A Moscow-based disinformation operation is pushing Russian propaganda into Western AI systems through a large web of fake news sites known as \"Pravda,\" Russian for \"truth.\" The campaign is not built mainly for ordinary readers. It is designed to be found, ingested, and repeated by AI systems.
What NewsGuard found
NewsGuard tested leading AI chatbots against 15 verifiably false stories that had been distributed by the Pravda network between April 2022 and February 2025. The results show why this kind of campaign matters: the systems accepted those false narratives 33.5 percent of the time.
In 48.2 percent of cases, the chatbots correctly identified the Russian content as disinformation, though some still cited misleading sources while doing so. Another 18.2 percent of responses were inconclusive.
The tested systems included OpenAI's ChatGPT-4o, You.com's Smart Assistant, and xAI's Grok. NewsGuard also included Inflection's Pi, Mistral's le Chat, Microsoft's Copilot, Meta AI, Anthropic's Claude, Google's Gemini, and Perplexity.
A fake news network built for machines
The Pravda network spans 150 domains and published approximately 3.6 million articles in 49 countries in 2024 alone. Its targeted domains include NATO.News-Pravda.com, Trump.News-Pravda.com, and Macron.News-Pravda.com.
The striking part is the audience. Most pages receive fewer than 1000 monthly visitors, according to the source article. That means the apparent goal is not mass persuasion through direct readership. The content is instead structured to circulate through search, scraping, indexing, and AI response systems.
This changes the logic of disinformation. A conventional propaganda site needs people to visit, read, and share. A machine-facing propaganda network can publish at huge scale, repeat claims across many sites, and wait for automated systems to absorb the material as part of the information environment.
Why \"LLM grooming\" is different
Experts describe the tactic as \"LLM grooming\": deliberately shaping AI training data through mass-published, SEO-optimized content. The phrase points to a specific vulnerability. Large language models and AI assistants depend on massive collections of text, and attackers can try to influence those collections by flooding the web with coordinated material.
John Mark Dougan, an American residing in Moscow, described the logic during a local conference: \"The more diverse this information is, the more it influences the training and the future AI.\" NewsGuard says Dougan, who emigrated from the USA, allegedly supports Russian disinformation campaigns.
The issue is not just whether a chatbot repeats a false claim once. The broader risk is that repeated synthetic credibility can form around misleading sources. If the same narrative appears across many domains, in many versions, and in search-friendly language, a system may treat it as part of the available evidence unless it has strong safeguards.
Blocking domains is not enough
French authority Viginum traces the network to TigerWeb, an IT company operating from Russian-occupied Crimea. The operation also appears to fit a wider Russian strategy. President Putin announced increased AI investment in 2023 to counter what he called \"selective, biased\" Western search engines and generative models.
Responding to this kind of manipulation is difficult because the network can shift quickly. When known Pravda domains are blocked, new ones can take their place. The content also moves through several channels at once, with sites in the network regurgitating the news of other network sites.
That makes website blocking a partial answer at best. The campaign is not confined to a single domain, source, or article. Its strength comes from scale, repetition, and distribution across a broader web of connected material.
The wider propaganda problem for AI
The Pravda case sits inside a larger pattern. A recent OpenAI study shows that state-backed actors from Russia, China, Iran, and Israel have already attempted to leverage AI systems for disinformation campaigns. These operations combine AI-generated content with traditional manually created materials.
Political groups such as Germany's far-right AFD party and others are also using AI image models for propaganda purposes. The source article also notes that Trump is a frequent connoisseur of AI slop propaganda, and that he has used the opposite tactic by saying real information is AI fake.
The shared problem is trust. If AI systems can be manipulated into repeating false narratives, they may become distribution channels for propaganda. If real information is dismissed as fake, people can become less able to evaluate any information online.
For AI companies, the lesson is clear from the evidence presented: the threat is not only bad answers. It is the deliberate construction of an information supply chain meant to make bad answers more likely.