Nearly half of US shoppers who bought something on an AI recommendation returned it within the past year. The cause lies in the product and fit data those models use, and apparel brands control more of that data than they think.
That finding comes from Blue Yonder's 2026 Consumer Retail Returns Survey, fielded in September 2026 across more than 5,000 consumers, including 2,000 in the US. It's the clearest read yet on what happens when shoppers act on AI advice, and for apparel brands, the numbers point in one direction.
Nearly one-quarter (24%) of American shoppers have purchased a product recommended by an AI assistant. Of those, 46% returned one in the past year. That's higher than the 41% return rate for purchases made on a social media recommendation, and 49% of shoppers who returned an AI-recommended item said they're more likely to return those purchases than items they bought directly from a retailer.
Duncan Angove, CEO of Blue Yonder, framed the cause directly in the release. He said it isn't really an AI problem, it's a data problem, and that an AI assistant is only as good as the product information it can find. Most product data, he noted, was written for a search bar rather than for a machine recommending something to a person.
For apparel, that gap has a specific shape.
The two leading reasons are that the item was different from its description (36%) and that the size or fit was wrong (23%).
In apparel, those two reasons usually describe the same failure. A shopper who says a jacket wasn't as described is often saying it didn't sit on their body the way they expected, which is a fit expectation that the product data failed to set. Sleeve length, how a size runs, whether a cut is relaxed or trim through the chest: none of that lives in a marketing description, and little of it lives in a standard size chart.
Blue Yonder's own analysis reads the split as AI struggling more with description accuracy than with fit. For apparel specifically, that reading understates the issue, because the description and the fit are describing the same garment on the same body.
The survey found that just one in five US consumers trusts AI to recommend the right size and fit.
That's the number apparel brands should sit with. Shoppers already use AI assistants to find clothing, and most don't trust the sizing advice they get. Trust that low shapes behavior before a return ever happens, because a shopper who doubts the recommendation either abandons the purchase or orders two sizes and plans to send one back.
Research from Rithum across US and UK shoppers found that more than half of consumers blame the retailer or brand when an AI recommendation contains incorrect product information. The assistant makes the error, and the brand absorbs the cost and the reputational damage.
The generational split is wide. Among US shoppers, 43% of Gen Z have bought a product on an AI recommendation, compared with 36% of millennials, 19% of Gen X and 9% of baby boomers. A recent Mastercard survey found nearly a third of teenagers trust AI-generated recommendations over a friend's.
Gen Z also returns the most of what AI recommends. More than half of Gen Z AI shoppers (56%) have returned an AI-recommended item in the past year.
The cohort adopting agentic shopping fastest is the one experiencing the fit failures most often. For brands building toward a younger customer base, that problem compounds as adoption grows.
When return rates climb, the instinct is to restrict the policy. The survey suggests that backfires.
More than half of US respondents (51%) chose one retailer over another purely because it had a better returns policy, and more than eight in ten would shop less often or stop entirely at a retailer that limited returns for returning too frequently. Nearly one-third have already stopped shopping somewhere that tightened its policy in the past year.
Free returns shipping (32%) and fast or instant refunds (30%) rank as the most important factors when shoppers return something. Returns policy has become a competitive surface that shoppers actively compare.
That leaves one place to work: preventing the return at the point of recommendation.
If product data written for a search bar is the problem, the fix is product data written for a machine. In apparel, that means more than better copywriting.
An AI assistant needs structured garment measurements at every point of measure, graded across the full size run, with tolerances and fabric behavior attached. It needs to know whether a chart describes a body or a garment. It needs consistent size-naming conventions across styles and seasons. And it needs a way to relate all of that to the specific shopper asking, rather than to an average.
That last requirement is where most sizing data stops short. A size chart tells an assistant what the garment measures. It doesn't tell the assistant how the garment will fit the person in the conversation.
Bold Metrics trained its AI models on the world's largest cumulative dataset of high-fidelity human body scans, paired with proprietary machine learning algorithms proven across thousands of fittings. Combined with garment data, purchase and return behavior, and customer body data, the system pinpoints body measurements with remarkable accuracy and generates a digital twin. The models keep improving, continuously refined using real-world purchase and return data.
What that looks like for your shopper is four to six simple questions, with no photos, no scans, and no measuring tape. From those answers, the platform determines 50+ body measurements and builds a digital twin of that specific person. Virtual Sizer™ maps the twin against your garment specs, so recommendations reflect how your jacket is actually cut. Smart Size Chart™ then shows the shopper how each size will sit across the critical points of measure, which is the fit expectation a product description was never able to set on its own.
The same data travels in the other direction. Apparel Insights® returns aggregate body data to your design and merchandising teams, so the next size run gets graded against the customers you actually have. And when a shopper works through an assistant instead of your product page, Agentic Sizing Protocol™ lets that assistant query real body data and real garment specs, returning an answer grounded in both while keeping your retailer credentials out of the agent's reach.
Across client deployments, that combination produces a 4x increase in conversion rate, a 22% increase in average order value (AOV) on recommended sizes, and an 18% average reduction in fit-related returns.
Three steps separate brands that benefit from agentic traffic from those that absorb its return costs.
Audit what an assistant can actually read. Pull your product data the way a machine would and check whether size charts are labeled as body or garment, whether points of measure are structured, and whether naming is consistent across the catalog.
Set fit expectations on the page, not just in the description. A shopper who understands how a size will fit through the chest and sleeve before ordering doesn't need to order two.
Make fit data queryable. Agentic traffic is growing from a 24% adoption base, and the brands with structured, accessible fit data will be the ones those assistants recommend with confidence.
The returns showing up from AI recommendations point to apparel data that wasn't built for this channel yet.
Ready to close the fit data gap before agentic traffic scales? Explore the Bold Metrics platform or request a demo.