TankieTube

Anti-Imperialist and Neo-Stalinist Subcultures on YouTube

Team Members

Luca Ghiselli

Ema Kinsel

Gavin Mueller

Ada Paul

Sean Ryan

Jo Walton

Links

Poster

Contents

Team Members 1

Contents 1

Summary of Key Findings 2

1. Introduction 2

2. Initial Data Sets 2

3. Research Questions 2

4. Methodology 2

5. Findings 2

6. Discussion 2

7. Conclusion 2

8. References 2

Summary of Key Findings

Although tankie YouTubers and Marxist-Leninist organizations ostensibly share political ideology, they express their politics in contrasting ways on YouTube. We view

• Online tankies differ discursively from Marxist-Leninist organizations in their preferred topics

• Marxist-Leninist YouTubers conform to platform dynamics more than organizations

• Organizations conform to social movement activism

• Aesthetic convergence between older organizations and newer YouTube-driven Marxism-Leninism is occurring, from both directions

1. Introduction

“Tankies” is a term that emerged on the British left in the 1970s to refer to defenders of the policies of the Soviet Union (those who supported the USSR sending in the tanks to quash uprisings in Hungary and Czechoslovakia). In contemporary internet parlance, it describes authoritarian anticapitalists who defend “actually existing socialist” states, such as Cuba, China, and Vietnam as well as the historical legacy of the USSR and figures like Stalin and Mao. Tankies subscribe to the ideology of Marxism-Leninism and prioritize anti-imperialism in politics, which tends to mean opposition to U.S. foreign policy, and which translates into a political focus on “geopolitics.” Tankies form a digital political subculture in online spaces: political positions are articulated less as a workable program and more as a set of signifiers that distinguish the group from mainstream political culture (liberalism), as well as other leftwing political subcultures (such as anarchists or democratic socialists) (Tuters and Mueller 2023). This has several implications. First, tankies are a source of extensive digital cultural production: memes, influencers, podcasts, short-form video, and so on. Second, they are a source of fringe political information, including conspiracy theories and misinformation, which then enter broader information ecosystems. Finally, tankies are a strong site of political subject formation. For these reasons, along with the general neglect of research into far-left subcultures in favor of the far-right, we believe tankies are an important object of study.

We also note that Marxism-Leninism persists in the form of small sectarian organizations with a lineage that predates the internet. While small, these organizations have maintained a visible profile in activist and leftwing spaces. This provides ideal grounds for understanding what is distinct in more “natively digital” (or “extremely online”) versions of tankie-ism, in contrast to the approaches of formal Marxist-Leninist political organizations.

2. Initial Data Sets

The Tankies dataset represents a full set of videos published by nine Marxist-Leninist political organizations and 23 “natively digital” YouTuber accounts, from which video transcripts and comment text were extracted. Channel information and comments were collected using a local instance of YouTube Data Tools (2015) resulting in a corpus of ~107.4K organization and ~2.4M YouTuber comments, enriched with video ID, channel ID, and channel title metadata. Postprocessing of the comment corpus normalized emoji and Unicode symbols to shortcode form for the purposes of standardized linguistic analysis and targeted searches for country-flag emojis (e.g., :Vietnam:, :United_Kingdom:). To protect commenter privacy, identifiable author usernames, channel IDs, and comment identifiers were irreversibly pseudonymized with SHA-256 hashing.

Over 6,000 hours of audio were extracted from the video corpus, transcribed, and diarized using a custom pipeline run on the SURF Snellius national supercomputer, requiring approximately 290 GPU hours. The pipeline first downloads audio using the YouTube scraper yt-dlp (2026) before passing it to the WhisperX (Bain et al., 2023) large-v3 model for transcription and speaker diarization. Diarized transcripts are then converted to ConvoKit’s multi-level JSON format for conversation analysis (Chang et al., 2020).

The Tenet/Rumble dataset used in this comparison comes from A Complete Dataset of Tenet Media’s Podcast Videos on Rumble. The dataset contains 560 podcast videos published by Tenet Media on Rumble between November 2023 and September 2024. It includes video metadata, user comments, and video transcriptions. Tenet Media was a U.S.-based right-wing media company that produced political and cultural content during the 2024 U.S. presidential election cycle. According to the dataset description, Tenet Media was later declared by the U.S. government to have been funded by Russia, making it relevant to research on outsourced state-sponsored information operations. This makes the dataset especially useful for this project because it provides a right-wing political transcript corpus, connected to a Russian-funded information operation, that can be compared against Tankie YouTube and organization transcript datasets. This comparison allows the project to examine possible rhetorical overlap between a right-wing pro-Russia media ecosystem and left-wing pro-Russia or anti-imperialist media ecosystems around geopolitics, anti-U.S. framing, Israel, and other shared topics.

