The era of AI-generated news and its influence: where do we go from here?
Artificial intelligence has already changed a lot about how we consume information, but a recent study brought a finding that is really hard to ignore.
Communication psychology researchers at University College London (UCL) discovered something that cuts right to the chase: people who watched deepfake videos continued to be influenced by the content even after being told that everything was fake.
Let that sink in for a second.
Knowing that something is a lie was not enough to erase the effect that content had already caused in people’s minds.
It is an uncomfortable paradox, right?
We grow up hearing that the antidote to a lie is the truth, but science is showing that the human brain does not quite work that way when the subject is fake news and hyper-realistic visual content created by AI.
And what makes all of this even more urgent is the fact that the tools to create deepfakes are becoming more accessible, faster, and more convincing with each passing month.
In this article, we will break down what this study revealed, why our brains fall into this trap, and what is being done in practice to deal with this scenario. 🧠
Receive the best innovation content in your email.
All the news, tips, trends, and resources you're looking for, delivered to your inbox.
By subscribing to the newsletter, you agree to receive communications from Método Viral. We are committed to always protecting and respecting your privacy.
What the UCL study revealed about deepfakes and influence
The research conducted by University College London was not a simple experiment. The researchers worked with groups of participants who were exposed to deepfake videos of public figures saying things they had never actually said. After the exposure, some of the participants were informed that the content was completely fabricated by artificial intelligence. What happened next was exactly what the scientists wanted to measure, and the results were surprising even to the researchers themselves who were involved in the project.
Even after receiving the information that the video was fake, participants continued to demonstrate beliefs and attitudes aligned with the content they had watched. In other words, the influence of the deepfake persisted even with an explicit correction being made right afterward. This phenomenon had already been studied in the field of communication psychology in more traditional contexts, such as rumors and disinformation texts, but the intensity observed with AI-generated videos was considerably greater than in previous experiments using other formats of false content.
What the data suggests is that the hyper-realistic visual format carries a different cognitive weight. When the brain processes a moving image of someone who looks real, speaking naturally, with convincing facial expressions and vocal intonation, it activates emotional processing mechanisms before any critical analysis even takes place. It is as if the message sneaks in through a back door, going straight into memory and the belief system, before rational judgment has a chance to intervene. That is the core knot of the problem that the study exposed with concrete data.
Why the brain cannot simply switch off the influence
Understanding why this happens requires a quick dive into communication psychology and how the human brain responds to visual stimuli. Humans evolved to trust what they see. For most of our species’ history, seeing something happen right in front of your eyes was sufficient proof that it was real. The problem is that generative artificial intelligence has learned to exploit exactly this feature of our cognitive system, producing visual content that deceives the senses with a level of precision that simply did not exist before this technology reached its current stage of development.
There is a well-studied concept in psychology called the continued influence effect, which describes exactly what the UCL study observed. When a piece of information is absorbed and processed, it leaves a trace in a person’s cognitive structure, even if that information is debunked shortly afterward. The correction has to compete with an already-formed memory, and in that competition, the original memory frequently wins out, especially when the original content was emotionally charged or visually striking. A well-produced deepfake video checks all of those boxes at once, which explains the persistence of influence even after correction.
On top of that, there is a factor of implicit trust that videos carry culturally. For decades, society has treated images and video as evidence. Phrases like a picture is worth a thousand words or seeing is believing are deeply rooted in the collective imagination. Even when people intellectually know that deepfakes exist and that AI can fabricate faces and voices, the brain’s emotional system still responds to the video as if it were real in the moment it is being watched. That gap between what a person knows rationally and what they feel instinctively is the fertile ground where fake news in deepfake video format plants its deepest roots.
The role of emotional memory in this process
Another aspect that deserves attention is how emotional memory works differently from factual memory. When someone watches a deepfake video with a strong emotional charge, whether it triggers outrage, fear, surprise, or anger, the brain stores not only the information but also the emotion attached to it. Even if the factual information is corrected afterward, the emotion remains intact in memory. That emotional residue continues to subtly influence future judgments and decisions, often without the person even realizing they are being guided by a reaction to something they themselves recognize as false.
Research in the field of cognitive neuroscience has already demonstrated that the limbic system, which handles emotional processing, operates at a significantly faster speed than the prefrontal cortex, which manages critical thinking and rational analysis. In practical terms, this means the emotion arrives first and settles in before reason has time to question what was just seen. In the context of deepfakes, this speed difference between emotion and reason is exactly what makes these videos so dangerously effective as tools of disinformation.
The current landscape of AI-generated fake news
The UCL study arrived at a moment when the volume of synthetic content circulating on the internet has reached proportions that would have been unthinkable just five years ago. Video generation tools powered by artificial intelligence that once required powerful servers and specialized teams are now available in smartphone apps, free or low-cost, accessible to anyone with an internet connection. This means the ability to produce convincing deepfakes is no longer restricted to tech labs or groups with significant resources — it is in the palm of the hand of any user who wants to use it, regardless of their intentions.
The impact on the information ecosystem is profound. Social media platforms have been scrambling to develop automated detection mechanisms for synthetic content, but the race is asymmetric. Generation tools evolve continuously, often faster than detection tools can keep up. Digital security researchers report that detection models trained to identify visual artifacts typical of deepfakes become outdated in a matter of months, because newer generations of AI models correct the very patterns that detectors learned to recognize. It is an attack-and-defense cycle that feeds on itself with no definitive solution in sight in the short term.
