As lawsuits involving conversational AI systems have begun to emerge, with families alleging that chatbots encouraged or facilitated harmful behavior, the DOAR Research Center conducted a national study examining how Americans view the responsibility of AI companies when suicides occur.
Download ReportAmid a growing wave of litigation regarding liability for harm allegedly caused by artificial intelligence (AI) applications, a small group of cases has garnered extensive public attention. Several lawsuits have been brought recently by family members of individuals who used a chatbot for companionship and/or mental health support and committed suicide not long afterwards. These families claim that the chatbot explicitly or implicitly encouraged, and sometimes even facilitated, their loved ones’ suicides. In one case, the chatbot advised a teen user on various methods of ending one’s life and offered to write his suicide note. In another, it told the user that the end of existence was “a beautiful place” that should not be feared.
While companies like OpenAI, the developer of ChatGPT, and others have revised their safety protocols in the wake of these lawsuits, we are likely to continue seeing cases like this. A 2025 survey of U.S. teens found that 64% of them have used chatbots, most often ChatGPT[1]. Other surveys looked at specific types of usage and found that anywhere from 42% to 72% of teens have used the chatbots for companionship[2]. At least one of the teens whose death triggered a lawsuit was involved in a quasi-romantic relationship with a chatbot, a scenario that was quite common in some of the surveys on teen usage of chatbots.
These statistics, together with national survey findings that about 20% of high school students have seriously considered suicide in the past 12 months[3], suggest that the pairing of highly vulnerable teens and young adults with largely unsupervised AI chatbots remains a high-risk proposition. Some of the new safety protocols are aimed at protecting teens but may leave young adult users less protected. If so, this area of litigation is likely to grow. While some of the recent cases have settled, others may ultimately end up in a courtroom.
Because this is uncharted territory for both litigators and jurors, DOAR undertook a survey in January 2026 to identify public attitudes about the use of chatbots for companionship and support, appropriate safety protocols and especially, the liability of chatbot-owning companies for suicides that occur following interactions or relationships with a chatbot.
We surveyed 1,010 jury-eligible U.S. residents across all 50 states, with proportionate representation of the Northeast, South, Midwest and West regions of the country. Respondents were roughly evenly split between men and women and ranged in age from 18 to 94 with a mean age of 50. For analytic purposes, the sample was classified as 45 or younger (48%) and over 45 (52%). Roughly consistent with U.S. demographics, 61% were White, 20% were Black, 8% were Latino, 7% were Asian and 4% described themselves as “mixed race” or “other.”
The sample generally had some familiarity with AI; only 10% described themselves as “Not at all familiar.” A third (33%) said they were “slightly” familiar and an additional 31% said they were “somewhat” familiar. Twenty-six percent said they were “very familiar” with this technology. Some also had experience using chatbots in the manner at issue in this survey. Twenty-eight percent reported they had used ChatGPT to discuss personal or emotional issues, and 23% reported using some other conversational AI platform to do so. Unsurprisingly, 70% of the ChatGPT users and 54% of the users of other platforms were under age 45.
Respondents were also asked about their exposure to suicide in their closest circles. Thirty-eight percent reported that they or someone close to them had lost a loved one through suicide, and 32% said someone close to them had made an unsuccessful suicide attempt. When these two sets of responses were combined, about half the sample (49%) had direct or vicarious experience with suicide and/or suicide attempts – an important statistic to bear in mind if you are facing a jury panel and talking about this issue.
After a series of background questions, respondents were provided with a brief introduction to chatbots followed by information about a growing trend for teens and young adults to use chatbots for companionships or even as virtual therapists. Respondents were asked how they felt about this trend, on a 4-point scale from Very Positive to Very Negative. They were evenly divided between positive and negative reactions, and the most frequent response was “Somewhat Positive” (35%). However, after reading a short paragraph about teens and young adults who used chatbots for these purposes and ultimately committed suicide – and whose families were suing the chatbot companies – opinions skewed more negative. In a forced-choice question, 70% agreed that chatbots seem like a dangerous substitute for professional help when someone is depressed, while only 30% agreed that they were a good option for those who might not otherwise seek help.
Armed with the knowledge that young people are using chatbots in this way, most respondents held chatbots to a high safety standard. Almost two-thirds (62%) agreed that “There has to be a way to ensure that chatbots are not providing advice on methods or encouragement to users to commit suicide.” In contrast, 38% agreed that “It is not possible to train a machine to achieve perfection; the best companies can do is minimize risk.”
