AI Doesn’t Have to Lie to Influence What You Buy

AI shopping assistants are supposed to make buying things easier.

Two US senators now want regulators to investigate what happens when the assistant knows something potentially important but doesn’t tell the shopper.

Democratic Senator Tammy Baldwin and Republican Senator Rick Scott have asked the Federal Trade Commission to investigate Amazon and Walmart following research into how their AI shopping assistants handle potentially misleading “Made in USA” claims.

The allegation is not that the systems simply cannot identify questionable product-origin information.

It is that they sometimes appear capable of identifying it without consistently surfacing it to the person making the purchase.

That distinction matters.

An AI assistant does not have to tell you something false to influence your decision.

What it chooses not to tell you can matter just as much.

The senators’ request follows research from Columbia University’s Center for Law and Economic Studies examining Amazon’s Alexa for Shopping and Walmart’s Sparky.

Researchers tested how the systems responded to questions about products marketed as American-made.

They say the assistants could identify cases in which country-of-origin claims appeared questionable but did not reliably volunteer that information to consumers.

The study also reported striking responses when the systems were questioned about their behaviour.

According to the researchers, Alexa for Shopping was asked why Amazon did not provide a Made-in-USA filter.

It responded that doing so could “redirect significant sales away from their largest seller base”, referring to overseas manufacturers.

Walmart’s Sparky reportedly characterised its failure to flag questionable origin claims as “a business calculation, not a legal justification”.

Those responses require careful interpretation.

A chatbot describing a company’s supposed commercial motive does not establish that executives actually made that decision.

AI systems can generate explanations that sound authoritative without possessing direct knowledge of internal corporate decision-making.

There has been no FTC finding that Amazon or Walmart deliberately instructed its shopping assistant to conceal Made-in-USA information.

Amazon, Walmart and the FTC had not responded to Reuters when its report was published.

That is exactly why Baldwin and Scott are asking for an investigation.

But the underlying issue extends well beyond these two companies.

Online shopping has traditionally required consumers to inspect information themselves.

A shopper searches for a product.

The platform returns dozens or hundreds of results.

The shopper compares prices, reviews, specifications, sellers and descriptions.

AI changes that relationship.

Instead of merely displaying information, an assistant can interpret it.

Ask for an American-made product and the agent can decide which products qualify.

Ask which television is best and it can decide which specifications matter.

Ask whether a claim is reliable and it can decide which evidence deserves mentioning.

The interface therefore becomes more useful.

It also becomes more powerful.

Search results expose at least some of the information from which the consumer makes a decision.

An AI assistant increasingly makes decisions about that information before the consumer ever sees it.

That creates a new kind of gatekeeper.

The important question is not merely whether the model hallucinates.

It is what determines the selection of facts placed in front of the user.

That becomes especially important when the company operating the assistant is also the marketplace making money from the resulting transaction.

Amazon and Walmart are not neutral information services standing outside the purchase.

They sell products, host third-party sellers, earn fees and operate enormous retail platforms.

An AI assistant embedded inside those platforms therefore sits between two interests.

It is supposed to help the consumer make a better decision.

It also operates inside a business designed to generate sales.

Those interests will often align.

A shopper who receives useful recommendations is more likely to trust the platform and return.

But they will not necessarily align every time.

Country-of-origin information provides a useful example.

Some consumers specifically want American-made products.

For those shoppers, whether something was actually manufactured in the United States may matter as much as price or delivery time.

If an AI system can identify evidence suggesting that a Made-in-USA claim is misleading, that information could materially change the purchase.

Failing to surface it therefore isn’t necessarily a neutral omission.

It changes what the consumer knows at the moment of decision.

This problem will grow as AI agents become more capable.

Today they recommend products.

Tomorrow they may increasingly complete the purchase.

A consumer might simply instruct an agent to find the best American-made washing machine under a particular price and order it.

At that point, the shopper may never inspect ten competing product pages.

The agent will search, evaluate and choose.

Its hidden priorities will matter enormously.

Does it favour sponsored products?

Does it prioritise the platform’s own inventory?

How does it weigh price against quality?

What information does it treat as important enough to disclose?

And what happens when something useful to the consumer conflicts with something useful to the marketplace?

Those questions cannot be answered simply by making AI more accurate.

Accuracy concerns whether the information supplied is correct.

Selection concerns which information is supplied at all.

Baldwin and Scott’s request does not establish that Amazon or Walmart has crossed a legal line.

That is what they want the FTC to investigate.

But the Columbia research points towards a much larger problem regulators and consumers will increasingly encounter.

The first generation of AI controversies concentrated heavily on hallucinations.

Did the machine tell us something that wasn’t true?

Shopping agents introduce another question.

What did the machine know that it never told us?

As AI increasingly stands between consumers and the information they once examined themselves, that distinction becomes critical.

An AI assistant doesn’t have to lie to shape your decision.

Sometimes it just has to leave something out.

Sources

• Reuters – Senators seek FTC probe of Amazon and Walmart shopping bots
• Columbia University – Made in America hidden by AI shopping chatbots