Embodiment, Context, and the Limits of AI in Therapy

There is an increasingly common assumption emerging within discussions of artificial intelligence and mental healthcare that if a system can produce coherent, empathic, and psychologically plausible language, it may be approximating something akin to psychotherapy. This assumption deserves careful examination. While large language models can generate remarkably sophisticated linguistic responses, psychotherapy is not reducible to language exchange alone.

Human communication, and particularly therapeutic communication, unfolds within a far richer contextual ecology that extends well beyond words, explicit cognition, or symbolic meaning.

A substantial proportion of human communication occurs outside direct verbal content. While popular claims about exact percentages of verbal versus non-verbal communication are often overstated or misapplied, there is broad agreement across communication science, psychotherapy research, affective neuroscience, developmental psychology, and embodied cognition that meaning is conveyed through multiple interacting channels. Tone, prosody, pacing, pauses, rhythm, hesitations, speech inflections, vocal intensity, facial expression, eye movement, bodily posture, gesture, autonomic shifts, muscular tension, orientation toward or away from another, and patterns of interpersonal synchrony all contribute meaningfully to human communication. Communication is not simply something spoken, it is enacted.

In psychotherapy, this becomes even more consequential. A client may verbally state that they are coping well, yet their voice may flatten, their posture collapses, their gaze drifts, their breathing shortens, or their emotional tone becomes constricted. The therapist is not merely interpreting a sentence. They are participating in a multi-layered contextual encounter in which language is only one informational stream among many. Meaning emerges not from semantic content alone, but from the relationship between words, body, affect, timing, interpersonal history, and situational context. This is part of what makes therapy fundamentally embodied.

Human beings do not communicate as detached information processors, exchanging abstract propositions. We are embodied organisms whose cognition, emotion, and relational processes emerge through lived interaction with environments, histories, and other people. Philosophical traditions such as phenomenology have long emphasised this. Maurice Merleau-Ponty argued that the body is not merely an object we possess, but the primary site through which we encounter and make sense of the world. Cognition is not disembodied computation occurring in abstraction from lived experience. It is grounded in perception, movement, temporality, and embodied presence.

This perspective aligns with later work in embodied cognition and enactivism, where cognition is increasingly understood as arising through dynamic interaction between organism and environment rather than purely symbolic internal processing. Thinkers such as Francisco Varela, Evan Thompson, and Eleanor Rosch advanced the notion that cognition is enacted through embodied participation in the world, not merely computed internally. This has direct implications for psychotherapy, where understanding often emerges relationally rather than through detached analysis.

Therapy also contains an implicit dimension that is often difficult to fully articulate in language. Much of what occurs therapeutically exists beneath or alongside explicit verbal narrative. There is atmosphere, felt tension, rupture, safety, pacing, hesitation, emotional resonance, discomfort, synchrony, silence, withdrawal, openness, and co-regulation. Therapists often respond not only to what is said, but to what is sensed, inferred, embodied, and relationally unfolding in the moment.

Eugene Gendlin’s work is particularly relevant here. His concept of the felt sense points toward a pre-conceptual, embodied knowing that often precedes formal language. People frequently know something in their bodies before they can articulate it verbally. Therapy can involve helping clients contact this unclear but experientially real implicit knowing and gradually bring it into awareness. This is not simply interpretation of linguistic content. It is engagement with embodied phenomenology.

Developmental psychology reinforces this understanding. Human communication begins long before language acquisition. Infants communicate through gaze, movement, vocal tone, affective synchrony, contingent responsiveness, and bodily regulation. Allan Schore, Daniel Stern, Colwyn Trevarthen, and others have extensively described how relational development is built upon pre-verbal attunement systems. Emotional communication and regulation emerge first through embodied interaction, with language developing later within this scaffold. Psychotherapy, in many ways, recruits these same deeply human relational processes This creates an important limitation for AI systems, particularly large language models.

