AI Recommendations Drive 46% Return Rate, Blue Yonder Survey Finds

What happened

Nearly half of U.S. shoppers who bought an item based on an AI recommendation have returned at least one item in the past year. The figure, 46%, comes from Blue Yonder's 2026 Consumer Retail Returns Survey, which Retail Customer Experience reported from a Blue Yonder press release.

The most common reasons for returning an AI-driven purchase were that the product was different from its description and/or the size or fit was wrong. Only one in five U.S. consumers said they trust AI to recommend the right size and fit, and more than eight in 10 said they would cut back on a retailer that limits returns.

The fourth annual edition of the survey polled more than 5,000 consumers across the U.S., Australia, France, Germany and the U.K., including 2,000 Americans. Blue Yonder, which provides an AI-driven platform and multi-tier network for supply chain strategy, published the findings.

Why it matters

The story is less about whether AI can make a recommendation and more about what happens after the click. Recommendation engines now shape discovery and purchase decisions, but the product data, sizing information and descriptions behind them are often not accurate enough to support that influence. When a purchase is returned, the cost lands in logistics, customer service and margin, and the customer's confidence in the next recommendation falls with it.

As Duncan Angove, CEO of Blue Yonder, put it in the release, "Shoppers are taking AI's advice faster than retailers are getting ready for it." For leaders in experience and digital transformation, that gap between adoption and operational readiness is the real issue, and returns data is now a feedback loop that shows exactly where it sits.

By the numbers

• 46% of U.S. shoppers who bought based on an AI recommendation returned at least one item in the past year.
• One in five U.S. consumers trusts AI to recommend the right size and fit.
• More than eight in 10 would cut back at a retailer that limits its returns.
• More than 5,000 consumers polled across five markets, including 2,000 Americans.

The Renascence take

Most retailers will read this as a sizing problem and reach for a better model. That is the wrong starting point. The survey points to a trust problem created by the experience around the recommendation, not by the algorithm alone.

Shoppers are not punishing AI for being clever; they are punishing it for being confidently wrong about things they cannot check until the parcel arrives. Trust is built by reducing the cost of a mistake, which is why a restrictive returns policy works against a recommendation engine rather than protecting margin. Before tuning another model, operators should audit fit data and product descriptions and route every return reason back into the recommendation logic. Treat returns as the most honest customer research you already own.