Senators press Amazon and Walmart on AI made-in-USA label errors

An AI shopping assistant will happily answer a question your listing never answered. US senators have now asked Amazon and Walmart about assistants attaching "made in USA" claims to products where the origin does not support it, and the liability for an origin claim sits with whoever owns the listing.

What happened

Senators wrote to Amazon and Walmart over their shopping chatbots producing country-of-origin answers that do not hold up, including "made in USA" labelling snags (Reuters, September 17).

The underlying problem is structural. Assistants summarise whatever is in the listing: title, bullets, A+ content, attributes, reviews and seller answers. When those sources disagree, the summary picks one. "Designed in California" in a bullet and a China country-of-origin attribute are not a contradiction to a human reader. To a summariser they are two candidate answers.

Why it matters

Origin claims are regulated. In the United States an unqualified "made in USA" claim requires that all or virtually all of the product be made domestically, a standard the Federal Trade Commission enforces (FTC, Complying with the Made in USA Standard).

If an assistant repeats a claim you cannot defend, the first place anyone looks is your listing. The practical exposure for a seller is not the chatbot. It is the ambiguity in your own catalogue that the chatbot made visible.

Where the ambiguity usually lives

• A country-of-origin attribute left blank or inherited from a template.
• Patriotic phrasing in bullets or A+ content that reads as an origin claim: "American brand", "built in the USA", flag imagery next to a product shot.
• A variation family where one child is assembled domestically and the parent copy speaks for all of them.
• Reviews and customer answers asserting an origin that no attribute supports.
• Copy written before a supplier moved, still live two sourcing changes later.

What to do this week

1. Export your catalogue and sort by country of origin. Every blank is a gap an assistant will fill for you.
2. Search your own copy for origin words. USA, American, domestic, imported, and the flag emoji, across titles, bullets, descriptions and A+ modules.
3. Qualify anything that is partly true. "Assembled in the USA with imported components" is defensible. An unqualified claim usually is not.
4. Ask the assistants what they say about your top ASINs. Rufus, a general chatbot, and the Walmart assistant. Record the answers with the date.
5. Fix the attribute before the copy. Structured data is what machine readers trust first.

The Nova angle

Listing problems are cheapest to fix while they are still content problems. Nova's listing health view flags weak and missing listing content across a catalogue so gaps surface before a customer, an assistant or a regulator finds them. See Nova listing health.