It is easy to read today’s headlines about artificial intelligence and misinformation and feel the ground shifting beneath your feet. We are told that chatbots will flood our feeds with falsehoods, that synthetic voices and faces will make seeing indistinguishable from believing, and that democracy itself is on the verge of being hacked by machines. But the growing body of research on this subject tells a more complicated and, in some ways, more hopeful story. Acerbi, Altay, and Mercier set the tone by asking whether we should be fighting misinformation at all, or instead fighting for information. Their point is not that lies are harmless; it is that the most effective way to reduce the power of falsehood is to make good information more available, more attractive, and more trustworthy. Altay and Acerbi push this further by showing that many people believe misinformation is a serious social threat because they assume other people are gullible, even when they do not see themselves falling for it. Mercier’s work on human reasoning adds a crucial insight: we are not passive sponges that absorb whatever we are told. We are socially alert animals who evaluate claims based on who is saying them, what motives we suspect, and what our communities believe. Simon and Camargo’s autopsy of the word “infodemic” reminds us that a single metaphor can distort our thinking, making the problem seem more uniform and apocalyptic than it actually was. And Orben’s research on technology panics shows how familiar this rhythm is: radio, television, video games, social media, and now artificial intelligence have all triggered waves of fear. The question is not whether misinformation exists, but how much power it has, who is most vulnerable, and which responses actually help. The experts surveyed in this literature disagree about definitions and priorities, which is a sign that the problem is real but not simple. What unites them is a shared belief that human beings are not isolated reasoning machines; they are embedded in relationships and institutions that shape what they trust and share.
So what does the evidence actually tell us about the scale of the problem? First, fake news in the narrow sense of fabricated articles is not nearly as common as the panic suggests. Allen and colleagues found that fabricated stories made up only a tiny fraction of the information Americans consumed, even during politically intense moments. Grinberg and colleagues studied fake news on Twitter during the 2016 United States election and found that exposure was heavily concentrated among a very small group of users. Guess’s work on online media diets adds another nuance: most Americans, even in politically divided times, consume a surprisingly mixed and moderate mix of news, not an endless echo chamber of extreme sources. This does not mean misinformation is harmless. A small number of highly engaged people can have disproportionate influence in a close election, and false stories about vaccines or emergencies can lead to real harm. But the data challenges the image of an all-powerful algorithm injecting lies into millions of passive brains. When people do share fake news, their motives are often more social than cognitive. Osmundsen and colleagues found that political polarization and partisan identity are among the strongest predictors of sharing false political stories on Twitter. Sharing is not necessarily an expression of belief; it is a way of showing loyalty, attacking opponents, and performing a political identity. Research on political microtargeting points in the same direction: personalized digital ads are less persuasive than many people assume. Tappin and colleagues tried to quantify the persuasive returns of microtargeting and found them to be modest at best, while Zarouali and colleagues showed that personality-tailored messages have subtle, conditional effects. All of this suggests that human judgment is not as fragile as we sometimes fear. People filter information through their identities, their emotions, and the people they trust. That is not a reason to be careless, but it is a reason to focus less on exotic techniques and more on the ordinary social and institutional conditions that make falsehoods plausible.
Nowhere is that clearer than in the study of conspiracy beliefs. It is tempting to dismiss conspiracy believers as ignorant or mentally unwell, but research increasingly shows that context matters as much as cognition. Alper found that conspiracy beliefs are more common in more corrupt countries, and Cordonier and Cafiero observed a similar pattern across a wide range of societies. When public officials lie, take bribes, or cover up their crimes, a belief that hidden forces are running the world is not irrational; it is an honest reading of experience. Ecker and colleagues provide a comprehensive review of the psychological drivers of misinformation belief, including cognitive biases, emotional states, social identities, and worldviews. But they also stress that the problem is not simply a lack of knowledge. The so-called “deficit model,” which says that ignorance causes misinformation and facts will cure it, is too simple. Mercier’s work explains why: people reason primarily in social contexts, and they often hold beliefs because those beliefs connect them to the groups they depend on. A person who rejects the mainstream scientific consensus on vaccines may be demonstrating loyalty to a political tribe more than making a scientific judgment. Hameleers adds another layer by warning that endlessly emphasizing the threat of misinformation can backfire. When people are told that everything is fake, they may become more cynical about all sources, including reliable ones. This is the liar’s dividend in reverse: if the world is nothing but lies, then no one can be held accountable. The takeaway is profound. We cannot fix misinformation just by issuing more corrections or fact-checks. We have to address corruption, strengthen institutions, support honest journalism, and rebuild trust between strangers. When institutions are perceived as fair, false claims have less fertile ground in which to grow. Misinformation is not just a psychological problem; it is a symptom of damaged democratic infrastructure.
