Picture this: you are scrolling through social media and a video appears showing a political leader saying something shocking. The voice sounds right, the face moves naturally, the setting looks authentic. Yet none of it is real. The entire clip was generated by artificial intelligence, and it has been designed to deceive. This is no longer the plot of a science fiction film; it is a growing reality in today’s information environment. Deepfakes can be used to smear reputations, manipulate public opinion, or simply make people doubt everything they see. Faced with such content, how can ordinary citizens tell truth from falsehood? Why are some people more easily fooled than others? And is it possible to train people to spot manipulation before they are misled? These are the questions at the heart of a new European research project called COMPASS, which brings together experts from eight countries to study disinformation from every angle. At the University of Duisburg-Essen (UDE), a team led by Junior Professor German Neubaum is investigating the psychological side of this urgent challenge. The project has received around €3.5 million through the European Union’s Horizon Europe programme, with approximately €443,000 going to the psychological research conducted in Duisburg. The name COMPASS stands for “COuntering Manipulation while Preserving Authentic Speech and Society,” and that title captures a delicate balance: how can democracies defend themselves against manipulation without destroying the open, free and privacy-protected public sphere that they are trying to protect?
Artificial intelligence has reached a level of sophistication that makes it possible to generate remarkably convincing images, voices and videos. A deepfake may still contain minor inconsistencies, such as an unnatural shadow or an awkward blink, but these flaws are quickly disappearing. Before long, AI-generated content may be virtually indistinguishable from authentic recordings. At the same time, the temptation to use this technology for malicious purposes is growing. Deepfakes can be aimed at politicians, journalists, celebrities or ordinary citizens. They can be used to create false evidence, to discredit opponents, or simply to flood the public sphere with so much confusion that people stop trusting anything at all. This is why COMPASS takes a broader view. It does not focus merely on identifying individual fake videos or fact-checking suspicious posts. Instead, it examines the entire system of information manipulation: how disinformation is crafted, how it spreads across platforms, why it finds receptive audiences, and what kind of countermeasures are possible. Underlying all of this is a fundamental democratic question: how can societies protect themselves from manipulation without restricting freedom of expression or invading the privacy of their citizens? To answer that question, the project brings together eight partners from eight European countries, each contributing a different perspective. Psychologists study how people perceive and evaluate digital content, legal scholars examine the limits of regulation, and social scientists investigate the broader societal conditions that allow disinformation to flourish. No single discipline can solve this problem alone, because disinformation is not simply a technical failure or a legal loophole; it is a human phenomenon.
At the University of Duisburg-Essen, Professor Neubaum and his team are responsible for the psychological research within COMPASS. Their work package is titled “Co-creating Democratic Resilience: from Evidence to Empowerment,” which signals a commitment to moving beyond diagnosis and toward practical solutions. The core of their research involves a series of experiments in which participants are exposed to different types of deepfakes. After viewing the content, participants are asked whether they believe it to be genuine, and the researchers carefully measure which factors best predict their judgements. Several kinds of influences are considered. Personal characteristics such as age, educational background and digital experience may play a role. Cognitive abilities and critical thinking skills are likely to matter as well. But the researchers are equally interested in less obvious factors, such as political attitudes, general trust in institutions, emotional states and even the participant’s mood at the moment of exposure. Neubaum is careful to avoid simple explanations. “Why some people are more likely than others to fall for AI-generated misinformation probably cannot be explained by any single characteristic,” he says. “We expect a combination of different factors, and our experiments will explore how these interact.” That interaction is crucial. A politically engaged person with a university degree may still be fooled by a deepfake that confirms their existing beliefs, while someone with little digital literacy might be suspicious because they have learned to distrust online content in general. The aim is not to create a simple profile of the “typical” deepfake victim, but to understand the complex psychological processes that make people vulnerable, and to identify the moments where intervention might be most effective.
