Four of the biggest retailers in the country reported earnings this week, and all four spent part of the call on the same product: an AI assistant that answers shoppers' questions.
That kind of coordination usually just means a talking point. This time the independent data backs it up, minus one boundary none of the earnings calls mentioned. Consumer willingness to let AI make purchase decisions tops out at 11%, per a January 2026 Gartner survey of US consumers, while 31% will let AI narrow their choices.
Call that boundary the assistance line. Shoppers will hand AI the research, the comparison, the shortlist. They're keeping the decision for themselves.
Four earnings calls, one product
Walmart's CEO of US operations, John Furner, told analysts that the number of customers using its Sparky assistant is up 70% from last year, and that Sparky users spend 40% more per order than customers who skip it. That second number stacks two different groups of shoppers against each other, so treat it as a correlation: the heavy spenders might just be the ones who bother trying the assistant in the first place.
Target CEO Michael Fiddelke said digital traffic from external AI platforms is growing at more than 3.5 times the industry rate, then flagged in the same breath that the total is "still small in total today." Lowe's CEO Marvin Ellison reported that MyLow has fielded over 25 million questions since launch, from customers and store associates combined, and that online customers who use it convert at triple the rate of those who don't. That's a correlation too, and Ellison called it one. At Home Depot, online president Jordan Broggi answered an analyst's question with the loosest figure of the four: Magic Apron handles "millions of questions per month."
Three of those four numbers show up nowhere in the companies' official press releases. Only Target put its figure in print. The rest are call-only remarks, which is exactly where retailers tuck the figures they want analysts to hear but don't want held up against next quarter.
What this means: Shopper AI just became earnings evidence at the four biggest home and mass retailers in the country. Before you benchmark yourself against their multiples, check the definitions first. Two of the four are user-vs-non-user correlations, and one doesn't even have a denominator.
AI referrals are growing. The pie is not.
Adobe Analytics, which measured over a trillion visits to US retail sites, found traffic arriving from AI sources grew 393% year over year in the first quarter of 2026, and that this traffic converts 42% better than visits from other channels. The May reading, reported by Digital Commerce 360, still showed 138% growth on a base that's been compounding for two years.
E-commerce reached 17.1% of total US retail sales in the second quarter, per the Census Bureau. But July retail sales fell 0.6% from June, and online was the biggest single-category decliner at minus 2.2%. AI referrals kept growing all year while the total shrank in July. Put those together and it looks like share moving between retailers, not new demand showing up.
Adobe's same report shows who's positioned for the fight: the average US retail homepage scores 75% on machine readability, meaning a quarter of its content is invisible to the AI tools now steering shoppers around. The best-performing sites score 82.5%. The worst score 54.2%.
What this means: AI referral is a share fight, and the entry fee is technical. If an AI tool can't read your site, it's going to recommend the competitor it can.
The assistance line
In Gartner's January survey, willingness to let AI narrow the choices ran 31% for household supplies and 28% for electronics. Willingness to let AI make the purchase decision peaked at 11%, in the lowest-stakes categories Gartner tested. A companion Gartner survey of 846 US consumers found 54% of shoppers who used AI double-checked everything it told them, and 62% said the information wasted their time at least once.
Alchemer's 2026 Retail Report (1,002 US shoppers) found 48.5% used an AI tool to research a purchase in the past year, and 17.4% found their most recent purchase through one, ahead of in-store browsing. Yet only 35.4% mostly or completely trust AI's recommendations. Checkout.com's agentic commerce study found 27% of consumers trust no organization at all to run a shopping agent on their behalf, and puts AI agents at 3% of current transactions. Checkout.com is a payments company with a stake in agentic commerce and doesn't publish a sample size, so weight it accordingly.
Sparky, MyLow, and Magic Apron answer questions, compare products, and shorten your research. None of them buys anything for you. Four merchandising organizations that test everything settled on the same design: do the work, hand back the decision. We build to the same spec at GrowthFactor, where the agent pulls together the evidence for a site decision and the real estate committee makes the call.
What this means: The winning AI products in retail own the work, not the decision. Build for the 31% who want a shorter list, and the 11% who want a robot buyer will find you anyway, eventually.
The other side of the line
The same technology gets punished the moment it stops assisting and starts talking at people. A vendor survey from Rival Technologies of 901 Gen Z consumers across the US and Canada found 72% have taken direct action against a brand over AI-generated marketing: unfollowing, unsubscribing, or dropping the brand outright. Rival sells research software, the report's landing page doubles as a demo-request form, and Canadian respondents skewed a lot more negative (84%) than US ones (65%), so treat the 72% as a direction, not a precise US number.
The survey also can't tell apart AI that works for the shopper from AI that performs at them. An assistant that finds you the right torque wrench sits on the earning side of the line. A synthetic ad campaign sits on the other.
What this means: Treat AI assistance and AI-generated marketing as separate bets, with separate risk. One's compounding usage at the biggest retailers in the country. The other's a brand liability you should test a lot more carefully before you touch it.
-Andrew
Founding Team Member, GrowthFactor