3. Research Questions

In what ways do “tankies” constitute a distinct digital political subculture online?

How does tankie content differ from the content produced by Marxist-Leninist organizations? What relationships or contrasts between the two can we find?

How are platform dynamics reshaping radical politics?

4. Methodology

Explain your methodology / approach.

Computational Content Analysis with BERTopic

We used computational methods to explore the discourse of tankie YouTubers and Marxist-Leninist organizations, with the aim of identifying the main themes, recurring patterns, similarities, and differences across the two corpora. More specifically, we applied BERTopic as an exploratory form of computational content analysis, using topic modelling to map the thematic structure of the video transcripts and to guide subsequent qualitative interpretation.

Topic modelling was conducted with BERTopic separately on the YouTuber video transcripts and on the Organizations video transcripts. The two corpora were analyzed independently in order to preserve their respective thematic structures and to enable comparison between the discursive patterns associated with each group. BERTopic was chosen because it combines transformer-based sentence embeddings and class-based TF-IDF topic representation, allowing semantically similar videos to be grouped together and subsequently interpreted through distinctive topic terms (Grootendorst 2022).

This embedding-based approach is particularly suitable for the study, as the discourse of tankie YouTubers and that of Organizations may overlap while differing in vocabulary and narrative framing. Rather than relying only on surface-level word co-occurrence, as in traditional bag-of-words topic models such as LDA, BERTopic allows topics to be shaped by semantic similarity. This made it useful for identifying both shared thematic repertoires and more subtle differences in how similar themes are articulated across the two corpora (Yija, 2024).

Since YouTube transcripts can be very long, each transcript was segmented into shorter textual chunks before topic modelling. This ensured compatibility with the context limits of many embedding models, including those commonly used in BERTopic pipelines, and followed the rationale that BERTopic generally seems to work more effectively on shorter textual units than on long documents. After generating the embeddings, BERTopic’s default pipeline was used for dimensionality reduction and clustering. The default UMAP and HDBSCAN parameters were retained for these steps.

Topic modelling was therefore treated as a quantitative and exploratory form of content analysis, intended to guide subsequent qualitative interpretation rather than to produce fixed or definitive categories. To improve the robustness of the exploration, two BERTopic pipelines were applied separately to both the YouTubers and Organizations corpora. The two pipelines differed in chunk length and embedding model. Comparing the outputs of different pipelines made it possible to test the stability of emerging themes and to explore the corpora at different levels of granularity.

1) In the first pipeline, each video transcript was divided into non-overlapping chunks of 90 words. This relatively short chunk size was used to keep each document within the context limits of the embedding model. The chunks were embedded using BERTopic’s default multilingual sentence-transformer model, “sentence-transformers/paraphrase-multilingual-MiniLM-L12-v2”. A multilingual model was selected because, although the YouTube channels selected were primarily Anglophone, some transcripts contained material in other languages.

2) In the second BERTopic pipeline, we used the multilingual embedding model “ibm-granite/granite-embedding-97m-multilingual-r2” (Awasthy et al., 2026). It offers a suitable compromise between long-context processing and computational efficiency: it supports a maximum sequence length of 32,768 tokens while remaining relatively lightweight, with 97M parameters. This made it appropriate for use in a Google Colab GPU environment, where larger embedding models would have been more computationally expensive. In this pipeline, each video transcript was divided into non-overlapping chunks of 600 words. This allowed us to analyze longer textual passages than in the first pipeline and so to explore the corpus at a coarser level of granularity. At the same time, 600 words are long enough to preserve meaningful context, but not so long that multiple topics would be likely to overlap within the same document.

Building on the results of the initial topic modelling, we identified a set of significant topics for more focused computational and qualitative analysis. Since the full video transcripts were too extensive for close qualitative inspection, we created topic-specific contextual abstracts in order to reduce the analytical scope and focus only on the portions of the videos where the selected themes were explicitly discussed. The abstracts were created through a keyword-based filtering procedure. First, the transcripts were searched for a set of topic-specific keywords derived from the relevant topics. For each keyword match, the sentence containing the keyword was extracted together with the two preceding and two following sentences, in order to preserve the immediate conversational context. The resulting abstracts therefore consisted of contextual abstracts from the videos rather than full transcripts.

