The behavioral science of agentic commerce: Why trust is the missing link

This Insights series explores agentic commerce – a transformative shift where AI agents take the lead in shopping, payments and customer decision-making.

Agentic commerce won't amount to anything unless consumers actually start using it.

Rapid advances in generative AI technology, coupled with easy access to tools like ChatGPT, have triggered a widespread outsourcing of thought to chatbots and digital assistants. People are becoming increasingly comfortable using these tools for conversational search and ideation, everyday life hacks, boosting productivity at work – and even as personal therapists.

Recent data underscores the rapid consumer adoption of GenAI. As of July 2025, ChatGPT had more than 800 million weekly active users and processed about 2.5 billion prompts a day.

Integrating commerce into these interfaces is therefore thoroughly logical. After all, these platforms have enormous, highly engaged audiences – the majority of whom have regular shopping needs. Why wouldn't they seize the opportunity to add shopping to the long list of tasks they already delegate to their AI assistants?

The core challenge here is that agentic commerce represents a step change in trust that consumers need to place in their large language models. Where before, the worst-case scenario might have been bad advice, a poorly written email or a questionable recipe, the outcome of an unfavorable purchase is an unwanted charge on your bank statement. This is outsourcing of action – and the jury is out on how comfortable consumers can get with it.

"This is outsourcing of action – and the jury is out on how comfortable consumers can get with it."

As powerful as technology is, purchases involve nuance, emotion, impulse, personal risk and social context, which AI cannot necessarily model comprehensively. Moreover, the prevalence of AI errors (or hallucinations) is well-recognized by consumers. As many as 96% of internet users are aware of them, with 86% having experienced them personally according to a recent study from Tidio. The demand for trust in commerce use cases underscores the acute need for frontier models to minimize hallucination rates.

Many would argue that a lot of people actually quite enjoy shopping. Browsing through hundreds of search results lends a degree of satisfaction to the product you eventually land on – somewhat reminiscent of the IKEA effect which posits that people place higher value on products they have created or assembled themselves. Prospect theory and loss aversion may also play a role. Consumers might irrationally outweigh the risk of suboptimal AI outcomes and prefer to default to 100% certainty in their own decision.

There's an interesting parallel to be drawn here between agentic commerce and voice commerce. The concept of voice commerce emerged in the mid-2010s during the smart home boom, during which consumers were rapidly embracing smart speakers and home automation for menial tasks like setting timers, creating grocery lists, switching on lights and adjusting thermostats. The broader vision, though, was not just to sell a lot of devices but to monetize consumer interactions with those devices – and since running ads through them was quickly deemed too intrusive, this meant investing in driving voice-activated purchasing journeys. It was a compelling idea, and one that generated substantial industry buzz at the time, with many hailing it as "revolutionary" and "the next big thing.” Ultimately, though, adoption fell well short of expectations – and behavioral scientists have pointed to the fundamental discomfort that manifests when purchases are made without the gratification of visual aids and confirmation as the primary constraint.

A more optimistic analogy is to consider the behavioral upheaval that occurred after the invention of the internet and the advent of e-commerce. At the time, the notion of entering your card details into a web page then having your product arrive at your doorstep was utterly fantastical, and e-com had to battle against deeply entrenched behavioral norms and pervasive trust concerns in order to gain traction.

I believe that we are now in the throes of an equivalent behavioral transformation with AI agents: If e-com managed it, why can't a-com?

"I believe that we are now in the throes of an equivalent behavioral transformation with AI agents: If e-com managed it, why can't a-com?"

If there's one thing we learned from the e-com revolution, however, it's that consumers are unreceptive to sudden, jarring changes in their shopping journeys. That's why, even today, e-com is littered with skeuomorphic attributes: You add your items to a cart, you proceed to the online checkout and you pay with your digital wallet. It's also why more radical innovations like Jet.com's dynamic pricing, Kactoos' group buying model, and Wanelo's "Swipe to Shop" feature fell flat with consumers and were ultimately phased out.

In summary, if agentic commerce is going to succeed, trust and familiarity must be deeply embedded in the user journey – in the agent's ability both to discover the best products for you, and to buy them for you securely and reliably.

That's why, at Worldpay, we're committed to delivering the trust infrastructure our ecosystem needs – empowering consumers to embrace Al-powered shopping with confidence, knowing that safety and transparency are built into every transaction.

Previously in this series:

Part 5: Who do consumers trust with their data?

AI agents are shopping for us. But trust is deciding who gets to power the purchase.

Coming up next:

Part 7: Where agentic commerce goes next

Consumers are already embracing AI agents to shop, book travel, manage subscriptions and more—offering a clear signal of where agentic commerce is headed across retail, travel, gaming and other key sectors.