By Munware Alam Nirjhor
•••
Every year, when Oxford University Press releases its Word of the Year shortlist, it is more than a curiosity for language lovers. How often a word is used, in what context and by whom tells us what a society is thinking about and what worries it. In November 2026, Oxford’s language experts analysed a corpus of about 32 billion words and picked four — doomerism, microdrama, post-literate and slopaganda. Public voting is open until 24 November. At first glance, the four words come from four different worlds. But to an information warfare researcher, they are four pieces of the same picture. One speaks to the psychology of fear, one to a new content format, one to the decline of reading, and the fourth names the weapon that can exploit the other three to change what people believe and how they decide. This article focuses on that fourth word, slopaganda.
Oxford defines slopaganda as low-quality propaganda created with generative AI and widely shared online, especially on social media. The word blends slop and propaganda. Slop, in this sense, made Oxford’s 2024 shortlist and refers to low-quality AI content spread without discrimination. Propaganda means the systematic, often biased or misleading, promotion of a cause. According to Oxford, the word’s relative frequency rose by more than 1,300 percent over the past year, and it often appears as AI slopaganda or slopaganda videos. The concern centres on AI-generated political messaging, its power to shape public opinion, and a wider crisis of trust in society’s ability to tell fact from fiction.
When a word moves from an academic journal to a Word of the Year shortlist in about a year, it means the thing it describes is no longer confined to expert debate. It has entered everyday life. In Facebook feeds, vertical TikTok videos and forwarded WhatsApp clips, people now see images and videos that no camera ever captured. Slopaganda gives that experience a name. And naming something is the first step to analysing it.
The word slopaganda was first used in this sense by three researchers, Michał Klincewicz of Tilburg University, Mark Alfano of Macquarie University, and independent researcher Amir Ebrahimi Fard. Their paper, Slopaganda: The interaction between propaganda and generative AI, was published in 2025 in volume 12 of the Swedish philosophy journal Filosofiska Notiser. The authors define propaganda as the intentional manipulation of beliefs and attitudes to achieve political ends, and slop as unwanted AI-generated content. Put together, slopaganda is unwanted AI-generated content spread to manipulate beliefs for political goals.
The paper opens with a figure that shows the scale of the problem. Citing The Guardian, the authors note that News Corp Australia was producing around 3,000 local news stories a week using generative AI. In other words, work that once needed a full newsroom can now be automated. The researchers argue that this kind of AI slop will create epistemic challenges in the years ahead, meaning it will damage the very process by which people come to know things, decide what to believe and decide whom to trust.
The researchers also describe the social structure behind slopaganda. At the top of the pyramid are those who set the message, give direction and plan its spread. In the middle are the implementers who handle the technical work of producing and distributing content. At the bottom is the mass audience, who are both the targets and, often without knowing it, the distributors. An analysis from the University of Technology Sydney points out that, like traditional propaganda, slopaganda floods the information space with plausible-sounding material to distract, mislead or shape ideology. The difference is that it spreads instantly and at unprecedented scale through social bots and algorithms, it can be tailored to individuals, and anyone, anywhere, can make it at any time.
The first step is the collapse of production costs. Writing a convincing propaganda article used to take a skilled writer, language ability and time. A 2024 study by Goldstein and colleagues in PNAS Nexus showed how this has changed. They ran a pre-registered survey experiment with 8,221 respondents in the United States, comparing real propaganda articles from Iranian and Russian covert campaigns with articles written by GPT-3. The AI-written articles were almost as persuasive as the real ones. When the researchers edited the prompts or picked the best outputs, the AI articles became, in some cases, just as persuasive as the originals. Their conclusion was that propagandists could produce convincing content with very little effort. It is worth remembering that this was a fairly old model. Today’s models are far more capable.
