A large share of social media users cannot properly distinguish between comments written by AI and posts written by real people.

TechRadar reported on Sunday that in an experiment run by cybersecurity company Surfshark with master’s students at Malmo University, only 53 percent of 710 participants identified bots more accurately than humans, while 47 percent did not pass the task.

The experiment was conducted as a timed simulation called "Bot or Not". Participants acted as content reviewers and had 120 seconds to find 10 bot-written comments across four topics. The tool was made by Malmo University master’s students in interaction design for the UNFOLD exhibition at Milan Design Week.

Differences by topic were clear. On data centres, a topic with less emotional involvement, the bot detection rate was 71 percent and accuracy was 76 percent, the highest. In the pineapple pizza debate, the detection rate was 64 percent and accuracy was 69 percent. On immigration, which was more emotional and political, the detection rate fell to 54 percent and accuracy to 63 percent. On women’s rights, the detection rate dropped to 49 percent and accuracy to 61 percent. Cases of mistakenly judging real people as bots also increased.

A gap by age was also confirmed. Participants aged 20 or younger found about 65 percent of bots and recorded accuracy of more than 71 percent, the highest performance. That trend continued through those in their 20s and 30s, but in the 41 to 50 group the detection rate fell to 42 percent and accuracy to 59 percent. Those aged 50 or older remained slightly better than that.

Luis Costa (루이스 코스타), Surfshark's research lead, said the experiment could not be explained only by reading ability or media literacy in the traditional sense. He said emotions can cloud judgement in filtering out suspicious content when debates overheat. He added that to respond to automated deception, a calm attitude and awareness of one’s own vulnerabilities matter more than sophisticated sentence analysis.

The influence of bots is already growing. Industry estimates show about 23 percent of political discourse on X during election periods is accounted for by bot amplification. In a previous Surfshark study, major platforms were estimated to delete more than 6.3 billion fake accounts each year. "Bot or Not" is currently available to anyone on a web browser, and participants can compare their scores with those of the 710 people.