Susceptibility, however, is not the same as sharing. A person may watch a deepfake and suspect that it is not fully authentic, yet still decide to spread it to others. This is one of the most subtle and important aspects of the UDE team’s research. The researchers are investigating how people’s beliefs and motivations shape their responses to questionable content. Consider a voter who strongly supports a particular political party. They come across a video that appears to show a rival candidate making a racist or offensive remark. Even if the voter is not entirely convinced that the video is real, they may feel that the message is too important to keep quiet. It fits neatly into their worldview. It reinforces what they already believe about the other side. Sharing it feels like a way of participating in democracy, or at least like striking a blow for their own cause. This is exactly the kind of situation where deepfakes become most dangerous. False content spreads not only because people are naive, but because they are motivated by social identity, emotion and a desire for their views to win. The UDE team wants to understand this tension between truthfulness and perceived importance. If a person considers a message to be socially important, are they more likely to share it even when they have doubts about its authenticity? Almost certainly. But how strong is that tendency, and can it be weakened? What role do emotions such as anger, fear or hope play in the decision to share? These questions are essential because they explain why disinformation campaigns are so effective at polarising societies. They also show why simple technical solutions, such as adding warning labels to videos, will never be enough. Human beings are not passive receivers of information. They actively interpret, evaluate and spread content, often for reasons that have little to do with factual accuracy.
The next part of the UDE research agenda focuses on prevention. Once the researchers understand what makes people vulnerable, they can begin testing ways to protect them before they encounter harmful content. This is where inoculation and prebunking approaches come into play. The idea is not to teach people about every specific fake video, because new ones appear all the time. Instead, participants are introduced to the common strategies behind deepfakes and the suspicious cues that may reveal manipulation. They learn how AI-generated faces can struggle with certain angles or lighting, how audio can be subtly out of sync, and how the origin and publication history of a video can provide clues about its reliability. The researchers then assess whether this prior knowledge helps participants identify new forms of manipulation that they have never seen before. This approach is often compared to a psychological vaccine. A vaccine introduces a weakened form of a virus so that the immune system can learn to fight it. In a similar way, exposing people to a small, controlled dose of manipulative strategies may train their minds to recognise similar patterns in the future. But the researchers are careful not to promise more than they can deliver. “Such approaches are often compared to a psychological vaccine. But they cannot make people completely immune to disinformation,” explains Dr Jana Dreston, a member of the research group “Psychological Processes of Education in Social Media” at UDE. This is an important caveat. No single intervention is likely to make a person permanently resistant to all forms of manipulation. What the team hopes to discover is which cues actually help people in real-world situations, whether the protective effects persist when people encounter entirely new types of deepfakes, and whether the benefits last over longer periods of time. If these questions can be answered with evidence, it will become possible to design educational programmes and public awareness campaigns that are genuinely useful, rather than simply well-intentioned.
The COMPASS project is ambitious in both scope and duration. It will continue until August 2029, allowing the research teams to conduct long-term studies and develop recommendations that are grounded in solid empirical evidence. The interdisciplinary consortium is coordinated by Tallinn University in Estonia, and it brings together partners from eight European countries. The funding for the UDE contribution, approximately €443,000, is a relatively small part of the overall €3.5 million budget, but the psychological research in Duisburg is central to the project’s mission. Psychology is, after all, the discipline that connects the technical creation of deepfakes with the human consequences of their spread. Without a deep understanding of how people perceive, judge and share AI-generated content, any attempt to regulate or counter disinformation would be built on guesswork. The ultimate goal is not to create a society in which nobody is ever deceived. That would be an unrealistic and perhaps even undesirable ambition. Instead, the hope is to make manipulation more difficult, to make citizens more aware of their own psychological vulnerabilities, and to make public debate more resilient in the face of uncertainty. In an age when a plausible-looking video can be manufactured in minutes, the ability to pause, question and verify may be one of the most important democratic skills that a person can possess. The research being conducted in Duisburg is a step toward turning that skill from an individual instinct into a collective capacity. If the project succeeds, Europe’s democracies may be better equipped to protect themselves from disinformation without sacrificing the openness, freedom of expression and respect for privacy that make them democratic in the first place.