Deepfakes can fool voters, but new research points to a deeper danger: genuine political footage is increasingly being dismissed as fake. As Victoria heads towards its November election, scepticism itself may become a vulnerability.
The Victorian election is still months away, yet voters are already being shown events which never happened. A machete attack at a petrol station, a woman giving birth on the roadside, floodwater pouring down the steps of parliament: all ‘deepfakes’ generated by artificial intelligence (AI), and all reportedly appearing as paid advertising ahead of voters going to the polls.
Previously described as a potential ‘sleeper issue‘ for Australian politics, deepfake political content is rapidly becoming a very real electoral challenge.
We have been studying political deepfakes – and how people respond to them – in the lead-up to the Victorian state election. Our preliminary results provide important insights into people’s ability to detect deepfakes and will help voters navigate the increasingly fraught information ecosystem as electoral campaigning ramps up.
Deepfakes once required specialist technical expertise and software to convincingly produce. Today, they can be produced by anyone using commercial and open-source AI tools. And they’re getting harder and harder to spot.
We conducted a meta-analysis that brought together findings from previous published studies from 2018 to 2026 to understand political deepfakes more and what this could mean for upcoming elections.
Across 19 studies and more than 24,000 people, the results (which are yet to be published) identified deepfakes can appear just as persuasive as other political content, including authentic footage and conventional misinformation.
We also found that deepfakes are associated with lower trust in news and democratic institutions, potentially making people question not only what is fake, but what is real – ultimately damaging trust in democracy itself.
At a broader level, this uncertainty can contribute to the ‘liar’s dividend‘. This is where individuals dismiss genuine content as inauthentic.
Rejection of the real
In an experiment we did with 411 Australians, participants viewed an authentic political video, a low-quality deepfake, or a high-quality deepfake. Our preliminary findings are cause for concern.
People who consumed more news were better at detecting low-quality deepfakes, but not sophisticated ones. More strikingly, as news consumption increased, people became less accurate at identifying authentic content.
We found that roughly 75 per cent of our sample could correctly identify deepfakes, regardless of whether they were high or low quality. However, this still leaves around one in four people who are unable to correctly identify a deepfake or are unsure about its authenticity.
Of even greater concern, almost 29 per cent incorrectly labelled the real video as fake, while a further 18 per cent were unsure.
Put another way, almost half of participants could not confidently and correctly identify genuine political content as real.
The danger, then, is not only that voters may believe content that is fake, but that they may also be becoming too ready to disbelieve content that is real.
Who is better at spotting a deepfake? It’s complicated
Some people are likely to be more susceptible to deepfakes than others.
Previous research suggests political deepfake detection declines with age, which may be particularly relevant in Australia, given the substantial proportion of older voters.
Political orientation may matter too.
Research from the United States suggests left-leaning individuals can be more accurate at detecting political deepfakes.
Our preliminary experimental findings add an important nuance: differences emerged only when the deepfake depicted a political actor aligned with the viewer’s own political orientation.
For example, if you are a Labor supporter you’re more likely to be able to tell a video of the current prime minister is a deepfake. This suggests susceptibility may depend not simply on a person’s politics, but on who is being depicted and their political alignment with the viewer.
Feeling confident doesn’t mean you’re right
Research provides a mixed picture regarding the role of confidence in detecting deepfakes.
Some studies suggest greater confidence can help people identify deepfakes, while others find confidence has little relationship with correct identification.
Our latest findings add another layer.
Among moderate and higher news consumers, confidence became increasingly disconnected from accuracy. These participants may have felt well equipped to judge political content, but that confidence did not necessarily translate into getting it right.
Therefore, feeling confident that something is AI-generated is not evidence that it actually is. Confidence may even become a vulnerability.
This matters beyond whether one video is correctly classified. If voters become increasingly suspicious of authentic evidence simply because it could have been generated by AI, deepfakes can potentially undermine trust even in their absence.
How can you spot a political deepfake?
There is no foolproof way to spot a deepfake. As AI improves, voters should rely less on whether something ‘looks fake’ and more on verifying its source and supporting evidence.
First, don’t rely on appearance. Visual or audio flaws can reveal poorer deepfakes, but these cues are becoming less reliable.
Second, trace the source. Find where the content originated and consider whether the source is credible.
Third, check other evidence. Look for independent reporting or official sources confirming the content.
Fourth, question your confidence. Feeling certain something is fake does not mean you are right.
Finally, don’t amplify it. Avoid sharing suspicious content to ask if it is real. Verify it first.
As political deepfakes become more sophisticated, spotting visual imperfections will become increasingly difficult. The challenge for voters is therefore changing. We need to become better at questioning what we see, without automatically disbelieving it.
In an age of increasingly convincing AI, healthy scepticism matters. But so does knowing when that scepticism has gone too far.
Republished from The Conversation
Rory Mulcahy
Rory Mulcahy is an Associate Professor at UniSC the School of Business and Creative Industries. His research focuses on topics such as consumer vulnerability, emotions, technology, and sustainability. He has collaborated with various organisations, including government, NGO’s, and start-ups. He has published his works in various journals, including Journal of Service Research, European Journal of Marketing, and Industrial Marketing Management, among others.
Aimee Riedel
Dr Aimee Riedel is a Senior Lecturer in Marketing at Griffith University and Co-Director of the Social Marketing Insight Lab. She is an emerging research leader whose program examines how vulnerability is created and amplified in digital and service environments, particularly in relation to artificial intelligence, misinformation, and complex service systems. Dr Riedel’s research focuses on understanding why some individuals and communities are more susceptible to harm and how organisations and policy can better anticipate and respond to these risks.
Asha Worsteling
Asha Worsteling is a PhD student at the University of Queensland and a research assistant at QUT and Griffith University, working on projects funded by the Department of Social Services and by the Office of National Intelligence. Asha’s work is united by a broader interest in how service design can be harnessed to tackle societal challenges; whether that’s building social connection and inclusion, improving economic participation for people with disability, or protecting public trust from AI-generated deepfakes.
Lucas Whittaker
Dr Lucas Whittaker is a Lecturer in Marketing in the Department of Management and Marketing at Swinburne University of Technology. His research explores the implications of deepfakes and Generative AI as an emerging innovations for marketing communication and how individuals appraise uses of such technologies in persuasive contexts. He has published research in journals including the International Journal of Information Management, Psychology & Marketing, and Technovation.
Renee Barnes
A journalist of more than fifteen years, Renee has worked for numerous media organisations including the ABC, The Age and Australian Associated Press. Renee’s passion is online journalism and her research and teaching focus on this field. She still regularly contributes to The Age and provides training and consultancy to the journalism industry on online journalism and social media. Renee’s current research is focused on online participatory behaviours.