The second step is the move from the lab to the real world. Another study, published in PNAS Nexus in 2025, examined a state-linked propaganda website with ties to Russia and found that after adopting generative AI, the site increased both the volume and the range of the disinformation it produced. In short, AI has already begun to change the size and reach of state-backed campaigns. Tech companies’ own threat reports tell the same story. In its June 2025 report, OpenAI said it had shut down ten operations misusing its tools within three months, four of which likely originated in China. One of them, which OpenAI called Sneer Review, used ChatGPT to write short comments posted on TikTok, X, Reddit and Facebook in English, Chinese and Urdu. The mention of Urdu matters for South Asian readers, because it shows the region’s languages are already on this factory’s production line. The report also noted that these operations created both a post and the replies to it, so that the discussion would look natural.
The third step is distribution and algorithmic spread. Slopaganda’s power lies not in the quality of any single post but in sheer numbers. Platform algorithms reward engagement, and emotions such as anger, fear and surprise generate the most engagement. So emotional AI content spreads fast even when it is poor quality. A single operator can now produce hundreds of versions, see which one performs best, and then push that one further. What the advertising industry calls A/B testing can now be done for propaganda at almost no cost.
The fourth step is exploiting psychological weaknesses. Psychology has a well-established finding called the illusory truth effect people are more likely to believe a statement the more often they hear it. Hasher and colleagues first showed this in the lab in 1977. A 2015 study by Fazio and colleagues added a more worrying finding. People can come to believe a repeated false statement even when they already know the correct answer. The researchers called this knowledge neglect, meaning people fail to use what they know when faced with information that feels familiar and easy to process. Slopaganda targets exactly this weakness. When a false claim keeps returning across hundreds of different images, videos and posts, each one may be weak on its own, but together they create a sense of familiarity, and that familiarity passes for truth.
In the fifth step, slopaganda industrialises an old technique. In 2016, Christopher Paul and Miriam Matthews of the RAND Corporation described Russia’s propaganda model as a firehose of falsehood. In their view, it has four features, it is high-volume and multichannel; it is rapid, continuous and repetitive; it has no commitment to truth; and it has no commitment to consistency. Conventional communication theory says that truthful, credible and consistent messages persuade best. But RAND showed that psychological research also explains why this opposite model works. In 2016, running such a firehose took state funding, troll farms and hundreds of staff. Generative AI has pushed that cost close to zero. Slopaganda can therefore be called the democratisation of the firehose: a technique once reserved for great powers is now in the hands of a party’s IT cell, a content farm, or even a single person.
The sixth step is the subtlest and perhaps the most harmful. In 2019, legal scholars Bobby Chesney and Danielle Citron introduced the idea of the liar’s dividend in the California Law Review. Their argument is that the more the public learns about deepfakes, the more easily liars seeking to escape accountability can dismiss real evidence as fake. An analysis by Daniel Schiff, published by Brookings, notes that several studies found politicians can gain public support by falsely claiming that real scandals about them are fake. This means slopaganda strikes twice. First, it makes lies believable. Second, it makes the truth look doubtful. The second blow is more dangerous in the long run, because it weakens the value of evidence, which is the foundation of journalism, the courts and accountability.
Which theoretical framework does slopaganda belong to? There is no single answer. Information warfare is essentially a fight for control of the information environment, which narrative dominates, which information reaches people and which is buried. Psychological warfare has a narrower aim, changing the emotions, morale and behaviour of an opponent or a target population. Slopaganda sits where the two meet. It floods the information space, which is an information warfare tactic, and it changes behaviour by stirring fear, hatred and confusion, which is a psychological warfare tactic.
Recent military and academic thinking increasingly points to a third framework, cognitive warfare. Since 2021, NATO’s Allied Command Transformation has been developing a Cognitive Warfare Concept, with input from more than twenty member states. In NATO’s description, cognitive warfare combines cyber, information, psychological and social engineering capabilities, and its core aim is to attack and degrade rational thinking. NATO has also said that cognitive warfare is not the means by which we fight, but the fight itself. In a 2024 analysis in Frontiers in Big Data, Christoph Deppe and Gary Schaal of Helmut Schmidt University in Germany describe cognitive warfare as using human cognition and technology to disrupt, influence or change how people make decisions. Many researchers see it as part of hybrid warfare. China’s Three Warfares concept, which combines psychological warfare, public opinion warfare and legal warfare, belongs to the same family.
