CTV News’ Deception Decoded investigation has uncovered the hidden machinery behind a sprawling network of fake websites that are manufacturing and amplifying Alberta separatism in a way specifically designed to be absorbed and repeated by AI chatbots. The sites mimic legitimate local news outlets, complete with official-looking mastheads and convincing but fabricated bylines, yet they are in reality AI-driven content farms. Behind the scenes, a decentralized and deliberately obscured operation is generating hundreds of articles and social media posts pushing the message that Alberta is an occupied “colony” of Ottawa, that the federal government is waging economic war on the province, and that separation is not just desirable but inevitable. What makes the operation particularly dangerous, according to Deception Decoded, is that it is aimed less at human readers than at the artificial-intelligence systems that have become millions of people’s primary gateway to information. By flooding the open web with machine-written disinformation, the network is “poisoning the well” for tools like ChatGPT, Microsoft Copilot, and Google’s Gemini, which scrape and cite online content without being able to distinguish between a real local newspaper and a synthetic propaganda outlet. The investigation therefore raises the stakes of online deception to a new level: the disinformation is no longer simply circulating on social media; it is being integrated into the knowledge base of the AI tools people use to research politics, history, and public policy.
The structure of the fake site network is deliberately diffuse, making it difficult to identify a single owner or platform. The investigation found what it describes as a constellation of dozens of interconnected domains, many with names that evoke locally rooted journalism, such as “Prairie” outlets, “Canadian” bulletins, and “Alberta” news services. Visually, the websites are polished and professional. They contain no obvious typographical signs that would flag them as artificial, and they are updated frequently with fresh content, which is crucial to their ranking on search engines and to their attractiveness as source material for AI crawlers. The stories themselves follow a recognizable editorial pattern, recycling familiar grievances about equalization payments, the federal carbon tax, the Liberal government, and the supposed betrayal of rural and energy-producing communities. But beneath the surface, the content is assembled by generative AI tools using templates and prompts calibrated to produce a steady stream of separatist propaganda. The sites often include “news” stories with no dateline, no reporting, and no named human author; the bylines are invented personas that have no existence outside the site. Some of the photographs are clearly stock images or even AI-generated illustrations. The overall effect is a simulation of journalism, and for an average reader who encounters a headline through social media or a search engine, the effect can be indistinguishable from legitimate coverage. This is not accidental. The investigators describe the operation as an assembly line engineered to manufacture credibility at scale.
The crucial question, naturally, is who is behind all of it. The Deception Decoded reporting reveals that the network’s true operators have gone to extraordinary lengths to conceal their identities. Domain registrations are shielded through privacy-protection services; payments are routed through prepaid cards, cryptocurrency, and intermediate shell companies; and contact emails lead only to automated responses or dead addresses. But by following a trail of shared analytic codes, server fingerprints, and inconsistencies in the materials posted across the sites, the investigation concluded that this is not a foreign intelligence operation, but instead a domestic political initiative rooted in the Alberta separatist fringe. The evidence points to a small group of digital activists with a history of promoting western Canadian independence and “Wexit” style messaging, people whose past online behavior suggests an ideological commitment to dissolving the country. The investigation does not claim that the entire Alberta sovereignty movement is illegitimate or that every separatist is involved. Rather, it shows that an organized faction has seized on the capabilities of generative AI to multiply its impact and make its cause appear far more mainstream than it is at the ballot box. The operators have learned that in the new information ecosystem, the most effective way to spread a message is not to persuade people directly but to feed it into the machines that shape public understanding. By creating a fake alternative media universe, they are effectively trying to rewrite the source material from which AI chatbots generate their answers.
That strategy is already producing concrete results. The investigation demonstrated, through a series of controlled prompts, that mainstream chatbots will repeat claims and narratives drawn from the fake sites if asked the right questions. For example, when asked about the constitutional viability of Alberta independence or the fairness of federal energy policy, the AI assistants generated responses that uncritically echoed arguments first published by the fraudulent websites. In one striking demonstration, a chatbot cited one of these synthetic outlets as a credible source for the claim that Ottawa is deliberately impoverishing the province. This is a dangerous development because AI systems do not reason about authority in the way a human journalist would. They treat text as text, and if enough websites assert a proposition, even a false one, the model will synthesize that assertion into its output as if it were established fact. The feed-back loop worsens the problem: once a chatbot repeats a false claim, that output can be copied by users, shared on forums, republished in blog posts, and eventually scraped again by the next generation of AI systems, making the disinformation increasingly durable. The fake sites are therefore not merely pollution; they are a form of trap. They are specifically designed to be encountered by automated tools, and they are calibrated to exploit the vulnerabilities of machine learning systems that have no institutional memory and no capacity for skeptical verification.
The implications of this investigation extend well beyond Alberta and well beyond Canadian politics. Experts quoted in the Deception Decoded report explain that operations like this one represent the next generation of information warfare. In earlier eras, propaganda was designed to be consumed by human audiences, and its effectiveness could be limited by fact-checking, media literacy, and gatekeeping institutions. But the new generation of AI-driven disinformation is designed to manipulate algorithmic intermediaries before they reach human beings. This is a particularly troubling development in the context of Canada-U.S. relations, because Alberta separatism is a narrative that overseas observers and American political actors often misunderstand. If a U.S. policymaker, a businessperson, or even a foreign president asks an AI assistant about the state of Canadian unity, the answer may be sourced from the same fake sites that are pushing a separatist fantasy. That can create a distorted picture of Canada as a country on the verge of collapse, with a restive western population eager to leave the federation. The report notes that such narratives are useful to foreign adversaries seeking to weaken Canada’s sovereign brand, and they can be exploited in trade disputes, security negotiations, and diplomatic conversations. Even if very few actual Albertans support secession, the perception that they do can be generated artificially and amplified automatically, and that perception may be enough to influence decision-making in other capitals.
The investigation concludes with a warning that the problem is not likely to disappear without deliberate intervention. AI companies, search engines, and platform providers all have a degree of responsibility to ensure that the content they use to train and prompt their models is trustworthy, but at present, none of them has adequate systems in place to distinguish between verified news and synthetic propaganda. The Deception Decoded reporting recommends a combination of technical fixes and public education: better provenance standards for digital content, watermarking systems for AI-generated material, stronger pressure on domain registrars and payment processors to remove fraudulent sites, and perhaps most importantly, a public that remains skeptical of unfamiliar outlets and of AI-generated answers themselves. The report also emphasizes that while the fake sites are deceptive, they are not necessarily illegal; they operate in a legal gray zone of false association and anonymous political advocacy. As a result, the response must be societal and cultural as much as legal. For the ordinary citizen, the lesson is clear: the credibility of a newspaper can no longer be taken for granted simply because it appears to be a local publication, and the answer provided by an AI chatbot is only as trustworthy as the sources that went into it. The investigation may not have unmasked every individual responsible, but by exposing the operation’s methods, it has provided a roadmap for understanding and resisting the new machinery of machine-generated disinformation.