In combination with qualitative analysis, these topic-specific abstracts were also used for more focused computational analyses aimed at examining whether the discourse of tankie YouTubers differed from that of organizations on selected themes. For this purpose, the YouTubers and Organizations abstracts were modelled together with BERTopic, while retaining a corpus label for each document. This made it possible to compare how the two groups were positioned within the same thematic structure.

The abstracts were embedded using the model “ibm-granite/granite-embedding-30m-english”. BERTopic was then used to cluster semantically similar abstracts and generate topic representations. After topic assignment, the distribution of YouTubers and Organization abstracts was calculated for each topic. We computed both the raw number of abstracts from each corpus and their relative prevalence within the YouTuber and Organizations subsets. The difference between these corpus-specific prevalence scores was then used to identify topics that were more strongly associated with one group or the other. Finally, a two-dimensional UMAP projection of the embeddings was used to highlight the six most distinctive topics, defined as the topics most strongly associated with each corpus.

To gain understanding of the conspiratorial thinking characteristic of tankie discourse, a set of keywords associated with common conspiratorial narratives surrounding them was identified: conspiracy, CIA, psyop, aliens, color revolution, NED, State Department, regime change, Western-backed, CIA-backed, and Uyghur. The occurrence of these keywords was then compared between the YouTubers' transcripts and the organizations' transcripts, taking the corpus size difference into account. Then, the context in which these terms appeared was inspected using a concordance script.

As an additional comparison, the Tenet/Rumble transcript dataset was compared with the combined Tankie transcript dataset using TF-IDF. The combined Tankie dataset included both Tankie YouTuber transcripts and Marxist-Leninist organization transcripts. TF-IDF was used to identify high-importance overlapping words across both corpora, rather than beginning with a manually selected topic. The output included average TF-IDF scores, shared overlap scores, raw word counts, and video-level frequency counts.

After reviewing the shared word list, “Israel” was selected as a focused keyword because it appeared as one of the strongest overlapping terms across both datasets. Transcript sentences containing “Israel” were then extracted from both groups. A sentence-similarity matcher was used to pair each Tenet/Rumble quote with its closest Tankie quote. This allowed the analysis to move from word-level overlap to quote-level comparison and examine whether shared vocabulary also reflected similar claims, slogans, or political framing.

To explore how YouTubers and organizations might be talking about AI and automation, we first used BERTopic and visualisation using Umap. This wasn’t very informative, since the term didn’t show up as much as we expected, though this in itself was an interesting finding. Shouldn’t job losses from AI be a core socialist concern?

Using a randomized downsampled dataset to get a quick sense of how tankies might be talking about AI, we investigated the frequency of a geopolitical framing. The definitions were quick and crude: do any of a set of geopolitics-related terms appear in the same text span as any of a set of AI-related terms? The lists used were:

ai | artificial intelligence | altman | openai | anthropic | grok | gemini | chatgpt | claude | agentic | data ?center

china | chinese | israel | israeli | iof | idf | zionist entity | gaza | west ?bank | palestine | palentinian | ukraine | russia | russian | amerikkka | israhell

To improve on the AI probe, the full corpora could be tested. Other digital methods might include training Word2Vec models and examining the nearest neighbours of key AI and geopolitical terms, or constructing smaller datasets containing only AI-related passages and applying methods such as ConvoKit’s Fighting Words. Comparisons with other, non-tankie online communities and/or leftist corpora would also be useful.

Fighting Words: Lexical Distinction Across Communities

An adapted version of ConvoKit’s Fighting Words transformer (Chang et al., 2020) was used to identify lexical distinction across the comments and transcripts of both organizational and YouTuber communities. Following the distributional logic of Monroe, Colaresi, and Quinn’s (2008) “Fightin’ Words: Lexical Feature Selection and Evaluation for Identifying the Content of Political Conflict,” the Fighting Words transformer applies a statistical log-odds calculation to distinguish two groups of text from one another, adding an informative Dirichlet prior to account for how common or rare a word is overall, so that frequently used words are not misrepresented as necessarily distinctive.

