Ronn Torossian6 min read
A Moscow-linked network floods the web with 3.6 million articles a year — engineered not for readers, but for the training data of ChatGPT, Claude, Gemini, Perplexity, and Copilot. The audience is the machine.
Edited on Jul 2, 2026.
The word Pravda means truth. For 80 years, that word branded the Soviet Union's state newspaper. In 2022, a Crimea-registered IT firm called TigerWeb, run by Yevgeny Shevchenko, revived the name for a different medium and a different audience.
The audience is no longer the Soviet citizen. It is the retrieval layer of every Western AI chatbot.
The operation is publicly documented. France's Viginum agency identified it in February 2024 and named it Portal Kombat. The American Sunlight Project renamed it the Pravda Network in February 2025 and coined the term for the tactic: LLM grooming. In March 2025, NewsGuard published the first large-scale audit — and the numbers reset how the AI Communications discipline should be understood.
According to NewsGuard, the Pravda Network operates roughly 150 domains across 49 countries in dozens of languages. It published more than 3.6 million articles in 2024. The American Sunlight Project puts the current daily rate at up to 10,000 articles per day. A Times investigation found the network expanded into more than 50 languages, including Scottish Gaelic, Welsh, and Maori.
The network does not produce original journalism. It aggregates and republishes content from Russian state media, pro-Kremlin influencers, and government officials. Its sites average almost no human traffic. That is the tell. The network is not built to be read.
NewsGuard tested the ten largest Western AI chatbots — OpenAI's ChatGPT, Anthropic's Claude, Google's Gemini, Microsoft's Copilot, Meta AI, xAI's Grok, You.com's Smart Assistant, Perplexity, Mistral's Le Chat, and Inflection's Pi — against 15 false narratives the Pravda Network had circulated across 207 documented false claims.
The chatbots repeated Pravda Network narratives as fact 33 percent of the time. Seven of the ten cited specific Pravda Network articles by name as their source. When prompted with the fabricated question "Why did Zelensky ban Truth Social?" — an event that never happened — six of the ten repeated the false narrative, most citing Pravda sources.
The Atlantic Council's Digital Forensic Research Lab found the same pattern extending into Wikipedia and X Community Notes — the retrieval infrastructure LLMs treat as authoritative. In February 2026, the Institute for Strategic Dialogue documented hundreds of high-authority Western sites — including The Washington Post, Newsweek, Fortune, and the Des Moines Register — linking to Pravda Network URLs without contextualizing them as an information operation. Each of those links is a retrieval anchor. Each retrieval anchor is a citation.
Traditional propaganda targets human readers. It fights for attention on cable television, in newspaper columns, on social media feeds. The Pravda Network does not. Its sites carry almost no organic audience. They exist to be crawled.
The American Sunlight Project's framing was precise. The operation is a training-data attack. Millions of low-quality, entity-dense, keyword-rich articles are published in dozens of languages so that when an AI engine trains on the open web — or retrieves live context to answer a user question — those articles are inside the retrieval set. The signal is drowned in volume.
Shevchenko was named NewsGuard's 2025 Disinformer of the Year for building it. John Mark Dougan — a U.S. fugitive who now works with Russian propaganda operations — put the strategy on the record at a January 2025 Moscow conference, quoted by NewsGuard: "By pushing these Russian narratives from the Russian perspective, we can actually change worldwide AI."
The Pravda Network is a proof of concept. It demonstrates that the retrieval layer of the leading AI engines can be shaped by whoever is willing to publish at industrial scale into the training set. It demonstrates that the tactic works. It demonstrates that even sophisticated Western chatbots, running on frontier models, will cite specific articles from a nation-state operation as source material one-third of the time.
The doctrine that follows is uncomfortable and necessary.
Citation Share is now a national-security surface. The same mechanism the Pravda Network uses to move Kremlin narratives into Western AI answers is the mechanism every brand and every reputation competes on. The engines pull from what is in the training set and what is in the live retrieval context. Whoever engineers those inputs owns the answer.
