Public perception of artificial intelligence in healthcare shapes acceptance, implementation and everyday use. Evidence from a large regional cohort in Denmark spanning 2022 to 2024 shows how first-hand exposure to ChatGPT related to shifts in views about benefits and risks. Participants reported their perception of artificial intelligence before the public release of ChatGPT, then again two years later alongside self-reported use of the tool. Analyses compared people who had tried or often used ChatGPT with those who had not. Across the period, views became more cautious overall, yet exposure was linked to reduced uncertainty and a greater likelihood that previously undecided respondents would see more benefits than risks. The findings highlight how direct interaction with conversational AI can influence attitudes during a time of growing scepticism.
Survey Cohort and Exposure
The analysis drew on 5,899 adults who completed both waves of the Health in Central Denmark survey. Perception of artificial intelligence in healthcare was captured with four options: benefits outweigh risks, benefits equal risks, risks outweigh benefits or don’t know. In 2024, participants were also asked whether they used ChatGPT, with responses grouped as exposed (tried or often used) or unexposed. At follow-up, one in five respondents fell into the exposed group, while a sizeable proportion of the unexposed reported having never heard of the tool. The exposed group tended to be younger, more often men and more highly educated, with a slightly lower share living with diabetes than the unexposed group. Before ChatGPT’s release, those who would later be exposed already showed a higher share perceiving benefits and a lower share reporting uncertainty than those who remained unexposed, indicating differing starting points between the groups.
Must Read:Co-Designing Equitable Virtual Hospital-at-Home Services
Shifts in Perception Between 2022 and 2024
Population-level patterns moved towards caution over time. The share reporting benefits declined from 40% at baseline to 35% at follow-up, while the proportion perceiving mainly risks rose to 6.7%. Uncertainty remained common at both time points. Within this broad shift, exposure was associated with more decisive movement among those initially unsure. Almost three in ten exposed individuals who began with “don’t know” shifted to “benefits”, roughly twice the proportion seen among unexposed respondents starting from the same position. Positive views were also more stable among people who had used ChatGPT. A clear majority of exposed individuals who began with benefits maintained that stance at follow-up, compared with a smaller majority among the unexposed. Supplementary weighting and unweighted checks produced directionally consistent patterns, although the magnitude of some changes from uncertainty varied across approaches. Taken together, the trends suggest a general cooling of enthusiasm in the population alongside a distinct stabilising effect among those with first-hand experience of the tool.
Exposure Linked to Reduced Uncertainty
Adjusted, weighted models examined how exposure related to changes in perception between the two surveys. Among respondents who were undecided at baseline, exposure to ChatGPT was associated with markedly higher odds of moving to benefits rather than remaining uncertain, with an odds ratio of 3.21 and a confidence interval that excluded one. For people who started with a defined view, exposure was linked to lower odds of becoming uncertain at follow-up. This pattern held whether the starting point was benefits, equal balance or risks, with odds ratios below one in each case. Baseline characteristics also related to the direction of change. Shorter education, older age and female sex were associated with lower odds of moving from uncertainty to benefits, and with greater likelihood of shifting into uncertainty from other starting positions. In models adjusted for rather than stratified by baseline perception, exposure remained associated with lower odds of reporting don’t know, equal or risks than benefits at follow-up. These results indicate that using or trying ChatGPT was linked to fewer uncertain responses and more movement toward benefits among those previously undecided, even as overall attitudes edged towards caution.
Two survey waves spanning the public release of ChatGPT show a population trend toward more cautious views of artificial intelligence in healthcare, alongside a clear association between exposure and reduced uncertainty. People who had tried or often used ChatGPT were more likely to move from indecision to perceiving benefits, and less likely to become uncertain if they already held a view. Demographic patterns suggest that age, sex and education shape these transitions, underscoring the relevance of digital familiarity for public attitudes. The findings point to the role of first-hand interaction with tools like ChatGPT in supporting informed, stable perceptions during periods of rapid technological change, while acknowledging that broader scepticism has grown over time.
Source: npj digital medicine
Image Credit: iStock