Out of the box, ConvoKit’s Fighting Words transformer encountered memory issues when run on our large corpora. To accommodate our data, the tool was rebuilt to implement the same statistical formula outlined in Monroe, Colaresi, and Quinn (2008), with greater memory efficiency achieved by batching and aggregating word counts. Outputs from our adapted version were compared against outputs from ConvoKit’s transformer on a smaller sample of our data. Both versions identified the same sets of distinctive words with only a slight variation in order, confirming that our version matches the logic of the original. Our version also produces smaller numerical log-odds scores — though this discrepancy is only representative of a difference in statistical tuning: our version applies a larger prior than ConvoKit’s does in accordance with our corpus size.

5. Findings

We found notable differences in the discourse between online tankies (“YouTubers”) and formal Marxist-Leninist organizations (“orgs”) in their YouTube content (Figure 1,2,3 and 4). Through topic modeling, we identified the most prominent topics in each corpus. While there was some overlap – both groups discussed issues such as Marxist theory, the war in Ukraine, the rise of China, capitalism, and healthcare – there were significant divergences. Formal organizations discussed domestic political issues, such as labor organizing, racist police brutality, and trans and LGBT politics, which were not prominent topics with YouTubers. YouTubers discussed topics not prominent with orgs: North Korea, other political ideologies (anarchism, Maoism, accelerationism), and other political content creators (Contrapoints, Joe Rogan). Further, in shared topics such as China, there were marked differences in what issues were discussed. YouTubers were much more likely to defend China’s policy towards the Uyghur ethnic minority. Diving into the level of utterances, we found numerous conspiracy theories about the attention to the Uyghurs as a U.S. propaganda campaign. Orgs discussed Uyghurs and Xinjiang far less than YouTubers.

Building on these findings, we conducted a more focused qualitative and quantitative analysis of China-related discourse, examining only the videos that mentioned the keywords “China” or “Chinese.” This allowed us to investigate more closely how the discussion of China differed between YouTubers and organizations. The results of this additional focused analysis confirmed the differences already suggested by the topic modeling. In the YouTuber corpus, China-related discourse was strongly connected to the Uyghurs, whereas these themes were much less central in the Organizations corpus. Finally, although the broader topic modelling results show that Korea also appears as a topic in the Organizations corpus, where it seems to be discussed differently from the YouTuber corpus, the China-related discourse map suggests that Korea is a distinctive theme of the YouTuber discourse in this specific thematic context

Figure 1: Representative BERTopic clusters in the Organizations video corpus: 90-word chunks. The figure displays the twenty most prevalent topics in each corpus, based on video transcripts divided into 90-word chunks. The Y-axis lists the most representative terms for each topic, while the X-axis reports their relevance score based on BERTopic’s c-TF-IDF weighting.

Figure 2: Representative BERTopic clusters in the Youtubers video corpus: 90-word chunks. The figure displays the twenty most prevalent topics in each corpus, based on video transcripts divided into 90-word chunks. The Y-axis lists the most representative terms for each topic, while the X-axis reports their relevance score based on BERTopic’s c-TF-IDF weighting.

Figure 3: Representative BERTopic clusters in the Organizations video corpus: 600-word chunks. The figure displays the twenty most prevalent topics in each corpus, based on video transcripts divided into 600-word chunks. The Y-axis lists the most representative terms for each topic, while the X-axis reports their relevance score based on BERTopic’s c-TF-IDF weighting.

Figure 4: Representative BERTopic clusters in the Youtubers video corpus: 600-word chunks. The figure displays the twenty most prevalent topics in each corpus, based on video transcripts divided into 600-word chunks. The Y-axis lists the most representative terms for each topic, while the X-axis reports their relevance score based on BERTopic’s c-TF-IDF weighting.

Figure 5: UMAP projection of China-related BERTopic clusters: most distinctive topics by corpus. Each point represents a China-related YouTube video abstract projected into a two-dimensional UMAP space. Grey points show the full set of abstracts, while colored points identify the six topics with the strongest relative association with each corpus. Blue clusters are more characteristic of organization videos, whereas red clusters are more characteristic of videos by tankie YouTubers.

Frequency of keywords related to common conspiratorial narratives between the YouTubers’ and organizations’ corpora reveal significant overlap after accounting for corpus size (table 1). References to CIA and conspiracy appear at broadly approximate rates. Contrarily, the keywords regime change and western-backed appear more frequently in the organization channels. Meanwhile, YouTubers mention psyop, color revolution, NED, State Department, and CIA-backed more.

