A story labeled as AI-assisted but filled with AI-generated experts sparked backlash in Brazil, raising questions about the role of human expertise in journalism and the ethical limits of the technology.
What should have been an ordinary lifestyle story about diets quickly turned into a major scandal in Brazil over the use of artificial intelligence in journalism.
On June 26, both the portal Terra and the newspaper Estado de Minas published an article produced by the third-party content creator Giro10 about a trend known as "fibermaxxing," which involves gradually increasing the amount of fiber in one's diet.
The problem? The story, attributed to a writer named Jonas Moura and labeled as having been produced with the assistance of AI, was filled with quotes from what Giro10 described as "simulated interviews" with made-up experts, including "imaginary clinical nutritionist Dr. Rafael Martins," "fictional gastroenterologist Dr. Helena Costa" and "fictional public health researcher Dr. Mariana Lopes."
The backlash was quick, and the article was taken offline.
As the use of AI has become more common in newsrooms, one of the first recommendations from journalism experts has been that news organizations should be transparent about how the technology is used in reporting and content production.
But Giro10's fibermaxxing story underscores new ethical questions — and suggests transparency alone may not be enough. While the company disclosed that AI had been used to produce the article, the case raised broader concerns about the role of human expertise in journalism and where to draw the line when using AI to gather information.
Journalism ethics professor at Universidade Federal de Santa Catarina (UFSC) Rogério Christofoletti said reporters have always relied on statements, testimonies, opinions, accounts, and comments from human sources. He asked: Why should we turn to AI when we still have human experts?
“Journalism is a human product, made by humans, with humans, and for humans”, he told LJR. “Although many people are fascinated by advances in AI, its error rate is still high, we do not know how it processes the data it has, nor what data it is trained on.”
The incident led the National Federation of Journalists (FENAJ for its initials in Portuguese) to call on journalists and news organizations to establish rules governing editorial responsibility, transparency and ethical limits for the use of AI.
"The absence of any form of regulation, and the simple acceptance of AI as a legitimate tool for journalism, directly conflicts with the profession's core principles, including the credibility of information, independent fact-checking and the social responsibility to provide useful, trustworthy and verifiable information," a statement from FENAJ said.
Shortly after the scandal, Giro10 editor Carlos Vieira issued a statement saying the company takes full responsibility for errors found in some of the content produced by its team and published by Terra.
"These errors occurred because of failures in our review and approval process for content created with the assistance of artificial intelligence," he said.
Terra also issued a public apology, saying it removed the article as soon as it became aware of the incident.
"Although the material was not produced by Terra's editorial staff, we recognize that it was published on our platform and does not meet the standards of quality and reliability expected of all content distributed through our platforms," Terra said in a statement.
The outlet added that the incident exposed shortcomings in its oversight and review processes for content produced by partner organizations, particularly those using AI tools to assist with content creation.
Alexandra Moraes, the ombudsman at the Brazilian newspaper Folha de S.Paulo, also weighed in on the case. In a recent column, Moraes interviewed Vieira, who said Giro10 employs five journalists and that the bylines that appear on its content belong to real reporters.
According to Vieira, the agency, founded in 2025, uses a suite of AI tools operating simultaneously to produce its content.
"Fact-checking is carried out by humans. But there was human error,” Vieira told Moraes. “We have produced a total of 6,600 pieces of content, and errors occurred in six of them."
João Pedro Malar Massa, a journalism and advertising professor at Pontifical Catholic University of Minas Gerais, told LJR that the industry has seen a growing number of cases involving AI-generated errors that got past human review, or instances in which news organizations used AI without properly disclosing it, misleading audiences.
"In this case, the use of AI was disclosed in the articles themselves, and the stories made clear that the 'sources' were fabricated through 'simulated interviews.' I believe this was a test to gauge how receptive audiences would be to information generated by AI, but presented as if it came from a person, even though that person does not exist. It was a form of humanization," he told LJR.
Massa said the case is a classic example of the concept of "good enough," a principle widely discussed in management research on the adoption of new technologies and workplace automation.
"Companies know that a human journalist tends to produce higher-quality work than AI. But if they conclude that AI can produce something that is 'good enough,' then replacement becomes possible. That logic does not prioritize quality, and it could ultimately become a self-inflicted wound for journalism, an industry already facing a growing crisis of public trust."
The debate comes as often deceptive content featuring AI-generated experts is becoming increasingly common on platforms like YouTube.
A recent report by BBC News Brasil uncovered a global industry of fake AI-generated doctors that exploits fear to go viral among older adults in Brazil.
The nonprofit CTRL+Z, which helps hold technology companies accountable for consumer harms, also identified a network of fake doctors on YouTube that has amassed more than 70 million views, including 10 million in the past month.
To Massa, relying on AI tools that were not created or trained to serve as sources of specialized and reliable information is extremely problematic.
“AI systems do not fact-check; they have no commitment to the truth. They are good tools for reworking existing content, but terrible as sources of information,” he told LJR.“Certain tasks should not be delegated to AI, and the search for specialized information is one of them. It causes more problems than it solves.”