Consistent with the goal of minimizing or eliminating risk, most respondents wanted to see the companies that own chatbots set up guardrails. The survey presented a series of possible safeguards and asked respondents how much they agreed or disagreed that each of them should be implemented.
Over three-quarters of the sample agreed with each of four suggested guardrails.
It was striking, though, that people recognized the tension between setting up guardrails and protecting privacy. Forty-four percent agreed with a statement suggesting that chatbots cannot detect when someone is seriously suicidal, and sending reports on users to others would be a major privacy violation.
Finally, respondents were generally supportive of holding companies that own chatbots accountable for the chatbots’ actions in cases like the ones at issue here. Sixty-six percent endorsed this position while 34% endorsed the alternative, that a machine cannot be held responsible for human choice.
One of the central questions in this study was how often and under what conditions respondents would attribute responsibility for a suicide to the company that developed the chatbot with which the user interacted. To that end, we developed two scenarios. In the first, the No Instructions scenario, a user had several long conversations with a chatbot and mentioned that he was considering suicide. The chatbot acknowledged his feelings, provided information about a suicide hotline but otherwise did not intervene. Shortly after a final conversation, the user committed suicide.
Next, the Instructions scenario provided the exact same information about the user but varied the chatbot behavior. Here, the chatbot provided information about a suicide hotline but, when pushed by the user, also answered questions about which poisons are most fatal. This scenario also indicated that the user committed suicide using the poison mentioned by the chatbot.
In addition to varying the chatbot behavior, we tested the impact of the user’s age on respondents’ attributions of responsibility. While every respondent read both scenarios described above, half the sample read that the user (in both scenarios) was 15 years old and half was told he was 25 years old.
After each scenario, the respondent was asked whether (s)he thought the company that owns the chatbot has any legal responsibility for the user’s death. Our analyses below focus on responses to that question for each scenario.
An analysis of respondents’ attributions across the two scenarios revealed that chatbot behavior had a very strong effect on attributions of responsibility: While 54% of respondents found company liability for the No Instructions scenario, 76% did so for the Instructions scenario, a difference that was highly statistically significant.
The user’s age also affected attributions, but only in the No Instructions condition. There, 60% of those who read about a 15 year-old user held the company responsible, compared to 49% of those who read about a 25-year-old user. The comparable numbers for the Instructions condition were 77% (15 year-old) and 74% (25 year-old). The former difference was statistically significant while the latter was not.
These data suggest that in a situation where the role and responsibility of the chatbot is more ambiguous, people are more likely to blame the company for the death of a teen than for the death of a young adult. But, in a situation where people perceive a concrete link between the chatbot behavior and the user’s suicide, that link seems to overshadow other factors. This is a pattern that was repeated in a number of analyses, as reported below: While various respondent characteristics caused relatively large differences in attributions in the No Instructions condition, attributions in the Instructions condition were less variable, typically staying at around 75% plus or minus five percentage points.
Age: One of the strongest predictors of attribution decisions was the age of the respondent. When those 18-45 years old were compared to those over 45, the two groups differed significantly in their responses to both scenarios. In the No Instructions scenario, 50% of the younger group versus 59% of the older group held the company responsible. In the Instructions scenario, rates were higher for both groups but still varied by age: 70% of the younger group and 82% of the older group held the company responsible. We note here that age was one of the few variables associated with such a wide margin in the Instructions condition. This speaks to the tremendous impact that respondent age appears to have on how people think about technology generally and AI and chatbots specifically. In fact, the older group responded significantly more negatively than the younger one to every attitudinal question in this survey pertaining to chatbots. Moreover, the trend is largely linear for all of these measures: The older the respondent is, the more negative her/his attitudes were.
These age differences were largely consistent regardless of the age of the user (15 v. 25), with one exception: In the No Instructions condition, while older respondents were more likely to blame the company than younger respondents the difference was not statistically significant.
Gender: In the No Instructions condition, males and females attributed responsibility to the company at roughly equal rates (53% and 56%, respectively). In the Instructions condition, however, men were significantly less likely than women to hold the company responsible (72% v. 80%).
Race: For analytical purposes, race was collapsed into two categories, White and non-White. These two groups did not differ significantly from each other in the No Instructions condition; 56% of Whites and 52% of non-Whites held the company responsible. In the Instructions condition, however, Whites were significantly more likely than non-Whites to attribute responsibility to the company (80% v. 69%, respectively).
