AI chatbots are quickly becoming a popular way for people to find news. However, a system is only as reliable as its source, and is prone to misinformation and manipulation.
Meanwhile, the rise of AI is threatening the business models of news publishers, many of whom argue that creators of AI models are exploiting their results without fair compensation.
NewsGuard, a startup that evaluates the reliability of news sources, says it’s looking for ways to solve multiple problems at once.
On Tuesday, the company will launch a new product, NewsGuard AI, which aggregates information by only accessing information from sources it evaluates as trustworthy. Chatbot responses include citations and prominent links to news outlets.
Perhaps most importantly for publishers, the product promises to share the wealth. NewsGuard AI claims to be “the only chatbot that rewards every publisher for their content” through a 50/50 revenue share model.
NewsGuard hopes to generate revenue from users who think the chatbot is worth paying $6 a month for access.
First access is free, and NewsGuard works with a variety of marketing partners, including magazines and independent bookstores, to encourage usage.
NewsGuard co-CEO Steven Brill pitches the product: “Think about how existing AI chatbots work. In every way, NewsGuard AI does the opposite.”
In a Zoom interview, Brill and the startup’s other co-CEO, Gordon Crovitz, said NewsGuard leverages both a database of vetted news sources and research into the shortcomings of products like ChatGPT.
NewsGuard published a flurry of studies showing that popular chatbots are prone to sharing false claims and are vulnerable to propaganda campaigns.
“The idea that our nation’s leading AI models are susceptible to foreign disinformation operations is sobering, and they are not taking steps to remove themselves,” Crovitz said. “And that was one of the reasons why we saw an open market for services that would assess the credibility of sources and, of course, eliminate all sources of propaganda and disinformation.”
NewsGuard AI has similarities to specialized AI tools that are gaining traction in industries such as healthcare and law. But it’s also being marketed to the general public, albeit to people who care about accuracy.
In a demonstration ahead of the product’s launch, NewsGuard Chief Operating Officer Matt Skibinski showed how chatbots debunk medical misinformation and direct users to primary sources of information.
NewsGuard’s website advertises the new product as “your own personal fact checker,” noting that it also leverages the startup’s catalog of debunking more than 64,000 false claims.
But in a world already full of chatbot-type interfaces, convincing users to give it a try may be difficult.
That’s where marketing partners come into play. They spread the word about NewsGuard AI and receive a portion of the subscriptions they help sell.
Several news publishers are participating in the launch, including The Atlantic, one of the most highly regarded magazines in the United States.
“In the near future, nothing will be more important than the human ability to discern what is true, what is false, and what is fabricated nonsense,” Atlantic CEO Nicholas Thompson said in a statement.
“This is especially true when it comes to news, which is why we love that NewsGuard is launching an AI news product built on verified sources, clear citations, and a publisher reward model. We are proud to partner on this important issue and approach,” said Thompson.
This compensation model is interesting because major media companies are taking a two-pronged approach to AI, in some cases filing copyright infringement lawsuits and in other cases entering into content licensing agreements with AI companies.
“The stakes here are really, really high,” New York Times CEO Meredith Kopit-Levian said Monday at an Axios event.
“These companies that manufacture LLM stole our work. They used it without our permission,” Levien said of her company’s lawsuits against OpenAI, Microsoft and Perplexity.
CNN also recently filed a lawsuit against Perplexity, saying it prefers “sensible licensing arrangements” to legal battles, but that there are “no free options” for the AI model operators it builds with the help of news providers.
Perplexity responded to the lawsuit by saying, “Facts cannot be copyrighted.”
“We’re trying to stand shoulder to shoulder with the publishers, not stand shoulder to shoulder with them in court,” Brill said in a Zoom interview.
He said NewsGuard notified major news organizations in advance of its approach, but only two sources said they wanted to opt out.
The remaining members will be paid “based on how often their journalism is cited,” Krovitz said.
Queries often display links to 6 news sources. NewsGuard AI’s database also includes thousands of other vetted resources, including government websites.
Brill said NewsGuard does not store or train publishers’ content or peer behind a paywall. “We simply use what they make available to the public.”
About 10% of consumers use AI chatbots for weekly news, up from 7% a year ago, according to a study released last week by the Reuters Institute for Journalism.
The researchers concluded that “AI is more than just a vehicle for headlines.” “Some people use it to access the latest news, but many use it to research, summarize, and evaluate information, pointing to a broader role that combines access and interpretation.”
