Will Consumers Let AI Buy Things for Them? The Data Is In.

I have a recurring thought when I read agentic commerce adoption statistics: the people writing the press releases and the people reading the research reports are probably not the same as the people being surveyed. The tech-adjacent audience that follows this space is already comfortable delegating purchases to AI, or at least comfortable in theory. The actual consumer population is, reasonably, more measured about handing their credit card details to a system they don't fully understand.

So the trust data is useful precisely because it's not a tech audience survey. Bain's research on agentic commerce and consumer trust, Checkout.com's December 2025 study, and Quad and The Harris Poll's April 2026 survey are trying to capture where a general consumer population actually sits on this, and what they found is more nuanced than the headline numbers suggest.

The Numbers

Only 39% of Americans trust AI agents to make everyday purchases on their behalf. Only 34% are comfortable with AI-driven purchasing for larger items. Three-quarters say they would trust AI shopping results less if they discovered those results were sponsored. And while 72% of US consumers have used AI in some form, only 24% would be comfortable using it to make a purchase today.

If you stop there, it sounds like agentic commerce has a serious consumer adoption problem.

But then: 47% say they would use an AI agent for boring or repetitive purchases, a figure that rises sharply among younger shoppers. In the UK, a separate Checkout.com survey puts the equivalent figure at 40% of Brits willing to let AI agents handle routine shopping tasks. When asked what would make them trust AI features involving money, a Bidease survey of US consumers found: 48% want the ability to review and confirm before anything happens, 44% want easy access to settings or opt-out options, 41% need to know they can override or undo decisions.

These are not rejection signals. They're design requirements.

The Category Dependency

The most important nuance in the trust data is that willingness to delegate is highly category-specific, and follows a logic that makes intuitive sense.

Routine replenishment — printer ink, coffee pods, washing tablets, pet food — is where consumer comfort is highest. These are purchases where the only real variable is price and availability, the decision criteria are simple, and getting it slightly wrong is a minor inconvenience. "AI agent noticed you're running low on printer ink and reordered the same brand at the same retailer" is a proposition consumers understand and largely accept.

Fashion and high-consideration purchases are where comfort drops sharply. These involve subjective preference, fit, and context that's difficult to specify precisely, and the consequences of getting it wrong (a dress that doesn't work, a piece of furniture that's the wrong size) are considerably more annoying. "AI agent bought you an outfit for your anniversary dinner" is a proposition that makes most people genuinely nervous.

Electronics and considered goods sit in the middle: rational criteria, but enough complexity and enough consequence for errors that most consumers want to stay involved in the decision.

This is, more or less, the pattern of early e-commerce adoption. Tickets, books, and commodity goods went online first. Fashion took longer because trust in the product-to-image match needed to build. Furniture and big-ticket items took longer still. Agentic commerce is likely to follow a similar progression, except compressed by the rate at which AI capability is demonstrably improving.

The Transparency Problem Is Real

The sponsored results finding deserves particular attention. 75% of Americans say they would trust AI shopping results less if those results were sponsored, per the Quad/Harris Poll survey. That's a very large number, and it has direct implications for the commercial models that AI platforms are currently developing.

The tension here is structural. AI platforms have built significant distribution through trust: consumers use ChatGPT, Perplexity, and Google AI Mode because they believe the responses are optimising for their interests, not for advertiser revenue. The moment a consumer learns that a recommendation was influenced by a commercial relationship, the value proposition of using an AI for discovery is undermined.

This is a version of the search advertising problem, but sharper. People accepted that Google showed ads alongside organic results, because the separation was clear. An AI that presents sponsored recommendations as if they're genuine recommendations is doing something categorically different. Whether AI platforms find a model that maintains consumer trust while generating advertising revenue is one of the genuinely interesting open questions in agentic commerce, and the consumer data suggests there isn't much room for error.

What Closes the Trust Gap

The research is fairly consistent about the mechanisms that build trust in agentic purchasing contexts. Explicit confirmation before purchase is essentially non-negotiable for anything beyond ultra-routine reordering. Transparency about why a recommendation was made. Easy access to history and easy reversal of decisions. Brand familiarity: consumers are more comfortable with AI agents from retailers they already trust than from AI platforms they don't have a prior relationship with.

That last point has implications for how retailers should be thinking about AI agent strategy. The Walmart Sparky embedded-in-platform approach, where the customer's AI relationship is with a brand they recognise rather than a generic AI platform, probably has a trust advantage over the pure third-party agent model, even if it requires more investment to build.

The agentic checkout thesis isn't wrong. The willingness to delegate is there, for the right categories, with the right UX design and the right transparency. What isn't there yet is the design pattern and the trust-building track record that makes that willingness durable. Those take time, and the consumer data suggests the industry is at the beginning of that process, not the middle of it.

Which is probably fine. The infrastructure built over the last 18 months isn't going anywhere. The consumer habit formation is the long pole now.