The Third Actor in the Room: How AI Is Quietly Changing Therapy

Well. Now what?

This isn’t the first time that this has happened in session to me. A client may sit down at a first session and offer, “I think I have avoidant attachment. I asked ChatGPT.” A couple may argue about advice one partner received from an AI chatbot the night before. A teenager tells you they feel understood—“more than by anyone else”—because of how an app responds at 2:00 a.m. Another client said, “I have the whole fight with my wife recorded, and AI said she was wrong,”…and proceeded to send me the recording (despite my insistence not to send it). These are all too common occurrences in therapy. Artificial intelligence has entered therapy: just as therapists are using it to make workflow more efficient, clients are relying on it to process emotions, seek relationship advice, rehearse conversations, self-diagnose, interpret their perceived symptoms, and make decisions before they ever walk into the therapy room.

AAMFT recently highlighted these dilemmas in their hosted “LinkedIn Live” discussion called Navigating AI’s Complicated Role in Systemic Therapy. The panel explored the evolving role of AI in therapy, including tools that support efficiency to the ethical, relational, and clinical complexities they introduce. The topics covered included informed consent, emerging regulations, client use of AI, and what these technologies can (and cannot) replace in systemic work. This event provided commentary that was very thoughtful, provided a real-world perspective for clinicians, media, and the broader public. But questions still remain. How do systems therapists navigate and understand the new entity in the system: the machine?

It’s Not Just Information: It’s Relational Influence

What makes generative AI different from researching symptoms online or reading a self-help book is what we think sets us apart as a profession: the interaction. Clients do not simply retrieve information. They enter a relationship with the device, listen carefully to the guidance, and conversation from a self-appointed digital expert. These ideas about attaching to electronics were referenced by Hertlein and Twist here in Family Therapy Magazine in 2018. Part of the draw is how quickly technology responds. A client asks, “Why do I feel this way?” AI offers an explanation. The client feels understood and acknowledged so they ask another question. AI refines the answer. The client adds more information, receives more validation, and continues. Over time, the interaction can shape how clients understand themselves, interpret relationships, and decide what to do next. By the time they arrive in therapy, part of their explanation, recollections, and what they share with the therapist is likely already co-constructed with AI.

This is clinically important because AI can appear and sound coherent, confident, and affirming even when a situation is ambiguous or complicated. A client may frame a problem in a way that invites one interpretation and receive increasingly polished versions of that interpretation. Hertlein and Springer (2026) warn that AI can reinforce confirmation bias for clients and therapists alike when its outputs are treated as objective truth rather than one possible perspective. The systemic question, then, is not simply, “Is the AI right?” It is: “What is this interaction doing in the system?” What does it reinforce? What does it miss? What happens between people after the client closes the app?”

When AI Becomes a Regulator

For many, AI is becoming a place to go when they feel overwhelmed, lonely, ashamed, angry, or uncertain (Hertlein, 2025). It is available immediately. Functionally, the pattern can resemble familiar attachment-related experiences: consistency, availability, predictability, and responsiveness (Hertlein & Twist, 2018) but lacks the mutuality of a human relationship (Ferrario et al., 2024). A chatbot does not have needs that must be negotiated, nor does it require the client to tolerate disappointment, repair a rupture, or make room for another person’s perspective. This creates an important distinction for therapists. AI can support human connection when it functions productively. A client might use it to organize thoughts before a difficult conversation, translate complex information into accessible language, brainstorm ways to express a need, or reduce barriers associated with disability, anxiety, or executive functioning. In those cases, AI may help the client return to a human relationship more prepared to participate in it.

But AI can also support avoidance. A person may spend hours refining an argument about why a partner is wrong without ever testing that understanding in the relationship and come to conclusions that are wildly inaccurate but sound reasoned when printed back on a screen. They may prefer a system that validates them to a spouse who disagrees. They may outsource uncertainty, conflict, or vulnerability to an interaction that requires no reciprocity, accountability, and keeps the user in a place of perfect comfort, which is often in opposition to what really good therapy does (spoiler alert: it challenges us).

