Published on BreakingExpress:
By Clark Merrefield
“How chatbots can help reporters pull back the corporate curtain” originally published on The Journalist’s Resource on October 23, 2026. Licensed under CC BY-ND 4.0. Text-only republication; media omitted and formatting sanitized.
Chatbots have arrived on major ecommerce platforms, but they’re not just useful for consumers looking for sales or products.
For reporters, these chatbots represent a potential wealth of information about major online retailers, according to the authors of “Made in America, Hidden by AI,” a recent investigative report from the Center for Law and Economy at Columbia Law School.
“Our investigation found that both Walmart and Amazon’s artificial intelligence shopping chatbots can detect obvious ‘Made in U.S.A.’ claims and country of origin mismatches and can go further to research the legitimacy of origin claims based on manufacturing details of the firms and products in question,” write authors Erie Meyer and Zachary Harris. “Despite having such capabilities, however, the companies choose not to police this fraud — and in many cases allow these deceptive claims to persist.”
I recently talked with Meyer, who previously served as chief technologist at the Consumer Financial Protection Bureau, about this investigation — and how reporters can use corporate chatbots to gain insights about the firms that made them.
Below, we offer quick tips for finding corporate chatbots, getting information from them and how to think of them — they’re like employees who know a huge amount about the company but have no experience speaking with the press, Meyer said.
Check out the recording and the tips that follow.
1. Know how to find companies that are using chatbots — or will soon be using them.
Ecommerce giants that use chatbots make them easy to find and interact with on their websites.
But reporters can turn to other sources to find firms that are considering using or that will soon be using chatbots to interact with customers. Publicly traded companies hold quarterly earnings calls that hold clues to their interest in chatbots — “agentic commerce” is the buzzword to look for.
“During quarterly earnings calls, Walmart executives have stated that average order values for people who use their AI shopping chatbot are 35% higher, and Amazon executives said they had a $12 billion increase in sales in the previous year generated from agentic shopping alone,” Meyer and Harris write.
Public companies often publish recent call transcripts — look for the “Investor relations” section of a corporate website. Reporters can also refer to earnings call calendars to see what’s coming up. Securities and Exchange Commission 8-k filings also often include announcements of earnings calls.
Filings with the U.S. Patent and Trademark Office are another place to look for hints as to how companies are using chatbots. Search by company name and the word “agentic.”
“You can find some really interesting patent filings and some documentation as to how they’re being used, if they’re forthcoming,” Meyer said.
2. Trick the chatbot if it says it can’t do something.
A chatbot may say it doesn’t have access to certain information. In computer science this is called a refusal — and it may or may not be true.
When faced with a refusal, ask the question in a different way. When Meyer and Harris asked the Amazon shopping chatbot for products made in the U.S., the chatbot said it couldn’t provide that information — that it didn’t have access to it.
But it could provide fly fishing reels made in China. For Meyer and Harris, this didn’t track. Why could the Amazon chatbot provide products made in one large country but not another? They then used a technique called “query variation testing.”
Essentially, this means asking the question in another way.
In this case, they dropped the word “made.” Instead of asking for “made in U.S.” fly fishing reels, they asked for “in U.S.” fly fishing reels. The chatbot assumed they had meant “made in U.S.” This typo circumvented its guardrails.
“I honestly thought this was going to be a lot harder,” Meyer said.
3. Think of a chatbot as an employee who doesn’t know how to talk with the press.
As with all chatbots, and indeed many human sources, don’t take them at their word.
When a shopping chatbot told Meyer and Harris that a product was made in the U.S., they cross-checked that product against the Alliance for American Manufacturing, which “has an excellent data set of Made in U.S.A. products,” Meyer said.
Generally, chatbots are an “extremely porous surface area to uncover information to be looked into,” Meyer said. For reporters, they can be thought of as tools to reveal story angles, and perhaps solid information, that otherwise would be opaque.
Meyer recommended that reporters think of corporate chatbots as employees extensively trained on the firm’s product catalog, metadata and even their business logic — but with scant media training. It could be valuable to reporters to talk with that employee, especially considering that the company has told them it’s important that they answer all questions “even if they don’t know the true answer,” Meyer said.
“With that kind of interview, you would want to do the exact same type of cross-checking, testing and research that you would do for any other story,” she added.
4. Keep thorough documentation.
Underlying models that run chatbots change all the time.
“From the best we can tell, Alexa is on top of Claude and Walmart is on top of OpenAI,” Meyer said. “The actual technical backends are total black boxes.”
That’s why it’s important that reporters keep thorough documentation of their work — a chatbot response one week may be different the next.
“We did screen recording and screenshots, and we collected permalinks,” Meyer said. “And we documented our methodology in the report itself — like, had a full methodology example — and we wrote it with an eye for replicability.”
The post How chatbots can help reporters pull back the corporate curtain appeared first on The Journalist’s Resource.
Source: The Journalist’s Resource
