If you work at any large company, there's most likely an AI assistant on your website right now answering customer questions on your behalf. It's been tested for accuracy around things like the return policy, store hours, shipping fees. What nobody has likely tested is what it says when a customer asks why. Why is that the policy? Why don't you offer that? Nobody wrote those answers, because nobody thought of the assistant as a spokesperson. But every time it explains a decision, a policy or a piece of strategy, it's speaking on the company's behalf, without a brief, without a review, and without anyone in comms knowing what it said.

Amazon and Walmart found out this week what that answer sounds like when a senator plays that messaging tape back.

On Wednesday, Democrat Tammy Baldwin and Republican Rick Scott wrote to the Federal Trade Commission (FTC) asking it to investigate whether the two retailers' AI shopping assistants, Amazon's Alexa for Shopping and Walmart's Sparky, hide information about American-made products and fail to flag goods falsely labelled "Made in USA." The letter leans on a July study from a Columbia Law School center co-founded by former FTC chair Lina Khan. The letter quotes the chatbots: "When asked why there is no Made in USA filter, Alexa replied that it would 'redirect significant sales away from their largest seller base'," the senators wrote, referring to overseas manufacturers. The study also reports that when researchers asked Sparky why Walmart didn't flag suspicious origin claims, the bot pointed to limited FTC enforcement and called the company's inaction "a business calculation, not a legal justification."

An Amazon spokesperson told Reuters by email that "the suggestion that we intentionally withhold country-of-origin information from customers is fundamentally wrong." Walmart hadn't responded to Reuters by the time of publication. The FTC confirmed to The Hill that it received the letter and declined to comment. So, this is still a request for an investigation, not an investigation, and the FTC had already written to both companies last year about third-party sellers' origin claims, so the letter is an escalation but not a new fire entirely.

A word of caution here because Alexa didn't actually really reveal anything here. Ask an AI assistant why a company does something and it will give you a reason, whether or not that reason is legit. It's built to answer; it isn't built to know. The Columbia researchers even admit this in their own methodology, noting that communications teams would object to treating a chatbot's answer as an official statement. They're right, and I think we’d all very much agree.

But it also doesn’t really matter. It was said in Amazon's name, to a customer, in an answer to a question about Amazon's intentions, and now it sits in a letter to a federal regulator as evidence of those intentions. If a junior media relations staffer told a reporter "we don't offer that filter because it would cost our biggest sellers," you'd have immediately corrected the record and worked to figure out who authorized that statement and why. Nobody asked who authorized Alexa, because nobody thinks of the AI assistant as someone who speaks for the company.

But it does.

It's the only spokesperson you have that's on the record with millions of customers a day and has never once been briefed by the comms team.

There are two things in play here. 1) the senators' claim that Amazon hides American-made products and 2) the sentence Alexa said. Amazon's full official statement to Reuters was that "the suggestion that we intentionally withhold country-of-origin information from customers is fundamentally wrong." That denies the claim but it doesn’t say anything about why Alexa said what it said, whether Alexa still says it, or who's in charge of what Alexa says. So for as long as this story runs, every article will also include Alexa's words next to Amazon's official denial.

So here's what I'd do.

Pick the five policies you'd least like explained to a customer by a chatbot: the ones that show up all the time in customer complaints, the ones with real business sensitivity behind them. Then ask someone on the product or IT team to sit with you and put "why does XXYZ company do this?" to the assistant for each one. Read the answers as if a reporter had them, because one might.

What you'll get back will be a mix. Some answers will presumably be fine, some will be wrong in a harmless way, and most likely one or two will be materially problematic: a confident, fabricated account of your company's motives, in your company's voice.

Those are the ones you need to take to the next meeting with product and legal, and the meeting needs to settle three things.

First, which of these questions the AI assistant even be attempting to answer at all; "I can't speak to why, but here's the policy and here's who can" is a perfectly fine answer and probably most often the right one.

Second, who owns the wording when it does answer, because right now the honest answer is nobody.

Third, what happens when a bad answer surfaces — who can change it, how fast, and who gets told. Put a clock on that last one because a correction that takes three weeks to route through product isn't really a correction.

Lastly, write the holding statement now before you need it and make sure it does what Amazon's didn't. It has to answer the accusation and the sentence: "The assistant's explanation was not a company statement. Here is our position, and here is what we've changed." Without the second half, you're just denying the claim while the quote sits there.

Product will tell you the assistant is tested and it is probably but for accuracy. Nobody's testing motive, and until someone does, the bot is improvising it.