A chatbot at 2 a.m. shouldn't be the only one who knows

5 min read

Share

A felt smartphone.

What a new Israeli study of AI users can and cannot tell us

On Oct. 29, Medical Xpress ran the headline "Nearly half of severely distressed generative AI users turn to chatbots for emotional support, study finds" [1]. The story covers a study by researchers in Israel, published in the journal Cyberpsychology, Behavior, and Social Networking [2]. It is a useful number, and an easy one to over-read.

The short answer: In an online sample of 584 Israeli adults who use generative AI tools, 29.3% said they had turned to them at least once for emotional support, coping advice, or mental health assistance. Among those reporting severe psychological distress, 47.0% had [2]. The study does not show whether that helped or hurt anyone. A chatbot may be where someone starts. It should not be the only place.

If you are in crisis in the US, call or text 988 to reach the 988 Suicide & Crisis Lifeline, free and confidential, 24/7 [6]. In an emergency, call 911.

What the study found

The researchers recruited adults aged 18 to 70 through an Israeli online panel in November 2025. Everyone had used a general-purpose AI tool, such as ChatGPT or Gemini, at least once in the past year [3]. Participants answered one yes-or-no question about using these tools for "emotional support, coping advice, or mental health assistance" and filled out a standard distress questionnaire [3].

  • 171 of the 584 users, 29.3%, said yes [2][3].
  • The share rose with distress: 24.3% in the low-distress group and 47.0% in the severe group [3]. That is 22.7 percentage points higher, or 1.9 times the rate.
  • In the authors' model, distress was the strongest psychological factor linked to use. Younger people and women were more likely to say yes [2].

The authors describe general-purpose AI as "an accessible, low-barrier resource disproportionately used by distressed individuals" and call for systematic study of its safety and of how well it responds in a crisis [2].

Three surveys, three different questions

There is no single number for how many people lean on AI this way. Here is how the Israeli study sits beside two recent US surveys. Each asked a different group a different question, so the rates are not directly comparable.

SurveyWho, and whenWhat it askedWhat it found
Hadar Shoval, Refoua, et al. [2][3]584 Israeli adults, 18 to 70, who used an AI tool in the past year; online panel, nonprobability sample; Nov. 2025Had used AI for "emotional support, coping advice, or mental health assistance"29.3% overall; 47.0% with severe distress
McBain et al., JAMA Network Open [4]1,058 US youths aged 12 to 21, nationally representative; Feb. to March 2025Had sought "advice or help" from AI "when feeling sad, angry, or nervous"13.1% overall; 22.2% at ages 18 to 21
Rousmaniere et al., preprint [5]499 US adults recruited on Prolific who had used an AI tool and reported a diagnosed, ongoing mental health condition; posted March 2025Used AI "for mental health support or therapy-related goals in the past year"48.7%

The two highest figures come from groups already under strain: Israeli users with severe distress, and US adults with a diagnosed condition. The US adult study is a preprint that has not been peer reviewed, led by researchers at a marriage and family therapy program [5]. And the Israeli figures describe Israeli AI users, not Americans.

What the study does not show

  • Whether it helped or harmed. The study measured use, not outcomes [3].
  • How much. One message counts the same as months of daily conversation. The figures "reflect any such contact, from a single exchange to sustained use," the authors write [2].
  • Cause and effect. It is a one-time survey, so it cannot say whether distress leads people to AI or the other way around [3].
  • Everyone. It is a self-reported, nonprobability online sample from one country [2][3].

The authors' preprint makes one practical suggestion: clinicians may want to ask distressed patients about their AI use, because a chatbot might become a hidden source of reassurance that bypasses the people around them [3].

What you can do

  • Tell a person who can act. A chatbot is not someone who can follow up with you next week or change your care. Tell a clinician, a counselor, or someone you trust what you have been working through.
  • Show your doctor what you asked. If you asked an AI about a symptom, a diagnosis, or a medication, bring the questions and answers to your next visit or message. It shows your clinician what is on your mind and gives them a chance to correct anything wrong.
  • Check medication answers with a professional. Before you start, stop, or change a medicine based on something an AI said, talk with your clinician or pharmacist.
  • Know where to go when it's urgent. In the US, call or text 988 any time, day or night [6]. In an emergency, call 911.

A chatbot can be where someone starts at 2 a.m. It should not be the only one who knows. If you use Temi for a prescription, tell the physician reviewing your request what you have asked an AI about your medications.

Sources

  1. Medical Xpress, "Nearly half of severely distressed generative AI users turn to chatbots for emotional support, study finds," Oct. 29, 2026. https://medicalxpress.com/news/2026-10-severely-distressed-generative-ai-users.html
  2. Dorit Hadar Shoval, Elad Refoua, Karny Gigi, Yuval Haber, Inbar Levkovich, Iftach Tsafrir, Ori Pen, and Zohar Elyoseph, "AI-Mediated Mental Health Support: The Role of Attachment Orientation and Psychological Distress," Cyberpsychology, Behavior, and Social Networking 29, no. 11 (2026): 809 to 813, published online Sept. 29, 2026. https://doi.org/10.1177/21522715261494423 (abstract: https://pubmed.ncbi.nlm.nih.gov/42808327/)
  3. Dorit Hadar Shoval et al., "AI-Mediated Mental Health Support: The Role of Attachment Orientation and Psychological Distress," preprint, PsyArXiv, posted Feb. 18, 2026. https://doi.org/10.31234/osf.io/vfz2s_v1
  4. Ryan K. McBain et al., "Use of Generative AI for Mental Health Advice Among US Adolescents and Young Adults," JAMA Network Open, Nov. 7, 2025. https://doi.org/10.1001/jamanetworkopen.2025.42281
  5. T. Rousmaniere, Y. Zhang, X. Li, and S. Shah, "Large Language Models as Mental Health Resources: Patterns of Use in the United States," preprint, PsyArXiv, posted March 18, 2025. https://doi.org/10.31234/osf.io/q8m7g
  6. 988 Suicide & Crisis Lifeline. https://988lifeline.org/

Editorial Standards

This article has not been reviewed by a licensed physician. We cite peer-reviewed studies, academic research institutions, and medical associations. Our sources are listed at the end of each article.

If you notice an error or have a concern about our content, please email supportusetemi.com.

This article is for informational purposes only and does not constitute medical advice. Always consult a qualified healthcare provider for diagnosis and treatment.

Felt Temigami re-enacting famous revolutions, from crossing the Delaware to raising a pill-bottle flag.

The revolution will be emailed.

Once a week, roughly. We're taking the fight to a corrupt and broken health insurance system. Join up.

Get Temi's newsletter by email. Unsubscribe anytime. Privacy policy

Temi

Start online, and a licensed physician reviews your request

See if you qualify

A licensed physician reviews every request. Nothing on this page is a prescription.