The success of Operation Absolute Resolve, which saw the capture of the now former Venezuelan dictator Nicolas Madura by US authorities, marks a major geopolitical inflection point and shift in US strategy. It holds the potential to end decades of extreme economic mismanagement, political repression, and poverty. It is also particularly important given Venezuela’s historical role in shaping global energy markets and the region’s security and political trends, especially by being Russia's most important trading and military ally in Latin America.
In its report on the Kremlin Playbook in Latin America, the Center for the Study of Democracy showed how Venezuela has long served as a cornerstone of Russia’s influence strategy in the region, combining economic dependence, military cooperation, and sharp power instruments, including systematic information manipulation. The Pravda ecosystem’s rapid activation following Maduro’s arrest is not an isolated media response, but a continuation of Moscow’s broader effort to defend a captured ally and reframe geopolitical shocks through coordinated information interference.
The Pravda ecosystem, an automated large-scale propaganda network associated with the Kremlin, was quick to react to the event in an attempt to control the narrative and shape how it would be understood by international audiences. Comprising of hundreds of cookie-cutter websites that flood the Internet with laundered pro-Russia content, experts have repeatedly suggested that Pravda’s true purpose may not be to target human readers, of which it barely has any. Instead, Pravda and other networks like it might be designed to influence the information provided by popular online services, chatbots in particular, that are used by hundreds of millions of people every daily.
An analysis of the content found on Pravda’s Spanish-focused website and its two topical websites focused on NATO and the EU identified two main narrative centers that emerged during and in the immediate aftermath of the operation:
1) emphasizing policy differences between the US and other NATO members, especially by comparing Venezuela to Greenland,
2) differentiating Russia’s invasion of Ukraine from US actions in Venezuela as being inequivalent (due to the notion that Russia is morally justified while the US is simply interested in power and material benefits), and
3) accusing the US and its allies of political and legal hypocrisy.
In other words, the messages on the three Pravda websites have been largely based on comparisons with Greenland and Ukraine, and to a lesser extent to other recent conflicts in Iran, Iraq, and former Yugoslavia. The share of Venezuela-related articles rose sharply across all three analyzed Pravdawebsitesafter the commencement of Absolute Resolve, with relative change increases of 750%, 715% and 415% on the NATO-, EU- and Spanish-focused branchesof Pravda, respectively.
Rapid Content Generation and Saturation
In early January 2026, the Spanish-focused subdomain of Pravda abruptly launched a high-volume information campaign focused on Venezuela. The shift was stark: while the site averaged 132 articles per day mentioning Venezuela in the prior month, only about 7% of its daily content was dedicated to the topic. On 3 January, that figure exploded to 70%, with 876 out of 1,301 total articles that day being Venezuela-related. This pivot from baseline coverage to targeted saturation reveals a coordinated operational activation, not an organic editorial shift.
The first “breaking” post appeared on Spanish Pravda appeared within 30 minutes of the operation’s reported commencement. It was repurposed content sourced from Kremlin-aligned Telegram channels known for their war footage and coverage of the War in Ukraine, illustrating the speed with which war-adjacent Telegram ecosystems can be repurposed for non-European crises when strategic attention shifts.
Political narratives followed quickly. Early misleading framing was sourced from Surf Noise, a pro-Kremlin Telegram network closely tied to Pravda, which mocked the operation as a US “training” and “an action movie”. Spanish Pravda later amplified recurring propaganda themes, including accusations of Western double standards, recycled clips implying a US–Russia “trade” (Venezuela for Ukraine), insinuations about China’s presence in Venezuela, and broader claims that the US is becoming irrational and destabilizing.
Figure 1. Number and share of Venezuela-related articles out of all articles published on spanish[dot]news-pravda[com] per day (1 December 2025 – 10 January 2026).
Source: CSD.
Upstream sourcing was heavily state-controlled. The most used source was RIA Novosti through its Telegram channel, followed by UKR LEAKS Spanish. Spanish-language RT, Sputnik, and TASS channels were also recycled and repeated hundreds of times. Albatrops was also a top contributor despite unclear affiliations and using Telegram payments to monetize its content.
