More shoppers are letting AI suggest what to buy — and many are unhappy with the results. A SmartCustomer survey of nearly 1,200 U.S. consumers released Sept. 23 found that 76% used AI to make purchases in the past year, while 89% said they don’t fully trust AI recommendations. That gap — widespread adoption paired with persistent distrust — is now producing measurable buyer’s remorse and new reputational risk for retailers and platforms.

Survey highlights: where AI recommendations fail

One in three respondents said they made at least one purchase they later regretted after following an AI suggestion. Roughly half of those regrettable purchases were electronics; apparel and footwear together accounted for about 45%.

Consumers also worry about what feeds those recommendations. Fifty-three percent of respondents suspect companies intentionally release false or misleading information to influence AI outputs. An overwhelming majority — 98% — said AI vendors should bear some responsibility for validating the sources behind product recommendations and for confirming that recommended companies are legitimate.

The data show demographic differences in adoption and resistance. Men report using AI for shopping more often than women (63% vs. 53%). Generationally, about 21% of Gen Z and 19% of Gen X respondents said they will never use AI to shop, compared with 10% of millennials.

The SmartCustomer findings line up with other recent surveys that highlight trust as a barrier. YouGov found last year that 41% of respondents didn’t trust AI shopping assistants at all. An RTB House study reported that 35% of people would want a human to review transactions before an AI agent completes them (44% among baby boomers). A Gartner survey of more than 1,500 U.S. consumers found half prefer shopping with brands that don’t use AI in consumer-facing content, and more than two-thirds question whether the content they view is authentic.

Why this matters to marketers and retailers

For commercial teams, the takeaway is straightforward: AI-driven recommendations can drive acquisition, but the trust deficit creates operational and reputational exposure. When shoppers blame AI for poor purchases they can escalate friction to retailers — returns, refund requests and customer-service interactions — and to platforms, accusing them of manipulation or misinformation.

Operational priorities should shift accordingly. Product and personalization teams should treat provenance, verifiability and transparency as core features, not optional add-ons. That means three practical areas of immediate focus:

  • Data and source verification: Ensure the training data and signal sources that feed recommendation models are traceable and come from reputable sellers or verified feeds.
  • Recommendation transparency: Surface concise explanations for why an item was recommended (for example, “based on past purchase X” or “popular among shoppers who bought Y”) and show seller verification status where relevant.
  • Customer protections: Make return policies clear, provide fast human escalation paths, and monitor return rates specifically for AI-recommended items.

Marketplaces and aggregators that surface third-party or web-derived recommendations face particular scrutiny. SmartCustomer respondents expect AI vendors to validate sources, which raises reputational risk for services that do not clearly distinguish between verified and unverified signals.

Segmentation matters too. The survey shows attitudes vary by gender and generation, so messaging that explains how recommendations are generated and how mistakes are handled can reduce friction for skeptical cohorts.

Michael Lai, SmartCustomer’s CEO, summed up the tension plainly: AI use is becoming standard in shopping, but trust isn’t keeping pace — a gap that, according to the survey, has already led a third of shoppers to regret AI-driven purchases.

What to watch next: marketing and product teams should track emerging regulatory guidance and platform policies on AI transparency, monitor return rates tied to AI-recommended items, and A/B test simple provenance cues (for example, “recommended because you purchased X” or “verified seller”) to measure whether clearer sourcing moves both trust and conversion. Given current consumer concern, proving the quality and origin of recommendation inputs will be as important as improving model accuracy.