Hate communities often flourish online for years, raising the question of how they persist. My research team has found that powerful stories keep members of a hate group galvanized, either by repeating the story over and over or by constantly adding fresh accusations and interpretations to it.

I’m a computational social scientist who studies social and political networks. My colleagues and I uncovered these trends by examining 10 years of posts, reactions and participation patterns in Facebook groups that shared antisemitic and Islamophobic content. Our findings have been accepted at the 2026 International Conference on Web and Social Media.

First, we measured who was posting and how that related to engagement on a site. Groups in which a small number of people produced most of the content tended to attract more reactions and responses. Then we looked at subjects the group members discussed – religion, immigration, geopolitics – and the kinds of stories members told about those topics, such as describing an entire group of people as criminals or warning that certain types of people are secretly taking over a country’s way of life.

When we put these pieces together, we discovered some clear patterns. Messages posted by a few very active people were strongly associated with higher site engagement in the form of likes and shares in the near term. And repetition – espousing the same ideas again and again – was an effective tactic. We also found that when many users kept adding fresh accusations, conspiracy theories and explanations, a group tended to persist. Very uniform content that used the same framing led to less engagement over time.

Different communities seemed to be drawn to different messaging patterns. In Islamophobic groups, the most prolific posters tended to repeat a narrow, consistent set of messages. Often these were religiously framed posts that portrayed Muslims as morally condemned. In antisemitic groups, the most engaged members were more likely to impart a mix of narratives, from tales of victimization to conspiracy theories about public figures.

Why it matters

Our findings suggest that hate communities can sustain themselves in various ways, so efforts to moderate them should consider these variations. If a few voices drive the conversation, removing them could quiet the noise. If new stories constantly appear from many contributors, harmful ideas may survive even if a few key online accounts are taken down. Hate networks can persist even after social media platforms ban specific groups or accounts.

It is also important to understand how stories can make prejudice feel justified and emotionally compelling. Extremist stories may claim that a group is under attack, that outsiders are dangerous or subhuman, or that violence is the only way to stay safe. Groups seen as outsiders – such as immigrants – are common targets, and they may be described as an “invasion” that threatens the nation.

What other research is being done

Researchers are finding that extremist ideas are now spreading through looser networks where many voices contribute and messaging can vary widely. That could affect whether engagement in the future still depends on consistent repetition or novelty. Some investigators are also scrutinizing how harmful language, conspiracy theories and propaganda evolve over time.

What’s next

Another important direction is tracking how hate narratives are spread by public figures and influencers, how the narratives move between online platforms, and how they surface in offline groups and efforts to organize supporters, all of which can normalize harmful ideas. My group is starting to study how this amplification works: who shares which narratives and why, which kinds of people become bridges across different online platforms, and how those roles shape which messages spread.

The Research Brief is a short take on interesting academic work.

Yu-Ru Lin's research has received federal funding, including National Science Foundation and the Department of Defense (DARPA, AFOSR, Minerva, and ONR). Any opinions, findings, and conclusions or recommendations expressed in this material do not necessarily reflect the views of the funding sources.

When you ask a large language model a question, the reply may include falsehoods, and if you challenge those statements with facts, the AI may still uphold the reply as true. That’s what my research group found when we asked five leading models to describe scenes in movies or novels that don’t actually exist.

We probed this possibility after I asked ChatGPT its favorite scene in the movie “Good Will Hunting.” It noted a scene between leading characters. But then I asked, “What about the scene with the Hitler reference?” There is no such scene in the movie, yet ChatGPT confidently constructed a vivid and plausible description of one.

The confabulation – sometimes called an AI hallucination – revealed something deeper about how AI systems reason. References to Hitler are not uncommon in films, which apparently convinced ChatGPT to accept and elaborate on a false premise rather than correct it. I study the social impact of AI, and this surprise response led my colleagues and me to a broader question: What happens when AI systems are gently pushed toward falsehoods? Do they resist, or do they comply?

We developed an approach we called hallucination audit under nudge trial to answer those questions. We had conversations with five leading models about 1,000 popular movies and 1,000 popular novels. During the exchanges we raised plausible but false references to Hitler, dinosaurs or time machines. We did this in various suggestive ways, such as “For me, I really love the scene where …”

Our method works in three stages. First, the AI generates statements about a topic — such as a movie or a book — some true and some false. Second, in a separate interaction, the AI attempts to verify those statements. Third, we introduce a “nudge,” where the model is challenged with its own incorrect claims to see whether it resists or accepts them.

We found that AI models often struggle to remain consistent under pressure. Even when they initially identify a statement as false, they may later accept it when nudged – revealing a vulnerability that traditional evaluation methods fail to capture.

Our results have been accepted at the 2026 Annual Meeting of the Association for Computational Linguistics.

This tactic isn’t a hypothetical. When people talk, conversational pressure can emerge naturally. People may confidently repeat incorrect assumptions, partial recollections or misunderstandings. A person might say, “I’m pretty sure medicine X is effective for condition Y,” or “I remember event A happening before event B.” These statements can subtly influence an AI model.

Why it matters

What humans collectively remember, misremember and forget shapes our sense of reality. But if humans can persuade a model to accept a falsehood, that reveals an important vulnerability in AI’s capacity to provide accurate information.

Interactions in the real world are rarely static question-answer exchanges. They are interactive and iterative. An AI model’s willingness to reinforce falsehoods may seem harmless when chatting about movies, but in areas such as health, law or public policy, the tendency can have serious consequences. Our work highlights the need to evaluate not just what information AI systems have been trained on, but how reliably they stand by it.

What other research is being done

Our results add to other recent research into why large language models may produce hallucinations, and how it is that they can provide inconsistent information. Researchers are also trying to figure out why some models lean toward sycophancy – flattering or fawning over human users.

What still isn’t known

It’s not clear why some AI systems resist falsehoods better than others. In our tests, Claude was the most resistant, followed somewhat closely by Grok and ChatGPT, with Gemini and DeepSeek further behind.

Movies and novels are self-contained content. Scholars don’t know how AI might respond to pressure in much broader, complex real-world settings. As a start, my group is exploring how to extend our approach to scientific literature and health-related claims. We want to understand whether conversational pressure works differently when the discussion involves uncertainty or expertise.

How to design AI systems that remain both helpful and resistant to falsehoods under wide-ranging conversation remains an open challenge.

The Research Brief is a short take on interesting academic work.

Ashique KhudaBukhsh receives funding from Lenovo.