About NewsGuard AI

NewsGuard AI is the only chatbot specializing in providing reliable responses to prompts about topics in the news.

The large all-purpose AI models are infected by the hoaxes and propaganda spread by the thousands of unreliable sources on the internet, which helps explain why the top chatbots provide false information on controversial news topics one third of the time. These models have also lost access to top quality news from many of the best news publishers by refusing to compensate them for their journalism.

NewsGuard AI re-imagines the chatbot experience by giving readers responses only from trustworthy news sources—and by sharing the revenues with publishers, giving NewsGuard AI users access to the best journalism and providing the publishers with much-needed funds to maintain their news operations.

Curated by Professional Journalists

While NewsGuard AI is AI-powered, its content is always drawn from humanly curated sources. That’s because NewsGuard AI is a product of NewsGuard, a company whose journalists have rated online news providers since 2018 based on nine transparent, apolitical basic journalistic criteria.

Only those online news sources that are rated as generally reliable are used by NewsGuard AI to produce responses to prompts. For more information about NewsGuard’s leadership and team, click here.

NewsGuard AI’s ability to use the proprietary NewsGuard ratings data means that no hoaxes being peddled by Russian or Chinese disinformation networks will infect its answers. Nor will sites posing as local news that are actually being secretly funded by political activists on one side or the other. Nor will those pushing inaccurate healthcare information, phony claims about consumer products, or repeatedly pushing provably false claims to promote one side or the other in political debates from Boston to Beijing.

A Big Tent

NewsGuard’s inventory of more than 12,000 news and information sites is a big tent—excluding only sources that have been identified by NewsGuard’s analysts as meeting one or more of these three exclusion criteria:

  1. They promote provably false claims.
  2. They are operated by hostile foreign adversaries whose propaganda is spread on the internet and by the existing AI models.
  3. They fail to achieve an overall score of at least 60 out of 100 according to NewsGuard’s nine journalistic criteria related to credibility and transparency.

Venerable newspapers, magazines and broadcasters are included, along with ambitious local news startups, single-author news, commentary websites, Substacks, the informational websites of state and local governments, think tanks, hospital systems, and research universities that have been assessed as credible. Conservative publishers and liberal publishers with strong convictions are all included. But not if those convictions spill over into repeatedly promoting provably false factual claims to support those views.

Put simply, NewsGuard AI is for people who want real news from real journalists.

NewsGuard AI’s Guardrails

NewsGuard has another set of data that further safeguards the information NewsGuard AI provides to its users: In order for the NewsGuard ratings to be appropriately transparent, NewsGuard’s journalist-analysts write a Nutrition Label explaining in detail how they judged whether the website adheres to NewsGuard’s nine criteria. The most important of these criteria is whether a site “repeatedly publishes false or egregiously misleading news.” In order to substantiate that, the Nutrition Label includes descriptions of these false claims accompanied by a detailed, well-sourced debunking of the claim.

This enables NewsGuard to have the world’s only humanly curated, continuously updated catalog of provably false claims related to news topics circulating online.

This catalog, which includes 64,000 examples of false claims in the news spreading online in every language and geography, feeds into the NewsGuard AI answer process. This means that as NewsGuard AI will cite NewsGuard’s debunking of the claim instead of spreading it.

Solving the AI Misinformation Problem

NewsGuard is launching NewsGuard AI after more than a year of monthly reviews of how the existing top 11 AI models respond to prompts on news topics. They are highly unreliable, spreading false claims on controversial topics in the news we have debunked on average 35% of the time. (You can read our analysis here).

This is because the leading AI models rely on claims on the internet regardless of the reliability of the source. For example, one Russian influence operation infected the AI models by publishing 6.3 million articles in 2025 spreading the Kremlin’s favorite 207 false claims. By overwhelming the internet this way with false claims, the existing chatbots repeat them as accurate, from authoritative sources. To read more about our reporting, click here.

Readers Benefit from Access to News from Top Publishers

The AI models have refused to compensate news publishers for their journalism, with a few exceptions where AI models paid a handful of publishers for access to their content. This means the AI models do not have the right to provide this quality journalism in their responses to prompts. Instead, foreign propaganda, healthcare hoax and conspiracy sources are happy to make their false claims freely available for the AI models to spread.

In contrast, NewsGuard AI will split revenues with all of the news publishers whose work it cites, aligning the interests of readers and publishers. After a period of free access, we will ask the most frequent users of NewsGuard AI to subscribe to get full access, which will provide support for quality publishers.