Language models process symbolic representations and generate statistically plausible responses based on learned linguistic relationships. Even when multimodal systems incorporate voice or visual cues, they do not inhabit embodied human experience in the sense relevant to therapy. They do not possess interoception, autonomic regulation, phenomenological subjectivity, or lived relational embodiment. They do not experience silence as tension. They do not sense the subtle discomfort in a client’s shifting posture or the meaning contained in a slight pause before answering a difficult question. This is often what clinicians mean when they say AI misses context.

Context is not merely missing background information that could be supplied in a better prompt. Nor is it simply a matter of insufficient data. Human communication occurs within nested contextual layers: physical space, temporal sequencing, relational history, developmental experience, embodied states, emotional processes, social contingencies, cultural meaning systems, and moment-to-moment non-verbal exchanges. Communication unfolds within a substrate of context, not outside it.

A client’s words in therapy cannot be separated from where they are said, how they are said, what preceded them, what bodily signals accompany them, what silence surrounds them, and what relational meaning has already been established between therapist and client. Meaning is contextual, dynamic, and embodied.

Large language models may produce remarkably convincing therapeutic dialogue. They may simulate empathy, validation, curiosity, and reflective listening with striking sophistication. Yet simulation of therapeutic language is not equivalent to participation in therapeutic process. A convincing approximation of empathic response should not be confused with embodied relational understanding.

This does not mean AI has no role in mental healthcare. It may be highly useful in psychoeducation, behavioural prompting, structured intervention support, reflective journaling, symptom monitoring, between-session reinforcement, administrative augmentation, or widening access to psychological resources. These are potentially valuable uses. But psychotherapy itself is not simply the correct arrangement of helpful sentences.

Psychotherapy is relational, contextual, embodied, and phenomenological. Reducing therapy to language exchange risks misunderstanding both therapy and human communication itself. The relevant question is not whether AI can produce convincing therapeutic words, it clearly can. The deeper question is whether psychotherapy is fundamentally a linguistic process. The evidence suggests it is not.

References

Fuchs, T. (2019). The interactive phenomenal field and the life space: A sketch of an ecological concept of psychotherapy. Psychopathology, 52(2), 67–74. https://doi.org/10.1159/000502098

Gallese, V., Eagle, M. N., & Migone, P. (2007). Intentional attunement: Mirror neurons and the neural underpinnings of interpersonal relations. Journal of the American Psychoanalytic Association, 55(1), 131–176. https://doi.org/10.1177/00030651070550010601

Gendlin, E. T. (1996). Focusing-oriented psychotherapy: A manual of the experiential method. Guilford Press.

Koole, S. L., Tschacher, W., & Fuchs, T. (2020). The embodied self in psychotherapy: Perspectives from phenomenology, neuroscience, and psychology. Journal of Psychotherapy Integration, 30(1), 1–16. https://doi.org/10.1037/int0000148

Merleau-Ponty, M. (1962). Phenomenology of perception. Routledge.

Norcross, J. C., & Lambert, M. J. (2018). Psychotherapy relationships that work III. Psychotherapy, 55(4), 303–315. https://doi.org/10.1037/pst0000193

Ramseyer, F., & Tschacher, W. (2011). Nonverbal synchrony in psychotherapy: Coordinated body movement reflects relationship quality and outcome. Journal of Consulting and Clinical Psychology, 79(3), 284–295. https://doi.org/10.1037/a0023419

Schore, A. N. (2001). Effects of a secure attachment relationship on right brain development, affect regulation, and infant mental health. Infant Mental Health Journal, 22(1–2), 7–66. https://doi.org/10.1002/1097-0355(200101/04)22:1<7::AID-IMHJ2>3.0.CO;2-N

Stern, D. N. et al. (1998). Non-interpretive mechanisms in psychoanalytic therapy: The “something more” than interpretation. International Journal of Psychoanalysis, 79, 903–921.

Trevarthen, C. (1979). Communication and cooperation in early infancy: A description of primary intersubjectivity. Behavioral and Brain Sciences, 2(4), 321–347. https://doi.org/10.1017/S0140525X00063070

Varela, F. J., Thompson, E., & Rosch, E. (1991). The embodied mind: Cognitive science and human experience. MIT Press.