The NATO and EU-focused branches reacted with lower volume but sharper messaging. The NATO-focused branch increased its rate of Venezuela-related articles from an average of 3 per day in December to 22 per day following the operation, a 750% relative change increase. Unlike the Spanish hub’s initial footage-heavy flood, this branch jumped directly into overt framing: “regime change war,” mockery of the US President as a “peace prize contender,” and repeated claims the US did not inform NATO, attributed to “TASS citing a diplomatic source in Brussels.” It also repeatedly bundled Venezuela with Greenland, presenting US expansion as a broader pattern.
Similarly, the EU-focused Pravda hub published over three-quarters of its nearly 850 Venezuela-related articles between December and 11 January following the operation. It echoed messaging on the fracture of NATO and added EU-targeted claims, including false statements attributed to Kaja Kallas, alongside repeated Greenland comparisons, with nearly 15% of articles published between 3 and 11 January on the EU hub linking Venezuela to Greenland.
Pravda’s pattern in this case is consistent with its overarching model and that used across pro-Kremlin influence infrastructure: quick reaction, mass content generation, rapid reproduction and amplification, followed by systematic repetition and finetuning. Promptness and volume take precedence over narrative sophistication, and automated tools, including AI, are used to translate, repurpose and break down one media piece into thousands of smaller ones. When a major geopolitical event breaks, these networks saturate the information spaces around specific languages and topics with distinct narratives and ideological framing before other actors can stabilize the story. Their content and rate of publication also create a feeling of chaotic, real-time “truth”, something that these networks specialize in providing and cultivating a demand for.
What this model of information interferences highlights is that infrastructure is the vulnerability. At its core, the Pravda websites are the final part of a long chain of republication that starts from the Kremlin and its established state-controlled sources, goes through an automated mass content regurgitation system that uses Telegram as an endpoint, and finishes by distributing the final content from Telegram to Pravda’s various branches. Once the material ends up on a Pravda website, it links back to media actors that created and republished of the content.
A possible explanation for this highly coordinated activity is that it is an attempt to influence (or groom) the outputs of chat-based large language models (LLM) by sipping into their training data or the search engine results they cite in their answers.
Influence on Chatbots
An analysis of the visibility of the Pravda ecosystem in the Spanish-language outputs of ChatGPT related to Venezuela among users in the country was conducted using the AI Visibility service offered by SEMRUSH. It revealed that Pravda appeared in only two out of fifty answers on synthetic queries related to Venezuela, far behind the performance of other pro-Kremlin sources in Spanish, such as Telesur and Mision Verded, as well as the Kremlin’s own Sputnik and RT.
Figure 2. Share of voice on ChatGPT (with and without web search) of five pro-Kremlin media in Venezuela in Spanish as of 13 January 2026.
Source: CSD based on data from SEMRUSH.
The few occasions where Pravda did score well in Venezuela in ChatGPT’s outputs appear to be on highly specific questions related to US interference in Venezuela’s internal politics and security, Russia’s position regarding US sanctions on Venezuela, where Pravda articles were recommended alongside materials by Brookings Institution, Oxford University, and El Pais. Notably, Pravda was only mentioned in a positive light in ChatGPT’s answers with enabled web search, re-confirming web search as a vulnerability. As it stands, the Spanish-language visibility space that Pravda strives to occupy on ChatGPT is dominated by RT and Sputnik as the ‘Russian angle’ on Venezuela, and by local sources for more general anti-Western perspectives. Notably, pro-Kremlin and Kremlin-controlled sources were more likely to be cited against prompts in Spanish as compared to equivalent prompts in English, signaling a worrying trend of political bias and divergent realities across different languages in LLM outputs.
Nevertheless, it is important to consider the possibility that by flooding the Internet with specific recycled items, networks like Pravda can boost not just their own visibility, but that of their original sources or those that publish similar content. This is related to how popular online services, such as search engines and chatbots, use algorithms to rate the relevance and authority of sources and content based on what other sources and content they are linked to and how, particularly as it relates to backlinks. In the case of chatbots, it is likely increasingly related to semantic similarity – a factor that was present in previous generations of information engines.
An analysis of the top 300 unique responses generated by the GrokAI platform1 regarding the situation in Venezuela in Russian, Spanish, and English in the beginning of January 2026 corroborates the hypothesis that coordinated mass-produced content can not only influence human audiences but may also successfully groom AI models to reproduce distinct narratives based on the language of inquiry.