Table 1: Keywords related to conspiratorial narratives in Youtubers’ vs. Organizations

Concordance analysis reveals differences in how these topics are discussed as well. While organizations generally avoid discussing controversial, sensitive topics such as China’s alleged human rights violations against Uyghurs, YouTubers do so quite extensively. Concordance analysis on Uyghur, psyop, and CIA reveal that YouTubers tend to explain criticisms on communist nations as narratives fabricated by Western governments (table 2, 3, and 4). References to the Uyghurs frequently appear within arguments that the alleged genocide is a Western propaganda campaign. Terms such as psyop and color revolution are frequently employed to explain protests and media reporting.

Table 2: Concordance run on Uyghur in YouTuber transcripts

Table 3: Concordance ran on CIA in YouTuber comments

Table 4: Concordance run on psyop in YouTuber transcripts

Marxist-Leninist organization channels tend to refrain from commenting on contemporary controversial topics pertaining to communist countries, such as the alleged Uyghur genocide, as keyword frequency reveals. However, topics that scrutinize Western countries, such as the CIA or the Epstein files are as commonly, if not more frequently discussed. Furthermore, concordance analysis suggests that organizations are more likely to reference documented historical examples of Western intervention rather than openly speculating about it. For example, discussions of CIA involvement are typically framed within established historical contexts instead of being used as evidence for contemporary conspiratorial explanations (table 5).

Overall, while both groups express skepticism toward Western governments and media, YouTubers are more likely to frame political events through speculative conspiracy narratives, whereas organizations tend to ground similar critiques in publicly documented evidence, rather than speculation.

Table 5: Concordance ran on CIA in Organization transcripts

The Tenet/Rumble comparison showed that some geopolitical language used by Tankie media also appeared in a right-wing pro-Russia media ecosystem. The clearest overlap appeared around Israel and Gaza. After “Israel” was identified through TF-IDF as a shared high-importance term, quote matching revealed several moments of rhetorical convergence, including similar or near-identical slogans such as “Down with Israel,” as well as overlapping claims about Gaza, genocide, and U.S. support for Israel. This does not prove that the groups are the same or that they are coordinating, but it does show that politically opposed media ecosystems can converge rhetorically around certain anti-U.S. or geopolitical frames. More research is necessary to pull out particular convergences in their context and to theorize their emergence.

Community Vocabularies

Somewhat surprisingly, pronouns and prepositions (e.g., “the,” “she”) appeared in Fighting Words outputs (Figure 6) — words that are expected to be omitted by way of the log-odds frequency weights. Setting these words aside for future investigation, we filtered the output list down to the top 20 most distinctive topical words (Table 6). Within the transcript corpora, organizational channels are distinguished by historical and movement-oriented vocabulary (e.g., “workers,” “labour,” “movement,” “class,” “struggle”), while YouTuber channels skew toward informal “chronically online” speech (e.g., “fucking,” “video,” “guys,” “stream”) with a focus on North Korea (e.g., “dprk,” “kim,” “jong,” “juche”). While these findings reflect channel hosts’ discourse, the comment corpora represent the audiences’ lexicons. Here again, organizational audiences tend toward a more political, party-based vocabulary (e.g., “cpgb,” “acp,” “cpusa,” “org”), though theirs also includes more historical and online figureheads (e.g., “fiona,” “joti,” “lali,” “stalin”). YouTuber audiences follow a similar figure-driven pattern (e.g., “hakim,” “jason,” “vaush,” “peterson”), but appear more varied in their overall political discourse compared to their primarily North Korea-focused channel hosts. Their most distinctive terms reflect a contemporary, Western political discourse: “government,” “climate,” “money,” “companies,” “america.”

Figure 6: Fighting Words outputs for organizational and YouTuber comment corpora (left) and transcript corpora (right).

Table 6: Top 20 most distinctive topical terms derived from Fighting Words analysis.

6. Discussion

Our findings demonstrate that tankies as a digital political subculture differ markedly from the formal Marxist-Leninist organizations that ostensibly share their ideology. YouTubers seem drawn into more controversial topics as a kind of attention-grabbing “edgelordism” leads them to positions that orgs avoid, such as defending North Korea, taking Juche ideology seriously, and openly trafficking in conspiracy theories. This is highly reminiscent of the positioning of non-mainstream views as more authentic that Lewis (2020) found in her study of far-right influencers. This is compounded by the less formal and profanity-laced language used by YouTubers, which can be read as more authentic, though also plausibly more combative. In contrast to orgs, whose public presence tends to be staid and academic in tone, YouTubers emerge in a political ecosystem characterized by confrontational debate driven by charismatic personalities. Though more research needs to be done, we suspect that political positions tend to be personified through the “influencerization” of radical politics. It should be stressed here that this was already a tendency in official communism, which identified its programs as “Marxism-Leninism,” “Stalinism,” “Maoism,” and so on, and was no stranger to cults of personality.

