People say they'd trust an AI to shop for them — but do surveys about trust actually predict what people do when the AI can really spend their money?

This explores whether what people tell pollsters about trusting AI shopping agents matches what they do once an agent can actually spend their money. First, a direct limit: the collection has no study that follows the same people from a survey answer to a real purchase. It does have enough from both sides to show where the two come apart. Read closely, the surveys themselves already hint that stated trust is weaker than the headline numbers suggest. Accenture found 74% of people would trust an AI agent with purchases, but 56% would still specify the brand and many loyal shoppers wouldn't switch at all Do consumers actually want to delegate shopping decisions to AI?. So the trust is real but comes with conditions. Visa's poll shows the same gap from another angle: 72% of US adults use AI assistants, but only 23% trust them to make payments. That number rises to 61% when a familiar payments brand is in the middle Does naming a payments brand increase trust in AI transactions?. RTB House found a small group of 'super-trusters' who would let AI buy with no return protection How much do shoppers trust AI shopping assistants?. That suggests full hands-off delegation is a niche, even among people who say AI understands them better than their friends do.

The behavioral studies point to something surveys mostly don't ask about: whether a mistake can be undone. In a controlled study of people using a general-purpose agent, trust didn't fall because the stakes were high. It fell sharply when an action was irreversible and visible to others, like sending an email, even when people rated the agent's work as fine What makes people distrust AI agents they delegate to?. Buying something is exactly that kind of action. Someone can honestly tell a pollster 'yes, I trust it' and still pull back at checkout, because the question never asked them to picture a purchase they can't take back. Studies that use real ad click-throughs support this. People adopt high-autonomy shopping assistants because of concrete benefits, especially when products are scarce, not because they feel generally confident about AI What drives shoppers to accept autonomous AI assistants?. Context matters too. Comparing fewer products through a voice assistant made shoppers more satisfied and more willing to hand over control, the opposite of the usual pattern Does limiting product choices increase trust in voice assistants?.

A broader pattern across the collection: what people do is a better guide than what they say about themselves. In search interfaces, users trusted answers with more citations even when the citations were irrelevant Do users trust citations more when there are simply more of them?. Trust in practice runs on shortcuts that people wouldn't report if you asked. In agent-to-agent shopping, a seller agent worked out a buyer's real budget almost perfectly from how the buyer talked. When the buyer only had a stated number kept private, the seller's estimate was far weaker Can a seller infer what a buyer really wants to pay?. And AI personas built from real behavior data predicted which version would win an A/B test 75–90% of the time Can behavior-based personas predict A/B test outcomes?. The common thread is that behavior carries signal that self-reports lose.

The less obvious point is that stated trust may not be a fixed thing waiting to be measured. Bilgihan and colleagues argue that agents that actually make purchases shape the consumer's preferences over repeated conversations, rather than just carrying them out Do AI shopping agents reshape what consumers actually want?. Longitudinal work on personalized chatbots finds that each good interaction builds trust and also raises expectations, so later failures hurt more Does chatbot personalization build trust or expose privacy risks?. A one-time survey captures trust before any of that has happened. A better question than 'do surveys predict behavior?' may be 'what does a person's trust look like after their first purchase they couldn't undo?' The collection suggests that answer is what matters, and that surveys aren't built to capture it.