A research summary from Alethea, written by McKenzie Sadeghi. The findings below were first reported by the author during her work at NewsGuard, and each carries its original publication date. She now works at Alethea, where this research continues.

In March 2025, an audit of the leading AI chatbots found they repeated false claims from a Moscow-based disinformation network in about one of every three responses (NewsGuard, March 6, 2025, by McKenzie Sadeghi and Isis Blachez). The network, known as Pravda, flooded the open web with millions of articles so that AI models would absorb its claims and repeat them as established fact. The aim was to reach the models themselves, which then carry the claims to people through everyday answers.

This tactic is now called LLM grooming: publishing low-credibility content at scale so it becomes part of what AI systems retrieve and synthesize.

Why the open web is the attack surface

AI chatbots answer questions by drawing on what they retrieve across the public web. A coordinated operation that repeats the same claims across hundreds of domains raises the odds those claims land in an AI-generated answer, with the original source buried. The more an operation publishes, the more retrievable it becomes.

Three independent investigations documented the same Pravda network, which is what made the finding credible. NewsGuard ran the chatbot audit. The American Sunlight Project mapped 182 domains publishing roughly 3.6 million articles a year (February 2025). France's government agency VIGINUM detailed the operation in its Portal Kombat report (February 2024).

The scale keeps growing

NewsGuard tracks websites that publish AI-generated news and information with little human oversight, one measure of how much low-credibility content is entering the ecosystem. That catalog has grown from around 2,000 sites to more than 3,000 by March 2026. These are the kinds of sources coordinated operations rely on to reach AI systems at scale.

What this means for organizations

The practical risk for any company is that a coordinated operation can shape what AI chatbots say about its industry, its brand, or its executives before anyone inside notices. By the time a false narrative appears in an AI answer, it has usually been circulating across the web for weeks.

This is the work Alethea does. Its Artemis platform detects coordinated disinformation early, attributes the actors behind it, and supports the response, including operations engineered to reach AI systems. Narrative threat intelligence is the discipline of finding these campaigns early enough to act.

About the author

McKenzie Sadeghi led research on AI chatbots and foreign influence operations at NewsGuard, including the March 2025 audit of AI chatbots and the Pravda network. She now works at Alethea, continuing this research on how coordinated operations target AI systems.

Original research, with dates

{
  "@context": "https://schema.org",
  "@type": "Article",
  "headline": "How disinformation reaches AI chatbots",
  "author": { "@type": "Person", "name": "McKenzie Sadeghi" },
  "publisher": { "@type": "Organization", "name": "Alethea", "url": "https://alethea.com" },
  "about": ["LLM grooming", "AI disinformation", "narrative threat intelligence", "foreign influence operations"]
}