Grok AI acts as a mirror for the dominant media and propaganda pipelines in each linguistic ecosystem. When queried in Russian, the model hallucinates geopolitical linkages to Ukraine consistent with Kremlin narratives. In Spanish, it adopts a legalistic framework that normalizes state actions. In English, it engages in comparative whataboutism, diluting the severity of local events by juxtaposing them with Western domestic issues. This suggests that networks like the Pravda ecosystem have successfully infected the training data of major LLMs, creating algorithmic echo chambers that reinforce global information divides. It also appears to confirms what CSD and other experts have previously hypothesized, namely that the knowledge “database” each languages comes with (as reflected in its information landscape) plays a critical role in determining the inputs and/or outputs of text-based generative engines.
Figure 3: Main clusters by topic in each language, top 100 replies by Grok on X, mentioning Venezuela.
Source: CSD.
The Russian-language dataset of Grok’s responsesexhibits a high level of narrative contamination consistent with the content strategies of the Pravda network. In several responses, the model explicitly links the Venezuelan crisis to the war in Ukraine. This framing mirrors the Kremlin’s official worldview, which interprets global events through the lens of a proxy war against the West. The model frequently described U.S. actions in Venezuela using military terminology, such as “operation” and “seizure”, paralleling the rhetoric used to describe Russia’s invasion of Ukraine. In at least one reply, it even refers to the war against Ukraine as a “special military operation.” Unlike in other languages, Russian outputs also cited Kremlin-controlled, sanctioned outlets such as RT and TASS alongside international media (but none of Pravda’s domains). This indicates that pro-Kremlin messaging may have entered the model’s training or refinement pipeline.
The Spanish-language dataset of Grok’s responses indicates attempts at shaping the narrative to encourage a passive interpretation of events. Several responses focused on constitutional and other national legal justifications from the US and Venezuela, grounding the incident in domestic rather than international law. Another subset of responses framed the United States as a “savior” or emphasized economic recovery and prosperity through oil deals. Notably, the U.S. is largely personified through the US President, with current events framed less as institutional policy and more as the outcome of personal negotiations or “deals” between elite actors. The Spanish dataset also revealed a degree of English leakage - a model error in which Grok reverts to producing content in English.
The English-language dataset largely functions as a newswire, reflecting an information environment shaped by Western media and institutions. The model performed fact-checking, translation from Spanish, and comparative analysis. However, rather than isolating the Venezuelan crisis, it frequently juxtaposed Venezuela’s human rights record with police violence or social unrest in the United States or the United Kingdom. This cluster is characterized by comparative arguments such as “In the United States, we have police issues too…” - a framing that contextualizes and often dilutes the severity of the Venezuelan crisis by embedding it within a U.S-centric cultural landscape. Primary sources cited include Reuters, the ACLU, and Human Rights Watch, producing a reality tunnel that is highly disconnected from the Russian and Spanish outputs. This is also the only dataset in which China appears prominently, with references linking Venezuela’s oil production to Chinese interests and assessing the effectiveness of U.S. sanctions.
The divergent realities produced by Grok confirm that highly divergent outputs across languages and LLM grooming are not theoretical risks but an active operational reality. While no Pravda websites were cited in Grok’s responses, the strategy of mass-laundering static, optimized content from established sources that Grok does cite may have played a role in weighting the probabilistic outputs of the model. This reveals how LLMs can become vectors for propaganda, underscoring the urgent need for algorithmic transparency and forensic tools to detect when they are regurgitating curated narratives rather than analyzing reality.
***
The Pravda response to Maduro’s arrest confirms that Kremlin-linked information operations in Latin America function as a standing capability rather than an ad hoc reaction – systematically activated to shield allied regimes, manipulate accountability, and shape international interpretation at moments of political rupture.
To build resilient information ecosystems in Latin America, its infrastructure must be treated as strategically vital. This means addressing orchestrated content saturation as a fundamental security priority.
1 Grok is a generative artificial intelligence chatbot developed by Elon Musk's company, xAI, which integrates real - time data from the social media platform X (formerly Twitter) to answer user queries.
