respondents.
A survey of 2,273 US online shoppers found that 63% would rather use an AI chatbot to compare product features, 59% prefer AI for checking stock availability and 56% favour it for personalised recommendations. That pattern reversed when support shifted from research to resolving post-purchase problems.
For damaged or incorrect items, 69% said they would prefer a human representative. Another 67% wanted a person for late or missing orders, while 66% preferred human support for shipping or billing mistakes.
The findings point to a clear divide in how shoppers view automation in retail. Consumers appear willing to use AI tools when gathering information, but are less willing to rely on them when money, delivery and returns are involved.
"Shoppers have made the division of labor very clear. They're happy for AI to do the legwork of comparing products and checking stock. But the moment an order goes wrong, they want a person who can actually fix it. Retailers that trap customers in a chatbot loop during the busiest weeks of the year are putting their most valuable relationships at risk," Christian Holst, Co-founder of Baymard, said.
Retail groups have been adding AI chatbots and virtual assistants to websites and apps to handle large volumes of customer queries. The research suggests that approach may be more effective in product discovery than in customer care, particularly during periods when delivery mistakes and return requests rise.
Holst said: "Shoppers see AI as a research assistant, not a problem-solver. When a package is late or the wrong item shows up, people want someone with the authority to fix it. A chatbot that can't hand over to a human at that point doesn't save money. It turns a fixable problem into a lost customer."
Image trust
The study also examined attitudes to AI-generated product images, an area where retailers and brands have been experimenting with synthetic visuals for listings, lifestyle scenes and model shots. Here, too, respondents expressed caution.
Some 60% said knowing a product image was AI-generated would reduce their confidence in buying the item. Of that group, 25% said it would slightly reduce confidence, and 18% said it would greatly reduce confidence. A further 17% said they would not feel confident enough to buy the product at all, while 18% said it would make no difference.
The results suggest product imagery remains closely tied to trust in eCommerce transactions. Unlike marketing visuals in broader brand campaigns, shoppers often treat product photos as evidence of what will be delivered.
Holst said: "Shoppers don't treat product photos as decoration. They use them to check what they're actually going to receive: the color, the fit, the texture, the scale. An AI-generated image undermines exactly that. For one in six shoppers, it's enough to walk away from the purchase."
The issue is becoming more prominent as retailers adopt generative AI tools to create or alter imagery more cheaply and quickly. The findings indicate that while such tools may lower production costs, they can also introduce doubt at the point of purchase.
Holst said: "That doesn't mean AI has no place in product imagery. But the core product views should be authentic photography. Where AI is used, for backgrounds or lifestyle scenes, brands need to work actively to keep shoppers' trust."
Where AI fits
The broader lesson, Baymard said, is that shoppers accept AI in narrow, task-based roles where speed and convenience matter, but still expect clear escalation to people when problems carry financial or practical consequences. That distinction could shape how retailers design support systems and product pages.
The research forms part of Baymard's AI Engagement study, which covers product research, feature usage, trust, privacy, perceived accuracy, image confidence and responsibility when AI makes mistakes in shopping contexts. Baymard, founded in Copenhagen in 2009, describes itself as an independent eCommerce UX research organisation.
Holst said: "The retailers getting this right put AI inside product discovery, where shoppers welcome it, and make the route to a human obvious the moment an order goes wrong. Hiding the 'talk to a person' option behind bots and prompts is exactly the friction shoppers don't forgive."