From the perspective of communication psychology, the problem is not just technical. The speed at which fake news spreads on social media means that, in many cases, the correction arrives too late to reach the same audience that consumed the original content. Earlier studies in the field had already shown that false news spreads significantly faster than true news in the digital environment, and when the false content comes in the form of a high-production deepfake video, engagement tends to be even higher, amplifying the reach before any correction is published. That unfavorable timing is one of the biggest challenges for anyone working in the fight against disinformation today. 📱
The cascade effect on social media
One point that deserves special attention is the cascade effect that a single deepfake video can trigger on platforms like Instagram, TikTok, X, and YouTube. When a piece of false content goes viral, it does not circulate only in its original format. Users create clips, video commentaries, reactions, and even memes based on that material, multiplying exposure to the disinformation in derivative formats that completely escape automated moderation systems. Even if the original video is removed, these secondary versions keep circulating, extending the lifespan of the lie in an almost organic way.
Beyond that, platform recommendation algorithms tend to prioritize content with high emotional engagement, which is exactly the type of response that deepfakes provoke. This creates a situation where the very way social networks operate amplifies the distribution of potentially false content, even without deliberate intent from those running the platform. It is a structural dynamic that goes beyond individual content moderation and touches on the very architecture of information distribution systems that billions of people use every day.
What is being done to tackle this problem
The response to the challenges created by deepfakes and AI-generated disinformation is being built on multiple fronts simultaneously, because no single approach is enough to handle the complexity of the problem. On the technical side, major tech companies like Google, Meta, and Microsoft have been investing in digital watermarking systems, which is essentially an invisible watermark embedded in AI-generated videos, allowing verification tools to identify the synthetic origin of the content even after it has been shared and edited. Initiatives like the Content Authenticity Initiative (CAI) and the C2PA project are working to create open standards for digital content authenticity certification, something like an origin passport for images and videos.
On the media literacy front, researchers and fact-checking organizations are developing digital literacy programs focused specifically on how to identify signs of manipulation in videos. The goal is not to turn every internet user into a forensic expert, but to build more critical consumption habits, such as:
- Verifying the source before sharing any video content
- Seeking the original context of a video before believing what is being presented
- Using reverse image search tools to check the authenticity of visual materials
- Being skeptical of videos with very intense emotional charge that appear without clear context
- Consulting fact-checking agencies before forming a definitive opinion on sensitive content
Some studies in the field of communication psychology have shown that educational interventions carried out before exposure to false content, known as pre-bunking approaches, are more effective than debunking done after the disinformation has already been consumed. This finding speaks directly to the results of the UCL study on the persistence of deepfake influence, reinforcing the importance of preparing audiences before they are exposed to the problem, rather than trying to fix the damage afterward.
Regulatory progress around the world
On the regulatory side, governments around the world are making moves to create legal frameworks that hold platforms and creators of synthetic content accountable when it is used with the intent to deceive. The European Union, with its AI Act, requires AI systems that generate synthetic content to include clear identification that the material was artificially produced. This legislation represents one of the most comprehensive efforts so far to establish clear rules on transparency in the use of generative artificial intelligence in public communication contexts.
In Brazil, the regulatory debate around fake news and the use of AI in electoral contexts gained momentum especially after recent episodes involving deepfakes in political campaigns, leading the Superior Electoral Court to publish specific resolutions on the use of artificial intelligence in campaign materials. It is an important step, even though gaps in oversight and enforcement remain a real challenge for the institutions involved. ⚖️
The limits of technological solutions
It is important to recognize that no detection technology, no matter how advanced, will solve this problem on its own. Digital watermarking, for example, depends on widespread adoption by developers of generative AI tools, something that is still far from universal. Open-source tools, models shared on forums, and decentralized platforms frequently operate outside the reach of these standardization initiatives, creating significant gaps in the traceability system for synthetic content.
Similarly, automated detection based on machine learning faces what is known as the cat-and-mouse dilemma. Every time detectors improve, generators also evolve, often using the detectors themselves as adversarial training tools to produce deepfakes that are even harder to identify. This cycle suggests that the solution to AI-generated disinformation will not be purely technological, but rather a combination of technology, education, regulation, and fundamentally, a cultural shift in the way people relate to digital content.
The human factor at the heart of the issue
What the UCL study leaves us with as its most important lesson may not be technical or political. It is human. It reminds us that the influence of a piece of content does not disappear just because someone tells us it is fake. This places an enormous responsibility on the moment of consumption, on the environment in which people receive information, and on the habits that shape each person’s relationship with what appears on their screen.
In a world where artificial intelligence can fabricate visually flawless realities, awareness of how our own brains process these images may be the most valuable tool any person can develop. It is not about distrusting everything or living in a permanent state of digital paranoia, but about understanding that the human brain has vulnerabilities that can be exploited by synthetic content, and that recognizing those vulnerabilities is the first step toward not being ruled by them.
The question that lingers, and that the UCL study poses quite directly, is this: if knowing something is fake is not enough to cancel its effect, then what is? The answer is still being built by researchers, educators, and technology developers around the world. But one thing has already become clear: the approach needs to change. Debunking after the fact is not enough. Preparation needs to come first. And that means rethinking how we teach critical thinking, how we design information platforms, and how each person chooses to engage with the content they consume every day. 🧩