These questions are not about policing AI use but instead are about assessing the impact of this highly influential external entity.

A Shift in the Questions We Ask

We as family therapists cannot write AI software: we can, however, improve our systemic assessment to incorporate AI and more fully understand our clients. Hertlein and Springer (2026) argue that this shift requires couple and family therapists to think beyond AI as simply a tool. AI can become a “third actor” in the therapeutic system. That does not mean a chatbot is a person or that an AI interaction is equivalent to a human relationship. Instead of asking only, “Where did that idea come from?” therapists can ask, “What AI or online tools to think through this?” Therapists can assess the prompts used by the client, ask how the AI responded, what fit, and what happened next. These questions are not about policing AI use but instead are about assessing the impact of this highly influential external entity.

Who Is Making the Decisions?

AI also raises questions about autonomy. Hertlein, Springer, and Rajaei (2026) highlight both individual and collective autonomy among nine systemic ethical principles for AI use in couple and family therapy. Their framework also emphasizes safety, confidentiality with multiple stakeholders, transparency and informed consent, professional responsibility, systemic beneficence, fidelity to the therapeutic relationship, effectiveness, and systemic justice and fairness. For clients, a formative question may be more useful than “Did AI help?” Ask: “Who are you becoming when you use AI this way?” Does the interaction strengthen reflection and agency? Does it help the client communicate more honestly? Or does it increase dependence on external certainty and make difficult human conversations easier to avoid?”

The answer will not be the same for every client. Age, developmental stage, ability, culture, access, digital literacy, and family context matter. Hertlein, Springer, and Rajaei (2026) note that AI safety itself may be developmentally dependent and that therapists may need different conversations with youth and adults. Systemic assessment should therefore remain contextual rather than categorical.

When Therapists Use AI

The third actor is not only a client phenomenon. Therapists use AI for documentation, case conceptualization, psychoeducation, training, supervision, and administrative tasks. These uses may save time and broaden clinical thinking but require appropriate guardrails. If a therapist enters client information into an AI system, confidentiality is no longer a theoretical concern. In relational treatment, information may belong to multiple people in the treatment system. Hertlein, Springer, and Rajaei (2026) emphasize that clinicians need to understand how information is stored, shared, and protected and to be transparent with clients about the role AI plays in their care. The same principle applies to clinical judgment. AI can help generate hypotheses, but the therapist must retain ownership of case formulation and treatment planning. AI can help organize a note, but the therapist owns the accuracy of the record. AI can suggest psychoeducational language, but the therapist owns whether it fits this client, this culture, this relationship, and this moment. Perceived convenience must not displace professional responsibility.

Protecting the Human Competencies

The rise of AI gives our field an opportunity to become clearer about what human therapists uniquely contribute. Our distinctive competencies are relational: we are present, engage in mutual relationship and interactions, display accountability, attend to context, and consider the cases we have seen over the years and how it shapes what we think about the case in front of us

The AI competencies proposed by Hertlein and Springer (2026) are grounded in relational ethics and person-centered care precisely because technical competence alone is insufficient. AI can simulate responsiveness. It cannot participate in therapy with the same lived accountability and mutual vulnerability as a human clinician. That difference should not lead us to dismiss the technology. It should make us more precise about understandings where it helps for both clients and therapists, where it may be harmful, and what about therapy must remain human.

A Simple Place to Start

Therapy has always adapted to the systems people live in. AI is now part of those systems. The sooner we learn to ask how our clients are using it and questioning our own practices, the more effectively we can do the work we have always done: helping people make sense of their lives and relationships. As for me, I have started asking both Robert and his wife to tell me about the AI tools they use before session to obtain a fuller understanding of what information and position he is bringing into the room before we get started. It helps me understand how to ask questions that challenge their assumptions from AI and also shape their prompts to AI to use it to supplement therapy instead of becoming a contrarian to the process.