Even as AI therapy tools improve, people consistently choose human therapists for the deep work. The research explains why.
Reading time: 6 minutes
Key Takeaway
AI can produce empathetic-sounding responses that sometimes outscore humans in blind tests. But clients consistently prefer human therapists because the relational contract—accountability, lived experience, authentic presence—remains uniquely human.
The Rise of AI in Mental Health
Artificial intelligence has made significant inroads into mental health care. Chatbots like Woebot, Wysa, and Replika offer 24/7 support, mood tracking, and structured cognitive behavioural therapy exercises. Large language models can now generate therapeutic-sounding responses that, in some cases, clients rate as more empathetic than responses written by human therapists [1]. This rapid improvement has led some commentators to ask whether human therapists are becoming obsolete. The answer, grounded in emerging research, is more nuanced.
The Empathy Paradox: Preferring People Despite the Data
A fascinating paradox has emerged in recent research. When participants are shown side-by-side responses from AI and human therapists, they often rate the AI responses as more empathetic—yet they consistently say they would prefer to work with a human [1]. This phenomenon, termed the AI empathy choice paradox, reveals that therapeutic preference is driven by more than perceived empathy. Clients value the knowledge that their therapist is a real person who has chosen to be present for them, who has their own lived experience, and who is accountable for their well-being. This relational contract cannot be replicated by an algorithm, no matter how well it mimics empathic language.
What Only a Human Therapist Can Offer
The therapeutic alliance—the collaborative relationship between client and therapist—is consistently the strongest predictor of positive outcomes across all therapy modalities. While AI can simulate components of this alliance, several elements remain uniquely human. These include the ability to hold a client’s emotional experience in real time with authentic compassion, to notice subtle non-verbal cues that may contradict spoken words, to draw on personal and professional intuition when standard approaches are not working, and to offer appropriate self-disclosure that validates the client’s experience [2]. A narrative review of AI’s impact on the therapeutic alliance concluded that AI can augment but not replace the human relational factors that drive therapeutic change [3].
AI as a Tool, Not a Replacement
The most effective integration of AI into therapy is not replacement but augmentation. AI tools can handle between-session check-ins, track mood patterns over time, suggest relevant psychoeducational content, and even generate draft progress notes that the therapist reviews and personalises. This frees the practitioner to focus their energy on the deep relational work that only they can do. Many practitioners are now using AI to enhance their practice efficiency while maintaining the human core of their therapeutic work. The key is transparency: clients should know when and how AI is being used in their care, and consent should be obtained for any AI-assisted components.
How to Talk to Clients About AI
Clients increasingly arrive in the therapy room having already interacted with AI for mental health support. This is not a threat but an opportunity. A client who has used Woebot or ChatGPT for basic support is already engaged in self-reflection and may be more ready for the deeper work that a human therapist provides. When a client mentions using AI, respond with curiosity rather than defensiveness. Ask what they found helpful and what they felt was missing. This opens a conversation about the unique value of human connection in therapy and helps the client articulate what they hope to gain from working with you that no chatbot can provide.
References
- Wenger, J., Cameron, D., & Inzlicht, M. (2025). The AI empathy choice paradox: People prefer human empathy despite rating AI empathy higher. DOI: 10.31219/osf.io/ghw2v_v2
2. R, M., M, V., & K B, T. (2025). AI-enhanced therapeutic alliance: Bridging AI innovation and mental health care. Mental Health and Social Inclusion, 1–16. DOI: 10.1108/mhsi-10-2025-0281
3. Malik, S., & Srivastava, P. (2025). Impact of artificial intelligence on the therapeutic alliance in mental health care: A narrative review. AI and Sustainable Transformations, 189–194. DOI: 10.1201/9781003642886-26
→ Browse more research at whatwasi.com/students/research/