YouTubers’ focus on geopolitical issues at the expense of social justice issues, such as labor politics, racism, and LGBT issues that are central to leftwing politics in the West, could be similarly explained. Geopolitics, which treats nations as stable political actors, represents a legacy of Stalinism and the focus on building “socialism in one country,” but it also presents an attractively simplified Manichaeism, where countries can be divided into good and bad, which can then be cheered on or booed. Since most tankies are based in Western countries, lionizing rival nations – China, Russia, Iran – is also a way to express criticism of one’s own country in its strongest form, and a way to draw distinctions between the tankie position, and more “reformist” left positions that would seek to improve or rehabilitate bad imperialist nations domestically. In short, while the geopolitical orientation does not provide any clear guidelines for political action, it provides ample ground for subcultural distinction-making: what Jelfs (2022) calls “performative disambiguation.”

Due to tankie emphasis on defending AES states, even in ways that strain credulity, they are fertile vectors for disinformation campaigns. Indeed, multiple American anti-imperialist and antiwar organizations have documented links to Russia and China (Hvistendahl et al 2023, BBC 2024). More research needs to be done to explore possible links to state disinformation campaigns. We also strongly suspect that, even without direct state support, tankies are an origin source, or at least a prominent vector, for disinformation and conspiracy theories that are subsequently funneled through a left media ecosystem, similarly to the networked propaganda model developed by Benkler et al (2018).

7. Conclusion

Discuss and interpret the implications of your findings and make recommendations for future research and application, be it societal, academic or technical (or some combination).

This comparison between the Tankie transcript datasets and the Tenet/Rumble dataset suggests that this toolset could be useful for finding moments of rhetorical convergence between different political media ecosystems. By using TF-IDF to identify shared terms, then quote matching to inspect the surrounding transcript language, the analysis was able to locate overlaps around topics such as Israel, Gaza, geopolitics, and anti-U.S. framing. While this does not prove coordination or shared intent between Tankies and Tenet Media, it shows how computational methods can help surface unexpected similarities in slogans, claims, and political framing that would be difficult to find manually.

Overall we find that the dynamics of digital culture and platforms are inflecting left-wing politics. YouTubers have a much more substantial presence on YouTube, and plausibly exercise a greater gravitational pull over radical Marxist analysis of current events than formal organizations. This has a number of implications. We found that YouTubers stressed geopolitics more than orgs, which, ironically, is an intensification of the political outlook Marxism-Leninism took historically: to emphasize geopolitics, and, specifically, the foreign policy of the Soviet Union, as the fulcrum of class struggle, rather than worker-based movements within nations. In this sense, class struggle became “nationalized.” However, since the fall of the Soviet Union and the decline of official communist parties and sects, this has become less relevant. Contemporary Marxist-Leninist organizations continue to prioritize a critique of imperialism in their political line, but, in order to continue to recruit, they must participate in activist social movements, where labor and identity issues are more prominent. YouTubers also seek growth, but in viewcount and subscriber base, rather than activist cadre. This means that its topical focus conforms to platform dynamics rather than political trends: a greater focus on online “drama” and interpersonal controversy, as well as conspiracy theory and “debunking.” Again, we note that these practices bear a strong resemblance to practices rife in the historical communist movement: sectarian denunciation, conspiracies and subterfuge, and a critique of mainstream narratives about politics and history as “ideological.” More research needs to be done to differentiate the combined ecosystems of tankie YouTubers and Marxist-Leninist organizations from other leftwing tendencies to evaluate how characteristic these practices are to the wider world of Marxist-Leninist YouTube as a whole, and how they might differ from a more general Marxist and anticapitalist YouTube channels, and left-wing and socialist YouTube channels, respectively. One hypothesis, rather grand and at this point speculative, is that historical Marxism-Leninism planted many of the seeds that have now found especially fertile ground in the dynamics of platforms and the attention economy in the 21st Century. Discerning the contours of this odd convergence is the goal of a larger project.