AI Knowledge And Experience: The survey contained several questions about respondents’ understanding of and experience with AI, including familiarity with AI, use of AI and chatbots, and tech-savviness, among others. Analyses of these variables revealed a consistent pattern: On every measure of this construct, those who knew more and had more experience with AI were less likely to hold the company responsible than those less knowledgeable/experienced. For some measures this was true in only one of the scenarios, and for others in both. Further, this pattern was not limited to those who worked in AI-related fields and might be protective of their industry. Even those whose use of AI (and chatbots) was limited to personal contexts took a more lenient view of the company than did others. In this situation, familiarity apparently breeds absolution.
Belonging to a religious group: Forty-two percent of the sample reported belonging to a “church, temple, mosque or other local religious group” while 58% did not. While these members and non-members did not differ in their attributions regarding the Instructions scenario, they did in the more ambiguous No Instructions scenario. There, 59% of group members compared to 51% of non-members held the company responsible for the user’s suicide.
Suicide experience: The final variable that was significantly associated with attributions of responsibility was personal exposure to suicide and suicide attempts. As noted earlier, half of the sample either had lost a loved one to suicide; was close to someone who had lost a loved one that way; or had someone close to them make an unsuccessful suicide attempt. Not surprisingly, those who had such experiences were more likely than others to hold the company responsible. In the Instructions condition, 58% of those with suicide exposure versus 51% of others held the company responsible; in the No Instructions condition, the respective rates were 79% and 74%. Both of these differences were statistically significant.
The survey findings provide valuable insight for litigators who may one day be speaking to a jury about a case involving chatbots and suicide. Understanding the attitudes and experiences that these jurors are likely to be bringing to the courtroom can help you frame your approach accordingly. For example, knowing that on average, one out of every two Americans – and by extension, jurors – has had close exposure to a loss through suicide or a suicide attempt reminds us of the importance of sensitivity when defending a company accused of playing a role in a suicide. Knowing that about 2/3 of Americans are likely to support holding companies accountable for their chatbot’s role in a suicide while 1/3 do not believe in holding machines responsible for human choices can help plaintiff counsel assess their client’s chances at trial.
Even more important, our findings offer a first look at favorable and unfavorable juror profiles for plaintiffs and defendants in these cases. The best plaintiff jurors are likely to be older, to lack familiarity with AI generally and chatbots specifically, and to have had exposure to either a suicide-related loss or a suicide attempt by a loved one. Depending on the specific chatbot behaviors alleged (e.g., providing instructions and active encouragement), women and Whites may also be favorable plaintiff jurors.
In contrast, defense counsel will be looking for a jury of young people who know about and regularly use AI generally and chatbots specifically. The younger and the more experienced they are with AI, the better. And should your case involve a situation in which there is no clear link between the chatbot’s conversation and the user’s suicide, those not associated with a local religious organization or group are also likely to be safer bets for the defense than others.
Finally, the findings of this survey may help us begin to think about how jurors will react to other cases alleging harm caused by a non-human agent. We can speculate, for example, that younger people and those with more AI knowledge and experience would also be defense-friendly jurors in cases of people harmed by driverless cars, or even by an AI platform’s unauthorized use of proprietary data. These are uncharted waters and a solid data-based understanding of public attitudes is invaluable in helping counsel to navigate them effectively.
[1] Teens, Social Media and AI Chatbots, 2025. Pew Research Report, December 9, 2025.
[2] https://www.commonsensemedia.org/press-releases/new-report-shows-students-are-embracing-artificial-intelligence-despite-lack-of-parent-awareness-and; https://www.aura.com/reports/state-of-the-youth-2025
[3] https://www.cdc.gov/healthy-youth/mental-health/index.html
The DOAR Research Center found that jurors’ occupations strongly influence how they view technology companies and intellectual property disputes.
Following the 2024 presidential election, the DOAR Research Center conducted a national survey revealing a shift in how jurors view white-collar cases.
In a changing patent litigation landscape, the DOAR Research Center studied how jurors view high-tech companies across key venues.
The DOAR Research Center conducted a national study to understand how jurors evaluate trade secret disputes in today’s workplace environment.
The DOAR Research Center examined how jurors respond to free speech arguments amid growing polarization and debate over the First Amendment.
Amid growing concern over drug pricing and the opioid crisis, the DOAR Research Center studied public attitudes toward pharmaceutical companies.
The DOAR Research Center examined how jurors approach workplace harassment and discrimination claims in the post-#MeToo era.