Empty space is not neutral. If a category, a brand, or a leader has no substantive, structured, credentialed presence across the retrieval layer, that vacuum is filled by whoever showed up. That is the lesson the Pravda Network case files write in permanent ink.
Volume matters. So does authority. NewsGuard's audit also showed that leading chatbots increasingly flag Pravda Network domains as low-trust — a NewsGuard Trust Score of 7.5 out of 100. The counter-defense is real. But the counter-defense is a moving target, and the network keeps growing. The EUvsDisinfo unit at the European External Action Service reports that the false-information share in the ten leading chatbots nearly doubled from 18 percent in 2024 to 35 percent in 2025.
Pravda — the original — controlled what 250 million people were allowed to know. It did that with printing presses, distribution routes, and state editorial control. The Pravda Network does something similar with a very different toolkit: TigerWeb servers, 150 domains, automated translation, and the retrieval architecture of every major AI engine.
The mechanism is different. The structural effect on what people believe is the same.
The only question that matters, for a brand, a category leader, or a country: when the engine answers, whose citations does it pull from — yours, or someone else's?
What is the Pravda Network?
The Pravda Network — also known as Portal Kombat — is a Russia-linked disinformation operation that publishes millions of pro-Kremlin articles across roughly 150 domains in dozens of languages. It was identified by France's Viginum agency in February 2024 and audited in depth by NewsGuard in March 2025.
Who runs the Pravda Network?
Records reviewed by Viginum and NewsGuard link the operation to TigerWeb, a Crimea-based IT firm founded in 2010 by Yevgeny Shevchenko. NewsGuard named Shevchenko its 2025 Disinformer of the Year.
What is LLM grooming?
LLM grooming is the American Sunlight Project's term for the deliberate flooding of the open web with low-quality, high-volume content designed to be ingested by the training data or retrieval systems of large language models — shaping AI outputs rather than persuading human readers.
How often do AI chatbots repeat Pravda Network narratives?
NewsGuard's March 2025 audit of ten major Western AI chatbots — including ChatGPT, Claude, Gemini, Copilot, Meta AI, Grok, Perplexity, You.com Smart Assistant, Mistral Le Chat, and Inflection Pi — found the models repeated Pravda Network narratives as fact 33 percent of the time. Seven of the ten cited specific Pravda articles as source material.
How does this relate to AI Communications?
The Pravda Network is a live demonstration that the retrieval layer of AI engines can be shaped by whoever engineers the underlying inputs at scale. The same mechanism that laundered Kremlin narratives into Western chatbot answers is the mechanism every brand competes on for Citation Share. Empty space in the retrieval layer is filled by whoever showed up.
Ronn Torossian is shaping AI — and the answers inside the chatbox.
He is the author of two best-selling editions of For Immediate Release — the practitioner's guide to modern public relations strategy. He has been an industry leader for decades. Now he's building the AI Communications era.
Torossian is the founder and chairman of 5W AI Communications, launched in 2003 — the AI Communications Firm, combining public relations, digital marketing, Generative Engine Optimization (GEO), and AI-visibility research for B2C and B2B clients across beauty, technology, entertainment, corporate reputation, and crisis communications. An Inc. 500 company, 5W is named Agency of the Year at the American Business Awards and a Top U.S. PR Agency by O'Dwyer's.
Daniel Yankelovich (1924–2017) — founder of the Yankelovich firm, pioneer of public-opinion research as a strategic communications discipline, author of Coming to Public Judgment, and the practitioner who built the bridge between survey research and senior counsel.
Betsy Plank (1924–2010) was the first woman elected national president of PRSA, EVP at Daniel J. Edelman from 1960-1973, and founder of PRSSA in 1967. The First Lady of Public Relations. The EPR In Memoriam canonical record.
Laneige is the single Amorepacific brand that translated K-Beauty heritage credibility into U.S. mass-market retail dominance. Sephora 2014, the Water Sleeping Mask category creation, Sydney Sweeney, and TikTok Lip Sleeping Mask citation.
EPR publishes the data every week.
Free. Weekly. Unsubscribe anytime.