We note two cases from our corpus that speak to an intriguing convergence of orgs and YouTubers. First, the Communist Party of Great Britain (Marxist-Leninist) has a remarkably professional (by the terms of the platform) YouTube presence as “Proletarian TV.” The CPGB-ML is one of the most extreme Marxist-Leninist organizations, defending North Korea (the source of its foundational split), supporting Brexit, and carrying portraits of Josef Stalin in parades. It may have found YouTube to be a supportive environment for its positions. Second, several prominent YouTubers, including Haz and Midwestern Marx (Ethan Liger), have started a Stalinist organization called the American Communist Party (not to be confused with the Communist Party USA). This “hybrid organization” continues to traffic in inflammatory views (and is, notably, hostile to LGBT causes and openly nationalistic), while also engaging in highly documented “IRL” activities around mutual aid and self-defense training. This convergence of activist organizations and with internet-derived positioning and practices points towards the emergence of Stalinist “digital parties” (Gerbaudo 2018), with continued echoes through the left-wing politics and activism.

8. References

Awasthy, P., Trivedi, A., Yang, Y., Barker, K., Li, Y., Iyer, B., Franz, M., Bross, J., Doshi, M., P, V., Kumar, V., Ward, T., Daniels, A., Lee, M., Lastras, L., Sen, J., & Florian, R. (2026). Granite embedding multilingual R2 models. arXiv. https://arxiv.org/abs/2605.13521

Bain, M., Huh, J., Han, T., & Zisserman, A. (2023). WhisperX: Time-accurate speech transcription of long-form audio. Proceedings of INTERSPEECH 2023. https://doi.org/10.21437/Interspeech.2023-78

“Black rights activists convicted over Russian links.” (13 September 2024). BBC News. https://www.bbc.com/news/articles/c624ppg14jpo

Chang, J. P., Chiam C., Fu, L., Wang, A., Zhang, J., Danescu-Niculescu-Mizil, C. (2020). “ConvoKit: A Toolkit for the Analysis of Conversations”. Proceedings of SIGDIAL.

Gerbaudo, P. (2018). The Digital Party: Political Organisation and Online Democracy. Pluto Press.

Grootendorst, M. (2022). “BERTopic: Neural Topic Modeling with a Class-Based TF-IDF Procedure.” ArXiv, March. https://doi.org/10.48550/arxiv.2203.05794.

Hvistendahl, M., Fahrenthold, D. A., & Chutel, L. (2023, August 5). A global web of Chinese propaganda leads to a U.S. tech mogul. The New York Times. https://www.nytimes.com/2023/08/05/world/europe/neville-roy-singham-china-propaganda.html

Jelfs T. (2021). Will the revolution be podcast?: Marxism and the culture of “millennial socialism” in the United States. In: Phelps C, Vandome R, eds. Marxism and America: New Appraisals. Manchester University Press; :241-263.

Kurek, L., Zheng, K., Gilbert, E., & Budak, C. (2025). A Complete Dataset of Tenet Media’s Podcast Videos on Rumble [Dataset]. Zenodo. https://doi.org/10.5281/zenodo.14629410

Lewis, R. (2020). “This Is What the News Won’t Show You”: YouTube Creators and the Reactionary Politics of Micro-celebrity. Television & New Media, 21(2), 201-217.

Monroe, B. L., Colaresi, M. P., & Quinn, K. M. (2008). Fightin’ Words: Lexical Feature Selection and Evaluation for Identifying the Content of Political Conflict. Political Analysis, 16(4), 372–403. https://doi.org/10.1093/pan/mpn018

Rieder, B. (2015). YouTube Data Tools (Version 1.43) [Software]. Available from https://ytdt.digitalmethods.net.

Tuters, M., & Mueller, G. (2024). Revolution by other memes: on the playful subcultures of r/PoliticalCompassMemes. Information, Communication & Society, 27(12), 2260–2280. https://doi.org/10.1080/1369118X.2024.2423340

Yija, L. 2024. “Comparison of LDA and BERTopic in News Topic Modeling: A Case Study of The New York Times’ Reports on China.” Pacific International Journal 7, no. 3. https://doi.org/10.55014/pij.v7i3.616

yt-dlp. (2026). yt-dlp [Computer software]. GitHub. https://github.com/yt-dlp/yt-dlp

-- GavinMueller - 27 Jul 2026