Britons no better than Americans at spotting deepfakes

Britons were no better than Americans at identifying deepfakes despite being more familiar with the term, according to research commissioned by Veriff. The UK was the only country in the study where familiarity with deepfakes did not improve detection.

The survey, conducted by Kantar among 1,000 UK adults aged 18 to 64, found that 74% were familiar with the term deepfake. Yet only 53% scored above chance when asked to distinguish real from AI-generated content. Another 32% scored below chance, and just 16% reached the highest accuracy band.

The findings point to a gap between confidence and performance. Some 44% of UK adults said they were confident they could spot deepfakes, but nearly 29% of that group still performed below chance.

Men were more likely than women to say they were confident, at 52% versus 38%, although their measured accuracy was almost identical at 53.1% and 52.7%. Younger adults performed better: those aged 18 to 34 were nearly three times as likely as people aged 45 and over to reach the top accuracy tier.

Awareness alone did not solve the problem. Britons were 11 percentage points more likely than Americans to know the term, at 74% versus 63%, but were slightly less likely to reach the top band, at 16% versus 18%.

Routine checking habits also made little difference. UK adults who said they regularly verified suspicious content against another source were no more accurate than those who said they never did.

People who had created AI images or videos themselves were twice as likely as non-creators to describe themselves as confident, at 63% versus 32%. Their test scores, however, were no better.

Video problem

Video was the weakest area for respondents. Only 27% of UK adults correctly identified AI-generated video as synthetic, compared with 41% for face-swaps and 53% for still photographs.

The study suggests that the more lifelike the format, the harder it is to judge by sight alone. Respondents most often said they relied on unnatural-looking skin, odd movement in video and unusual details in hair, teeth or eyes, but those cues had only a weak link to better performance.

Almost half of UK adults said they relied on gut feeling. The research found no meaningful connection between instinct and accuracy.

No UK respondent achieved full marks. The best individual result on the stand-alone visuals was 14 correct out of 16, and only five respondents scored 13 or higher. On the side-by-side comparisons, 18 respondents answered all eight correctly.

Participants were shown 16 stand-alone visuals, split evenly between AI-generated and authentic material, as well as eight side-by-side comparisons across photographs, face-swaps and video. They were told in advance that some of the material had been generated by AI, so the results represent a best-case scenario rather than a real-world setting.

Trust gap

The findings also showed low confidence in online platforms' ability to identify manipulated media. Just 42% of UK adults said they trusted platforms to identify and label AI content, compared with 57% in the US.

Among people aged 45 and over in the UK, that figure fell to 22%. Meanwhile, 58% of UK adults said they believed they had already encountered a deepfake online, while 24% said they would not know if they had.

Concern about the wider effects of synthetic media was high. Three quarters or more of respondents said they worried about deepfakes spreading political misinformation, enabling fraud or scams, being used for harassment or exploitation, or eroding trust in what they see online.

Veriff, which sells identity verification tools, said the results suggest consumers can no longer rely on visual judgement alone as manipulated media becomes more convincing and harder to detect through simple observation.

Video poses a particular challenge in identity checks and other digital transactions because the issue is not only whether footage looks real, but whether its source and method of capture can be verified. Manipulated material may be delivered through virtual cameras, emulators or software injection tools rather than appearing as an obviously altered clip.

"Video creates a paradox. For people, motion can make synthetic content feel more real. A face that moves, blinks and speaks naturally is harder to question than a still image. For detection systems, however, video can provide more evidence, not less: frame-to-frame consistency, camera-source signals, sensor data, behavioural cues and signs that a stream has been manipulated," said Ira Bondar, fraud platform lead at Veriff.

Bondar said the debate still puts too much weight on individual judgement. "Much of the public discussion still assumes people can spot deepfakes by looking harder at the content. Our findings suggest the burden increasingly needs to shift from human judgement to systems that can test authenticity more directly."