Teens prefer ChatGPT over therapists for mental health advice

Young people are increasingly seeking mental health guidance online, often turning to artificial intelligence (AI) tools before consulting professionals. The rise of generative AI platforms such as ChatGPT has introduced a new layer of digital support, raising urgent questions about credibility, safety, and emotional impact.

The study ChatGPT as a Mental Health Advisory Service: Comparing Evaluations from Youth and Health Professionals, published in Digital Health, examines whether AI-generated responses can match or outperform professional advice when evaluated by both teenagers and trained health practitioners.

Youth show strong preference for AI-generated advice

The study recruited 123 young people aged 16 to 20 from Norwegian high schools and 31 professional health workers, including psychologists, social workers, and educators involved in youth advisory services. Participants were presented with authentic mental health questions originally submitted to a Norwegian government backed youth advisory platform. For each question, they reviewed two responses: one written by a human health professional and one generated by ChatGPT using GPT 4.

The participants evaluated each response across four dimensions: validation, relevance, clarity, and utility. Validation measured whether the response acknowledged feelings and concerns. Relevance assessed how well it addressed the specific issue raised. Clarity focused on ease of understanding. Utility examined whether the advice was actionable and helpful.

The results were surprising. Young participants rated ChatGPT’s responses significantly higher in relevance and utility compared to those written by health professionals. They were more likely to recommend the AI generated answers and expressed a clear overall preference for them. For many youth respondents, ChatGPT’s responses felt structured, concrete, and immediately applicable.

Young participants frequently highlighted the practical nature of the AI responses. They valued step by step suggestions, clear formatting, and language that felt accessible. Structured explanations appeared to reduce confusion and increase confidence in how to act. In the context of emotionally charged topics, clarity and actionable advice seemed to matter more than formal authority.

On the other hand, professional health participants did not show a strong preference for either source. Their ratings of ChatGPT and human-authored responses were largely similar across the four evaluation categories. While they acknowledged that AI responses were often well written and informative, they did not consider them superior to professional advice.

The study found fewer differences between the two groups than expected. Both youth and professionals rated responses similarly in validation, clarity, and overall usefulness. However, youth consistently leaned more favorably toward ChatGPT in areas related to relevance and actionable guidance.

Professionals raise concerns about relational depth and ethics

While the quantitative data showed broad similarities, the qualitative feedback revealed important differences in perspective. Professional health participants were significantly more critical when identifying weaknesses in the responses, especially those generated by ChatGPT.

Health professionals frequently pointed to concerns about relational nuance. They noted that some AI responses, while clear and comprehensive, felt overly factual or emotionally flat. In mental health communication, tone and subtle empathy are essential. Professionals emphasized the importance of validating a young person’s emotional state in a way that aligns with established health communication guidelines.

Another concern involved cognitive overload. Some professional participants felt that ChatGPT responses were sometimes too detailed or offered too many suggestions at once. Excessive information can overwhelm vulnerable individuals, particularly when they are already distressed.

There were also concerns about professional boundaries. Health professionals are restricted in how they provide advice online. They must avoid offering diagnoses or individualized treatment recommendations without proper assessment. Some AI responses appeared to move closer to clinical language or medical framing than is typically permitted in online advisory services. Although no major factual inaccuracies were identified, professionals were cautious about how such responses might be interpreted.

Youth participants, on the other hand, rarely raised these concerns. They were less focused on strict professional standards and more concerned with whether the answer felt helpful, understandable, and supportive. Many youth respondents described feeling seen and supported when reading detailed AI responses. Comprehensive explanations were often interpreted as signs of care and effort.

This difference highlights a subtle but important tension. Young users may equate depth and detail with empathy and attentiveness. Professionals, however, evaluate responses through the lens of ethical responsibility, communication guidelines, and clinical appropriateness.

Rethinking media richness in text-based mental health support

The analysis is based on Media Richness Theory, which traditionally argues that richer communication channels reduce ambiguity through cues such as tone, immediacy, and personal interaction. In face to face settings, nonverbal signals enhance understanding. Text-based communication, by contrast, is typically considered less rich.

However, the study suggests that in digital youth mental health contexts, richness may be defined differently. Rather than relying on facial expressions or real time feedback, richness in this setting appears to depend on clarity, structure, relevance, validation, and perceived credibility.

ChatGPT’s strength lies in delivering well-structured, logically organized, and easy-to-read responses. This format aligned with the communication habits of digitally native youth. By presenting actionable steps and clear explanations, the AI reduced uncertainty and made guidance feel immediately usable.

For youth, this form of structured clarity translated into perceived richness. For professionals, richness also required relational subtlety and adherence to established ethical norms. The two groups evaluated communication quality through overlapping but not identical lenses.

Importantly, the study did not find that professionals rejected AI advice outright. Instead, they viewed it as potentially useful but in need of careful oversight. Their skepticism was measured rather than dismissive. They recognized the strengths of AI in clarity and efficiency but emphasized the importance of human judgment.

A hybrid model for youth mental health services

AI systems such as ChatGPT should not replace human professionals in youth mental health services. Instead, they propose a hybrid advisory model that combines the scalability and clarity of AI with the ethical oversight and contextual expertise of trained professionals.

In such a model, AI could assist in drafting structured and accessible responses, helping to manage large volumes of inquiries. Health professionals would review, refine, and contextualize these responses, ensuring compliance with clinical guidelines and ethical standards.

This approach could expand access to mental health support while maintaining safety and quality. Youth facing stigma, fear of judgment, or barriers to professional care may benefit from AI-supported services that feel approachable and immediate. At the same time, professional involvement would guard against misinformation, inappropriate framing, or overreliance on automated advice.

Millions of users worldwide are already turning to AI systems for emotional support. As reliance grows, questions of trust, authority, and accountability become increasingly urgent. AI-generated empathy may feel genuine to users, but it is ultimately simulated. Ensuring that young people understand both the strengths and limits of such systems is essential.

Limitations and future research

The authors acknowledge several limitations. The study excluded highly sensitive topics such as suicidal thoughts and sexual abuse for ethical reasons. As a result, it remains unclear whether preferences would shift in more acute or crisis situations. Previous research suggests that users may prefer human guidance for highly sensitive matters.

The study was conducted in Norwegian and focused solely on ChatGPT using GPT 4. Results may vary with other models, languages, or cultural contexts. The professional sample size was limited to available participants within the advisory platform.