This article takes the view that slopaganda is best understood as a layered model. Information warfare is the arena, the combined information space of social media, messaging apps and news outlets. Psychological warfare is the mechanism, the techniques of emotion, repetition and social proof used to influence people. Cognitive warfare is the end goal, the decision-making capacity of individuals and societies. Slopaganda is the ammunition that works across all three layers at once, and its supply is now almost unlimited. There is also a commercial dimension. In many cases, there is no state or party behind slopaganda, only the lure of monetisation. In an Arab News report on the 2025 India-Pakistan conflict, one expert observed that the conflict had turned into content, and the content had turned into currency. Political propaganda and the business of viral income have become almost indistinguishable.
Oxford’s other three words make this framework clearer. Doomerism, the extreme pessimistic belief that disaster from threats like AI or climate change is inevitable, is slopaganda’s emotional fuel. In societies where fear and helplessness are widespread, conspiracy theories and panic-inducing content find easy footing. Microdrama, a melodramatic series of very short episodes made in vertical format for smartphones, is slopaganda’s grammar. According to Oxford, the word’s usage tripled in 2026, and the format originated in China. Slopaganda videos speak this same language, short, fast, emotional and story-driven. And post-literate, describing a society where literacy and the importance of the written word have declined significantly, is slopaganda’s fertile ground. Oxford itself notes concern about a possible link between declining reading, weaker critical thinking and misinformation. In a society that reads less and watches more, a written fact-check is fighting an uneven battle against a convincing video. Together, the four words form a complete ecosystem, emotion, format, vulnerability and weapon.
In South Asia, slopaganda is not a future threat; it has already been tested. After the Pahalgam attack in April 2025, as India-Pakistan tensions rose, AI-generated fake videos, old war footage and invented stories spread like wildfire on X, WhatsApp, Facebook and YouTube. One viral AI image claimed that Rawalpindi Stadium in Pakistan had been destroyed in an attack, and it was viewed millions of times. Fake videos of leaders admitting defeat and cloned voices also circulated. The Institute for Strategic Dialogue found that many false claims viewed millions of times faced no moderation, and some made their way into national news media. A Bellingcat investigation showed how one deepfake video was published by mainstream outlets, after which NDTV and The Statesman removed their reports without explanation. Fake content flowed from both sides, and fact-checkers in both countries documented the other side’s campaigns.
An analysis by Tech Policy Press identified the most dangerous process in this crisis. A social media forward reaches a newsroom, the newsroom broadcasts it and lends the fake content legitimacy, and the content then returns to social media carrying that legitimacy. In a conflict, speed often beats accuracy. The analysis also showed that standard verification methods such as reverse image search and metadata checks do not work against newly generated synthetic images, because those images have no earlier existence to trace. AI was also used in Pakistan’s 2024 election. Supporters of Imran Khan created videos using the jailed leader’s image and voice to urge people to vote, while another fake video showed him calling for an election boycott, confusing his supporters.
Several structural factors make this region more vulnerable. First, it has huge young, mobile-first populations, many of whom came online for the first time directly through video-based platforms. Second, digital and media literacy is uneven. Third, there are deep historical fault lines of religion, community and rivalry between states that are easy to inflame. Fourth, large language models can now produce near-flawless text and voice in Bengali, Hindi and Urdu, while platforms’ local-language moderation remains relatively weak. An analysis by the Global Network on Extremism and Technology, looking at the spread of AI content after Pahalgam, identified key platform weaknesses, no mandatory labels for AI content, and too much reliance on keyword-based moderation rather than context-aware detection. Fifth, when content spreads in encrypted, closed groups such as WhatsApp, it stays out of sight of researchers and fact-checkers.
