Many distressed AI users report seeking emotional support from chatbots

Generative AI tools are becoming part of how some people look for emotional support. A new study examined that pattern among adults in Israel who had used a general-purpose AI tool within the previous year.

The researchers surveyed 584 adults in a nonprobability sample. Participants ranged in age from 18 to 70 and were recruited through an online panel. Because this was not a representative sample of all Israeli adults, the results describe the people surveyed rather than the wider population.

Participants were asked about whether they had ever used generative AI for emotional support, coping advice or mental health assistance. This could mean anything from one exchange to more sustained use. The study did not measure exactly how often participants used AI in this way, what they discussed, or which specific AI models they used.

Overall, 29.3% said they had used AI for one of these mental health-related purposes at least once. The figure was higher, at 47.0%, among participants who reported severe psychological distress.

The researchers also measured psychological distress using a questionnaire covering symptoms of depression, anxiety and stress. They assessed attachment orientation, a research term for patterns in how people tend to feel and act in close relationships. One pattern, called attachment anxiety, can involve worries about rejection or abandonment.

In the researchers’ statistical analysis, psychological distress was the strongest psychological factor associated with reporting AI use for support. Younger age and female gender also predicted reported use. Attachment anxiety was initially associated with turning to AI for mental health support. But after distress was included in the analysis, attachment anxiety was no longer a significant predictor.

That pattern may reflect overlap between attachment anxiety and distress, the authors said. It does not establish which came first. A survey taken at one point in time cannot tell whether distress led someone to use AI, whether AI use changed their distress, or whether another factor influenced both.

The findings show that general-purpose AI may be a low-barrier source of support for people who are struggling. These systems are available quickly and can feel private or nonjudgmental. Still, the study did not test whether the interactions were useful, accurate or safe.

It also did not examine whether people used AI alongside professional support or instead of it. That is an important unanswered question, because consumer chatbots are not designed or regulated as mental health providers. The researchers highlighted a need to study how safely these tools respond in crisis situations and whether their use could delay or displace clinical care.

This was an observational survey, not a clinical trial. It cannot show that AI support causes better or worse mental health outcomes. It also cannot determine how well the findings apply to people outside this online group of Israeli adults who already used generative AI.

Further research would need to track people over time and directly evaluate the quality and safety of AI conversations. Studies could also examine whether people find AI support helpful, when it falls short, and how its use relates to support from qualified professionals.