“A Single Image No Longer Proves Anything Today”

What was the most surprising finding in the latest survey for you?

Alexander Godulla: What surprised me most was the speed of development. In 2022, 77 percent of people in Germany said they had never encountered a deepfake in the previous year. In 2025, only 27 percent said the same. Almost three quarters now report having come into contact with deepfakes. The fact that a phenomenon can develop from a marginal occurrence into an everyday experience within three years is remarkable even in the fast-moving digital world. At the same time, one thing is becoming clear: knowledge about deepfakes is growing much more slowly than the frequency with which people encounter them.

You speak of the “end of the authentic image.” Do we fundamentally need to relearn what counts as evidence?

Godulla: Yes, that is exactly what I mean. For more than 150 years, photography was regarded as a particularly credible witness, based on the idea that what can be seen in a photograph must have happened that way. That certainty has been irretrievably lost. A single image no longer proves anything today, because it may have been generated entirely synthetically. We need to assess evidentiary value differently: by looking at the source, the context, the provenance of a recording, and the question of who is distributing it and with what interests. As a society, we are therefore at the beginning of a collective learning process.

Political deepfakes are perceived more frequently, even though pornographic ones are much more widespread. What does this gap in perception mean?

Godulla: Political deepfakes encounter us openly in feeds and in news coverage, whereas pornographic ones circulate in spaces that many people never enter. In our survey, 74 percent of those who had encountered deepfakes mentioned political content, while only 19 percent mentioned pornographic content. What this means for the debate is that risks are assessed according to their visibility rather than their prevalence. Only 29 percent are concerned about being depicted themselves in a pornographic deepfake, whereas 62 percent see this risk with other forms of deepfakes. I consider this perceived distance from one’s own risk to be the real problem.

Isn’t the real danger rather that genuine recordings can increasingly be dismissed as fake?

Godulla: I actually consider that to be the underestimated side of the problem. If, in principle, any image can be fake, then conversely any genuine image can be dismissed as a forgery. The accusation of manipulation thus becomes a manipulation strategy in its own right. The insidious thing about this is that this “liar’s dividend” works without any technical forgery at all; it only requires widespread doubt. That is precisely why it is so important that we strengthen methods of verification.

Is there such a thing as “deepfake fatigue” – the point at which people simply stop asking themselves whether something is real?

Godulla: This danger is real, and we are seeing the first signs of it. People who perceive themselves as increasingly coming into contact with deepfakes are more likely to develop media cynicism – a blanket attitude that they are being lied to from all sides anyway. That is different from healthy skepticism. Skepticism asks questions and checks; cynicism dismisses things and stops distinguishing between them. That very indifference would be the real triumph of disinformation.

Can I learn to recognize a deepfake with the naked eye?

Godulla: To be honest, that is the wrong approach. Only around a quarter of respondents feel confident that they could recognize a deepfake themselves or distinguish a genuine video from a fake one. The technology is improving faster than our eyes can be trained. My advice, therefore, is not to rely on a trained eye, but on a verified source. The question is not what an image looks like, but where it came from.

We mainly talk about the dangers. Are there also useful applications for synthetic media?

Godulla: Absolutely, and the public sees this more differently than one might expect. In our survey, the use of synthetic media receives approval particularly in situations where no one is being deceived: in online retail, for example, when virtually trying on clothes; in education, where historical events can be visualized in vivid ways; and in film. For people, the decisive factor is less the technology itself than the question of integrity: its use is considered legitimate when it is disclosed that AI was involved and when the people depicted have given their consent. Synthetic media are therefore not inherently harmful. What matters is whether they are used to illustrate and enhance reality or to deceive.

You call for media literacy to start at an early age. What can schools or parents concretely do today?

Godulla: You do not need a reform for this; above all, you need conversations about images. Parents can look at content together with their children and practice asking simple questions: Who published this? Why now? What do other sources say about it? Schools can bring current cases into the classroom. What matters to me is the attitude behind this: The aim is not to teach children to distrust everything, but to develop healthy skepticism combined with routines for checking information.

The EU AI Act provides for labeling requirements. How effective can these be?

Godulla: Labeling requirements are not a cure-all, but neither are they ineffective. Of course, malicious actors outside the EU will not comply with them. Nevertheless, the AI Act establishes a standard for what constitutes legitimate use and requires major platforms to implement technical safeguards. Our data also show that more than 90 percent of people in Germany want deepfakes to be labeled. The key will be to combine labeling with technical provenance mechanisms and media literacy.

If you conduct the same survey again in five years, what would you be worried about – and where would you be optimistic?

Godulla: I am concerned that the liar’s dividend will increase – in other words, that genuine images will routinely be dismissed as fake and that parts of the public will retreat into separate realities. What gives me cause for optimism is that between 2022 and 2025, people’s confidence in their ability to deal with deepfakes has increased slightly. Regulation and technical provenance mechanisms are continuing to develop. And historically, every society has adapted to new media technologies. So I do not expect the end of truth, but rather a difficult yet manageable learning process.

The image used for this blog post was created using the AI-powered image generation technology Midjourney. It is intended for illustrative purposes and may contain stylized or abstract representations. All rights to the generated images belong to Futurium or the respective creators, where applicable. The following prompt was used: A single fictional person standing in an ordinary realistic environment, photographed straight on as a documentary portrait. The image splits vertically down the center into two perfectly aligned realities, same person, same pose, same clothing, same location. LEFT HALF: Authentic documentary photograph - natural human proportions, realistic skin, natural hair, believable hands, correct anatomy, physically correct shadows and light, genuine depth, subtle film grain, natural imperfections. RIGHT HALF: Extremely exaggerated synthetic reality. The person appears almost impossibly real but the physical world is wrong.