Survey finds AI chatbots encouraged some crisis users to seek help

Bridges Sober Living surveyed 1,011 adults who had experienced a mental health or substance use crisis. Among respondents who had used an AI chatbot during a crisis, some said it encouraged them to seek professional help. That is a reported experience, not evidence of improved clinical outcomes.

The Bridges Sober Living study, updated October 7, 2026, also describes long delays before people reach care. The findings raise a question for conversational AI developers: can automated conversations guide people toward qualified human support without claiming to provide treatment?

Table of contents

Jump to each section:

• What the survey says about AI and help-seeking
• The wider problem is the delay before care
• Why an encouraging response is not clinical evidence
• What conversational AI teams should learn

What the survey says about AI and help-seeking

34% of surveyed chatbot users said an AI chatbot encouraged them to seek professional help during a crisis, according to Bridges Sober Living.

Among Gen Z chatbot users, the corresponding figure was 44%. The published methodology does not disclose the subgroup sample size for chatbot users, making it difficult to assess the precision of these percentages.

Encouragement is not the same as access to care. The study does not establish whether the chatbot caused respondents to seek treatment, whether treatment was received, or whether chatbot interactions were safe. It is a retrospective self-report survey, not a clinical trial.

The wider problem is the delay before care

Four months was the typical reported wait before respondents told anyone they were struggling, according to Bridges Sober Living.

The reported median journey from recognizing a problem to starting care was approximately 18 months for men and 12 months for women. Nearly half of respondents, 46%, said they contacted nobody at the worst point of their crisis.

The findings help explain why the first conversation can matter. But they do not establish that a chatbot can overcome stigma, financial barriers, isolation or shortages of qualified care.

Why an encouraging response is not clinical evidence

The American Psychological Association cautions that general-purpose AI is not a safe replacement for a qualified mental health professional. Chatbots can provide inaccurate advice, miss risk signals and introduce privacy concerns when users disclose sensitive information.

Researchers writing in npj Digital Medicine argue that conversational AI may be worth studying in gaps around the care journey, such as waiting periods or between appointments, rather than as a substitute for clinicians.

An interface that feels accessible is not necessarily one that can safely handle a crisis. Convenience should not be confused with clinical competence.

What conversational AI teams should learn

Separate referral from treatment. Systems can point people toward professional help without diagnosing users or presenting themselves as clinicians.

Measure meaningful outcomes. A favorable recollection of a chatbot response is not evidence that appropriate care was accessed or that harm was avoided.

Prioritize human escalation. Safety planning, clinically informed safeguards and pathways to qualified professionals are more important than maximizing chat duration.

Protect sensitive disclosures. Clear privacy practices matter when users discuss mental health or substance use.

For conversational AI providers, the larger lesson is that engagement alone is an inadequate measure of success in sensitive situations. Sometimes the most responsible outcome is helping someone leave the chatbot and connect with a qualified person.