AI's 'Made in USA' Deception
Columbia Study: AI Bots Detect 'Made in USA' Contradictions, But Retailers Ignore Them
Anonymous
Key Points
1
Alexa for Shopping told researchers it didn't have access to 'Made in USA' information, but by rephrasing the prompt to 'in the USA' it produced a list of products labeled as Made in the United States.
2
Walmart's Sparky showed inconsistent behavior: it could rank items in a cart from 'high confidence' to 'low confidence/likely not U.S. made' after prompts, indicating variability in origin data disclosure.
3
One chatbot explicitly framed enforcement as a policy issue, with Sparky saying 'That's a business calculation, not a legal justification' when asked why it doesn't flag questionable labels.
4
Amazon says country-of-origin information, when available, is displayed on product detail pages and that it is continually working to improve Alexa for Shopping to surface more origin data.
5
The report notes widespread misleading 'Made in USA' claims, including products described as both domestic and imported, illustrating the scope of the issue across listings.
Amazon and Walmart's AI shopping tools can spot fake "Made in USA" labels — but they almost never tell shoppers about it. A new study from Columbia Law School's Center for Law and the Economy found that both companies' chatbots detected contradictions in product origin claims yet stayed silent, leaving buyers in the dark NewsTribune.
The center was co-founded by Lina Khan, the former chair of the Federal Trade Commission. Researchers say the findings reveal a troubling gap between what AI can do and what it actually does for consumers Retail Gazette.
Chatbots Spot the Problem but Say Nothing
Amazon's Alexa for Shopping first told researchers it had no access to "Made in USA" information. But when researchers rephrased the question to ask about products made "in the USA," the bot produced a full list of items labeled as American-made PPC Land. That inconsistency suggests the tool has the data — it just isn't sharing it upfront.
Walmart's Sparky showed similar behavior. After specific prompts, it ranked items in a shopper's cart from "high confidence" to "low confidence/likely not U.S. made" Storyboard18. That kind of ranking requires real knowledge about product origins. Yet Sparky doesn't offer it to shoppers by default.
'A Business Calculation, Not a Legal Justification'
When researchers pushed Sparky on why it doesn't flag questionable labels, the bot gave a telling answer. It said enforcement mostly targets manufacturers, not platforms. Researchers shot back that this was "a business calculation, not a legal justification" BigGo Finance.
Amazon's chatbot was even more candid. It acknowledged that false "Made in USA" claims harm U.S. brands. But it suggested the company had little reason to act unless the problem created direct costs for Amazon itself NewsTribune. Critics say that logic puts company profits ahead of consumer protection.
Misleading Labels Widespread Across Listings
The report found the problem is not limited to a handful of products. Researchers found listings that described items as both domestic and imported at the same time Retail Gazette. That kind of direct contradiction should be easy for AI to catch — and, as the study shows, it often is.
Amazon responded by saying country-of-origin details are displayed on product pages when available. The company added it is "continually working" to improve Alexa for Shopping to surface more origin data PPC Land. That falls short of what the Columbia researchers are calling for.
Columbia Center Calls for Stronger Policy Enforcement
The study's authors argue that AI capability alone is not enough. Having the tools to detect fraud means nothing if platforms choose not to use them Storyboard18. The Columbia center says stronger rules and real enforcement may be the only way to change that calculus.
The findings also raise a fairness question. Small businesses face strict rules on country-of-origin claims. Critics say major platforms should be held to the same standard BigGo Finance. Right now, the AI knows the truth — it just isn't telling you.
Sentiment
Mostly NegativeNEGATIVENEUTRALPOSITIVE
Publishers
14
Articles
42
Reach
56
A
AnonymousAuthor