LDWW published two numbers about AI shopping this month, and the trade press kept the wrong one. Nearly seven in ten Americans already use AI to shop for products and services, the release reported, and 71% expect to use the technology more in the years ahead. Both figures measure exposure to AI-mediated surfaces rather than any decision to hand a purchase over, and the retailers now budgeting against them are paying for interface saturation on the assumption that they are buying intent. The distinction is not pedantic. It determines what you build.

The same release carries its own correction, reported without alarm: 30% of respondents said they had used AI to buy a product or service. Forty points separate the Americans who describe themselves as shopping with AI from those who have completed a transaction through it, and that gap is where every agentic-commerce projection currently lives. LDWW is an integrated marketing and communications agency, and its survey is competent work; the purchasing figure sits in the release in plain sight. The compression happened downstream, in the slide from “used AI while shopping” to “shops with AI,” where a tool became a delegate.

The question cannot separate the two behaviors because the interface no longer separates them either. Someone who typed a product query into Google in February 2026 got an AI Overview whether they asked for one or not; someone browsing Amazon got Rufus; someone in an Instagram thread got Meta AI. Asked afterwards whether they had used AI to shop, they say yes, and they are telling the truth. Nothing in that yes distinguishes them from a person who opened ChatGPT, described a problem, and bought what it recommended. Only one of those two behaviors is worth restructuring a merchandising organization around.

Gartner asked the question that isolates the variable, and the answer came back at nearly the same magnitude. 72% of consumers told Gartner that generative AI “appears in my internet and app use whether I asked for it or not”. That is a finding about product decisions made at Google, Amazon and Meta, not about anything a shopper decided. When a measure of saturation and a measure of adoption land within a point of each other, the simplest explanation is that they are counting the same population.

An interface nobody can switch off produces usage statistics whether or not anyone wanted it.

Willingness to hand over the decision collapses the moment the question is put directly. Gartner’s January 2026 survey of 322 US consumers found 11% willing to let AI make purchase decisions in low-stakes categories such as personal care and household supplies. That is the easiest case anyone could construct, and it still found almost nobody. Letting AI merely narrow the choice set, a far weaker ask, drew 31% for household supplies. Kate Muhl, the Gartner analyst behind the work, said consumers “are not looking to outsource shopping decisions to AI” and want to keep final decision-making control for themselves.

A second survey with a much higher usage rate produces the identical split. Exploding Topics, owned by Semrush, surveyed 1,009 US consumers and found 77.6% had used AI for shopping in the previous six months, as reported by Search Engine Land, comfortably above LDWW’s figure. Asked how much they would allow an AI to spend without approval, the most common answer was zero and the median ceiling was $50. Different wording, different sampling, different headline rate, same shape underneath: wide contact, almost no authority.

The same survey found 68.64% of respondents saying AI had directly influenced a purchase they would not otherwise have made. That is a claim about influence, not about who did the choosing, and the two are not the same thing.


The strongest case against this reading is that exposure is the mechanism, which would make the distinction irrelevant. If an AI summary is where a shopper first narrows a category, that summary shapes what gets bought whether or not anyone consciously delegated anything, and a budget aimed at seven in ten is aimed correctly. Adobe supplies the best evidence for it: AI-referred traffic to US retail sites grew 138% year over year in May 2026, a peak that arrived with no seasonal lift behind it. That traffic now converts 54% better than non-AI traffic. A year earlier it ran at roughly half the non-AI conversion rate, so the gap has not just opened, it has flipped. For the counter-argument to hold, AI-surfaced results would have to change what people buy at something like the rate at which people encounter them.

They do not, and Adobe’s own numbers explain why. The conversion premium is a selection effect measured on visitors who arrived pre-qualified, having run their comparison inside the chat window before they ever clicked through. That population is small, unusually valuable, and nothing like seven in ten Americans. eMarketer puts AI platforms at 1.5% of US retail ecommerce sales in 2026, or $20.9 billion. A channel converting that well on one and a half percent of sales has earned engineering attention and a test budget, not a line item scaled to how many people have seen an AI answer.

The cost of the error is not only misallocated money; it is building for a shopper who is not there. Gartner’s late-2025 wave, a separate survey of 846 US consumers fielded between November and December, found that among shoppers who had used AI for a recent purchase, 62% said the information it gave them ended up wasting their time. That is what a surface pushed at people rather than chosen by them tends to produce. A retailer budgeting against saturation optimizes for the shopper who is present but incidental, and under-builds for the one who matters: the small, high-intent cohort arriving from an AI answer with the comparison already finished and no patience for a product page that contradicts it. Those two shoppers want opposite things from a website, and only one of them justifies a roadmap.

Retailers are not choosing whether to fund AI commerce. The choice is which number to fund it against. If eMarketer’s path holds and AI platforms reach 9% of US online sales by 2029, the channel will be worth building for well before it is worth reorganizing around, and the merchants who spent 2026 making their catalogs legible to machines will be the ones positioned to take it. Anyone repeating the seven-in-ten figure in a budget meeting this quarter is not citing evidence about what shoppers do. They are citing evidence about how many screens now have AI on them.