Then there is generative AI, the newest source of both fear and fascination. Chatbots and image generators have brought the future of synthetic media into the present with breathtaking speed. Hanley and Durumeric have already detected machine-generated articles on both misinformation websites and mainstream news sites, showing that AI-produced content is not a distant risk but an emerging reality. Goldstein and colleagues tested whether large language models can write persuasive political propaganda and found that AI-generated text can be at least as convincing as human-written text, and at a scale and cost that no human team can match. This makes it dramatically easier for bad actors to flood public conversations with crafted messages. Deepfakes add another layer of danger. Habgood-Coote describes the possibility of an “epistemic apocalypse,” not because everyone will believe fake videos, but because in a world where any recording can be dismissed as synthetic, evidence becomes fragile and truth becomes unstable. Schiff and colleagues capture this with the phrase “liar’s dividend”: public figures who are caught saying or doing problematic things can claim that the audio or video is a deepfake. Christopher reported on a real-world version of this in India, when a politician dismissed leaked audio clips as AI-generated. The technology does not have to succeed as a forgery in order to do damage; the mere possibility is enough to create doubt. Yet researchers also caution against overhyping the threat. Groh and colleagues found that people are not highly accurate at detecting political deepfakes, but Paris and Donovan point out that “cheapfakes,” simple edits using consumer software, often work just as well and are far more common. Doss and colleagues show that AI-generated images can even contaminate scientific knowledge, while Weikmann and Lecheler argue that fact-checkers need to move beyond debunking individual posts and instead build broader verification networks. Kapoor and Narayanan offer practical advice for the coming flood: rather than trying to detect manipulated content after it spreads, platforms should build provenance and authentication into the way content is created and shared. Again, the message is not that AI is harmless, but that its effects depend on the social and institutional environment. If trust is low and institutions are weak, generative AI magnifies existing fractures. If we prepare carefully, the damage can be contained.
What does all of this mean for journalism and policy? One of the most encouraging findings in the research is that people are drawn to quality news in moments of crisis. Altay, Nielsen, and Fletcher found that during the early phase of the coronavirus pandemic, many people turned to trustworthy, professional news outlets rather than fleeing to tabloid rumors. The image of a global infodemic was not false, but it was incomplete; alongside the misinformation, there was a genuine hunger for reliable information. This is a reminder that the fight against misinformation is not simply a fight against lies. It is a fight for a healthy information ecosystem. Bell has argued that whatever ChatGPT and other tools do to journalism, the role of reporters and editors becomes more important than ever, because audiences need someone to make sense of a world filled with both human and machine-generated claims. Becker and colleagues studied how newsrooms across the globe are developing artificial intelligence policies and found that these policies are fragmented and inconsistent, which points to the need for more public conversation about the values that should guide automated journalism. Arguedas and Simon describe the process as “automating democracy,” arguing that generative AI will reshape not just news production but the relationship between journalists, sources, and citizens. Gregory puts the issue in human rights terms, calling on journalists and human rights defenders to “fortify the truth.” On the regulatory side, the Council of Europe has warned that algorithmic processes can manipulate citizens at scale, and Broniatowski and colleagues found that Facebook’s vaccine misinformation policies had a real, if imperfect, effect, showing that platform governance is not futile. Shah and Bender imagine a different future for information access systems, one in which tools are designed to support a healthy web rather than to maximize engagement. The work of fact-checkers remains valuable, but it cannot be the only defense. What is needed is a toolkit: better verification standards, content authenticity, media literacy, transparent algorithms, public service media, and accountability for technology companies. The goal is not to create a sterile world with no falsehoods. The goal is to create a world where truth is easy to find and lies are hard to hide.
At the end of this long and sometimes dizzying literature review, the feeling that emerges is neither doom nor complacency. It is humility. The experts who study misinformation hold different views about what it is, why it spreads, and how to stop it, and that disagreement is itself a clue that the problem is multidimensional. There is no platform policy that will solve it, no AI tool that will magically identify every falsehood, and no school curriculum that will make citizens immune to manipulation. What we can do is pay attention to the deeper conditions that allow misinformation to thrive. That means reducing corruption, supporting local journalism, investing in education, funding scientific institutions, and designing digital spaces that are less addicted to outrage. Coppock’s work on persuasion shows that attitudes change slowly and in parallel, through repeated and subtle encounters, not through single dramatic interventions. Jungherr and Schroeder, in their work on digital transformations of the public arena, remind us that the current crisis is not only about technology but about the changing structure of the public sphere itself. If we focus narrowly on fake news, we miss the larger story: the way platforms reshape attention, the decline of local media, the crumbling of democratic norms, and the rise of inequality. Technology panics have happened before, as Orben shows, and they have always contained elements of both truth and exaggeration. But so too have societies learned to adapt. We still teach children to read even though books can contain falsehoods. We still fund libraries even though some writers are dishonest. We still rely on journalism even though reporters make mistakes. The most human response to misinformation is not to assume that everyone around us is stupid. It is to build institutions that reward honesty, relations that sustain trust, and habits that make us thoughtful sharers of information. The research collected here does not offer an easy fix. It offers, instead, an invitation to grow up together: to face the new world of synthetic media and algorithmic persuasion not with panic, but with the confidence that democracy, if it is worth having, is also worth fortifying.