Bangladesh’s 13th parliamentary election on 12 February 2026 is an important case study for regional research on slopaganda. An analysis by Zulker Naeen published in Global Voices called it the country’s first AI-saturated election. Reviewing the work of fact-checkers such as FactWatch, BOOM, Dismislab and Rumor Scanner, the analysis identified 72 pieces of AI-manipulated content between December 2025 and February 2026 that pushed false narratives and gained significant traction online. About 49 percent showed political leaders doing things that never happened, and about 28 percent attached fake quotes to AI-edited photocards designed to look like those of trusted news outlets. The BNP was the most targeted party, with 47 cases. Jamaat-e-Islami faced 13 attacks, and six counter-campaigns were linked to the Awami League. The National Citizen Party, especially Nahid Islam, was also targeted with edited photocards.
Other investigations show the same pattern. Between 14 December 2025 and 14 January 2026, The Daily Star documented 97 pieces of AI-generated content on Facebook, mostly videos, from 11 pages, 6 profiles and 2 groups. In a one-month period from mid-December to mid-January, the same newspaper documented 220 cases of disinformation spread by various party-aligned actors. A January 2026 investigation by Dismislab found that a Facebook page called Uttorbongo Television had posted 35 videos assessed to be AI-generated or AI-edited, in which lifelike characters were presented as ordinary voters. This is a new form of slopaganda, not a fake statement from a leader, but fake public opinion. FactWatch debunked a total of 94 false claims between 11 December 2025 and 17 February 2026.
A study by Activate Rights, titled Faith, Fear, and Falsehoods, concluded that coordinated narrative manipulation was a structural feature of the election, not a set of isolated incidents. The study found that AI-generated explicit images were used against women candidates, including Tasnim Jara and Rumeen Farhana. Women made up only about 4 percent of candidates, yet they faced a disproportionate share of online abuse. AI-generated minority characters were used to spread fear, and many narratives from across the border used old or out-of-context footage to portray Bangladesh as a hub of extremism and minority persecution. An AFP report quoted analysts and fact-checkers as saying that about 90 percent of disinformation about the country originated in India. In the same report, former Election Commission official Jasmine Tuli warned that AI-generated images posed a major threat to a country like Bangladesh, where awareness of how to check information is low, and that they could mislead voters. bdnews24.com reported that the government admitted it had no direct power to remove such content and that removal depended on the platforms.
The context matters for understanding the scale. According to DataReportal, Bangladesh had 68.3 million social media user identities in October 2026, equal to 38.3 percent of the population. StatCounter data show that Facebook accounted for about 84 percent of the country’s social media traffic in June 2026. In other words, Bangladesh’s information space is concentrated on a single platform, which makes it a convenient battlefield for slopaganda. Several analytical lessons follow from Bangladesh’s experience. First, photocard culture, the habit of reading news through image cards carrying a media outlet’s logo, has allowed slopaganda to hijack the credibility of trusted brands. This is a direct concern for media watchdog organisations. Second, slopaganda comes from many sources, domestic party networks, cross-border campaigns and content farms chasing views and income all operated in the same space. Third, it has a clear gendered dimension that can directly discourage women’s political participation. Fourth, the tactic of faking public opinion through fake voter characters is likely to become more sophisticated.
Global risk assessments reflect this concern. The World Economic Forum’s Global Risks Report 2026 ranked misinformation and disinformation as the second biggest risk over the next two years, just behind geoeconomic confrontation. The same report placed adverse outcomes of AI technologies among the top risks. A Forum analysis describes it as a risk that fuels or worsens almost every other risk. Against this background, we can outline several likely paths for slopaganda’s spread in South Asia. These are not predictions but analytical projections based on current trends.
The first path is automated persona networks. So far, most slopaganda is created and posted by people. But as AI agent technology advances, thousands of fake accounts will be able to post, comment and hold conversations on their own, each with its own personality, regional dialect and interests. The practice OpenAI observed of generating a post and its replies together is an early form of this. The second path is hyper-personalisation, one message in the Chittagonian dialect, another with a Sylheti accent, each tailored to a specific community’s fears and hopes. The third path is microdrama-style serial propaganda. Instead of single posts, there will be episodic fictional stories in which AI characters build a political or communal narrative week after week. Viewers will think it is entertainment, when it is really narrative-building.
