Chatbot news network

McKENZIE SADEGHI knows how to spot a fake. She’s sifted through countless articles, hunting for telltale signs the author wasn’t a flesh-and-blood writer but an artificial intelligence program.

Sadeghi works at NewsGuard, a watchdog organization keeping tabs on the state of online misinformation. Over the past year, she and her colleagues have observed a troubling trend: a steady rise in the number of news websites run with little to no human oversight.

When NewsGuard first reported the situation last spring, it identified 49 AI-generated news websites. Now, that number is 750 and counting.

The sites pose as legitimate news outlets, using innocuous names like Biz Breaking News and Daily Business Post and covering everything from celebrity news to politics to health. But they hide critical information about sources, oversight, and funding—and their growth increasingly leaves online readers vulnerable to inaccurate or manipulative content.

NewsGuard’s investigation found sites like harmonyhustle.com, which posts technology and self-help content under fake author profiles like “Alex” and “Tom.” (A recent headline: “Embracing Uncertainty: The Key to Personal Growth.”) Or countylocalnews.com, a rinky-dink site that accidentally posted a 2023 article headlined: “Death News: Sorry, I cannot fulfill this prompt as it goes against ethical and moral principles.”

Most of these websites are anonymously registered by people seeking “money and clicks,” Sadeghi said. They typically harvest authentic content from credible media organizations and use AI to rewrite it for publication on pages packed with advertisements in an effort to drive revenue.

In the internet age, with so many sources of online information, it already takes diligence for readers to discern fact from fiction. But Mike Webb of the nonprofit News Literacy Project said it’s become harder since the November 2022 rollout of ChatGPT, a powerful AI chatbot.

Clickbait isn’t anything new, but AI has made it possible to churn out huge amounts of it at record speeds, Webb noted. A major misinformation risk with AI-generated content is “hallucination,” when chatbots inadvertently make up things that aren’t true.

That has already led to at least one libel lawsuit against OpenAI. A Georgia resident sued ChatGPT’s creator after the chatbot allegedly falsely identified him as the subject of a fraud and embezzlement case.

Websites seemingly written by AI proliferate online and easily turn up in a Google search. WORLD independently identified one, The Washington Independent, and sent it to Sadeghi for review. The domain was not yet on NewsGuard’s list, but Sadeghi said it fit the criteria of an AI-curated website.

Until at least 2014, the URL washingtonindependent.com was home to a news site run by the former American Independent News Network, but archived webpages show that group’s site ultimately shut down. Today, a new Washington Independent welcomes visitors with a random array of content under bizarre headlines like “Celebrities With Underbites—Embracing Uniqueness” and “What Is Tax Avoidance—Navigating the Fine Line Between Legality and Ethics.” (“Compliance with current tax rules and regulations is essential for the legitimacy of tax evasion,” the article states in clunky prose.)

When I reached out to The Washington Independent’s listed email address to ask whether the site used AI technology, I received only a stock reply listing the company’s ad and promotional services.

Sadeghi pointed out some Washington Independent articles included chatbot error messages that suggest much of the site’s content is AI-generated. In this case, the AI slip-up “as of my knowledge cutoff in September 2021” appeared in at least three different articles.

When Sadeghi and her co-workers find a questionable site, they try to learn who’s behind it and ask for ­comment. But operators are typically elusive—they often provide invalid email addresses or don’t respond to inquiries.

A senior writer for Wired magazine, Kate Knibbs, did catch up with one. In a recent article, she interviewed Serbian “clickbait kingpin” Nebojša Vujinović Vujo, a former DJ who claims to run over 2,000 websites with the help of ChatGPT. Vujo told Knibbs he’s “just an ordinary guy” trying to make a living in a run-down economy.

While many site operators may simply be trying to turn a profit, Sadeghi said some have more nefarious purposes. NewsGuard has identified two “pro-Russian” sites in the past few months, she said. And on March 7, The New York Times reported on five Russian-backed sites circulating fake news stories.

Mike Webb said readers can avoid getting duped by checking to see if articles cite multiple sources and by cross-checking them against info from reputable news websites.

Meanwhile, the number of shady news sites on NewsGuard’s tracker continues to grow. Sadeghi said her group intends to “find these sites and provide readers with the proper, accurate context about them.”