The fourth path is slopaganda for hire. A market for paid campaigns, whether for elections, business rivalry or personal feuds, could grow further. The fifth path is the normalisation of the liar’s dividend. The more people know that videos can be faked, the more corrupt or criminal figures will dismiss real evidence as AI. For investigative journalism, this is an existential crisis. The sixth path is cognitive fatigue. When everything feels suspicious, many people stop trying to verify, turn away from politics, or retreat into their own side’s narrative. This is where slopaganda meets doomerism, distrust leads to despair, and despair leads to disengagement. And that disengagement is more damaging to democracy than any single piece of fake news.
We can also identify the situations in Bangladesh and South Asia where slopaganda is likely to be most active- local and national elections, moments of communal tension, natural disasters such as floods or cyclones, border incidents, the Rohingya issue, and diplomatic strains between states. Every crisis creates an information vacuum, and synthetic content reaches people faster than the truth to fill it. At the same time, the post-literate trend can create a vicious cycle, the less people read, the more they rely on video, and the more they rely on video, the more exposed they are to slopaganda.
There is no single solution to slopaganda. Klincewicz and colleagues themselves propose interventions at three levels- psychological, technological and political-economic. At the psychological level, one of the best-proven methods is prebunking. In a 2022 study in Science Advances, Jon Roozenbeek, Sander van der Linden and colleagues at the University of Cambridge showed that five short videos explaining common manipulation techniques, such as emotionally charged language, false dichotomies, scapegoating and personal attacks, improved people’s ability to recognise these techniques. The study involved 6,464 people across six controlled experiments and 22,632 people in a field study on YouTube. The researchers found the effect was consistent across political views, education levels and personality types. RAND made a similar recommendation in 2016, warning people in advance, or reaching them first with the truth, works better than debunking afterwards. Bangladesh now urgently needs prebunking programmes in Bengali, using local examples and formats, especially short videos.
At the technological level, platforms bear the greatest responsibility. Mandatory and visible labels on AI content, tools to verify where content came from, context-aware moderation in local languages including Bengali, and data access for researchers should now be minimum expectations. For newsrooms, the lesson is to break the cycle of social media to newsroom and back to social media. In a crisis, they should not broadcast viral videos without verification, should set up their own checks against photocard forgery, and should show readers how to recognise genuine cards. At the political-economic level, a legal framework is needed, but it must be designed so that it does not curb freedom of expression or itself become a tool to silence dissent. A code of conduct among political parties on transparency in using AI for election campaigns could also be a practical step.
For researchers and media watchdogs, the biggest responsibility is preserving evidence and analysing it consistently. That means building archives of slopaganda for every election and crisis, developing a shared classification framework, and building cross-border research partnerships among South Asian countries. The Global Voices analysis rightly notes that India, Pakistan, Nepal and Sri Lanka face similar risks, and that political operatives in the region are probably studying the tactics documented in Bangladesh closely. In other words, the other side is learning. Those resisting must learn just as fast.
Whether or not slopaganda becomes Oxford’s Word of the Year for 2026, the word captures a fundamental shift of our time. Propaganda is not new, and neither is fake news. What is new is the cost, speed, volume and degree of personalisation in producing it. What was a specialised weapon of state information warfare a decade ago is now in almost everyone’s hands. Its aim is not just to make people believe one lie, but to weaken society’s very ability to tell truth from falsehood. That is why slopaganda should be seen not merely as a tactic of information or psychological warfare, but as a central threat to cognitive security. Bangladesh’s 2026 election showed that this threat is not theoretical here; it is real and documented. If society, the media, platforms and the state do not prepare before the next election or crisis, the biggest casualty on this battlefield will be people’s collective judgement.