6.1 Interpretation of the results
6.1.1 FF1 – Possibilities and Limitations of AI
At the level of the tool, the possibilities of AI are real. As Chapter 5.1 has shown, the mechanism of action does not depend on the origin of the stimulus, as meaning arises in the client’s bodily experience (enactive approach, Chapter 2.3). At the level of symptoms, AI can therefore be effective, and technological progress is more likely to enhance this effect than to limit it.
At the same time, Chapter 5 has also highlighted the limitations of the tool: the effect is fleeting and no longer detectable after three months (Zhong et al., 2024); it remains inferior to traditional psychotherapy (Spytska, 2025); and in a crisis, the trained politeness becomes an acute danger (Chapter 5.3).
The limitation, however, is ontological in nature, and lies not at the level of the tool, but at the level of the encounter. Consequently, the thesis does not become obsolete with the next generation of models, as it is not based on what a system is capable of achieving. AI is not a counterpart, and even the best model cannot change that.
Nevertheless, many people derive genuine benefit from AI: it helps them organise their thoughts, takes the pressure off them, and enables them to make some progress. This is the most common objection to the thesis, and it is a valid one. However, it does not refute it. A tool can be useful without being an interlocutor. The benefit is real. What is missing is the encounter. The Federal Chamber of Psychotherapists (BPtK, 2023, cited in Augustin et al., 2026, p. 2) also emphasises that AI does not replace the fundamental factors of relationship and empathy.
This answers the first research question. As a toolFootnote 1 Here, ‘tool’ refers not to a technical but to a relational concept (Chapter 5.1): it is measured not by how autonomously a system operates, but by the fact that the human being is not faced with another subject. A tool that cannot be controlled is a risk, but it is not, for that reason, a counterpart. , AI is useful: people treat it as an object, an ’it’. But it fails when it comes to the encounter: it cannot be a ’you’. What remains beyond its reach is not an aspect of the relationship, but the relationship itself.
The question remains as to why this ‘you’ is irreplaceable and why it cannot be created by a tool.
6.1.2 FF2 – The Unavailability of Therapeutic Resonance
The answer rests on three points. The first is physicality. Only a living organism must sustain itself, and only when one’s own survival is at stake does anything become meaningful, helpful or threatening. What is meaningful is felt, and only those who feel can empathise. Fuchs sums it up: ‘Only that which can die can also feel fear’ (2026a, pp. 9–10).
An artificial system has nothing at stake. It lacks the bodily involvement from which meaning arises, and where nothing is meaningful, nothing can be empathised with either.
The second point is presence (chap. 3.3). Healing takes place through action in the here and now, in the ’moment of meeting’, not through retrospective interpretation. This temporal structure is bound up with bodily presence, with a moment that happens and cannot be calculated, with an event between two bodily subjects.
The third point is the dialogical ‘in-between’ (Chapter 2.2). The ‘Self’, as understood by Perls, Hefferline and Goodman, is not an isolated core within the individual, but arises through contact, at the boundary of contact between two embodied subjects. Since AI is not an embodied subject, there is no such boundary at the human–AI interface and thus no place where a ‘Self’ could form. The ‘you’ is therefore structurally excluded.
Resonance is not the same as an echo (Rosa, Chapter 2.2). AI reflects back the client’s utterance. It calculates the most likely appropriate response, rather than transforming what has been said through its own bodily experience, as inter-bodily resonance does. Thus, in place of a genuine encounter – in which something new, unexpected and sometimes confrontational is possible – there arises a pseudo-relationship in which the client ‘ultimately only ever confirms themselves’ (Fuchs, 2026a, p. 7).
Whether such a pseudo-relationship is truly insufficient depends on the following question. Is therapeutic empathy only valuable if the therapist, as a vulnerable subject, genuinely shares in the client’s experience, or is it sufficient if the client simply perceives that they are being cared for?
The question confuses the effect with the cause. What AI achieves, at best, is ‘extended empathy’ (chapter 2.3): the cognitive adoption of another’s perspective, which statistically recognises which reaction is appropriate and, as a result, often validates the client with astonishing accuracy. What it lacks is the foundation: primary, bodily empathy – the involuntary resonance between two bodies (Chapters 2.3, 4.1). The more perfectly the machine mimics a human being, the greater the risk of deception. Technical developments do nothing to change this. If, in future, a system records voice, facial expressions and posture, it turns these into measured values, not experience (Fuchs, 2024b; cf. Chapter 4.1). And the implicit level on which a relationship unfolds is not information that can be captured: it takes place between two people (Chapter 4.2).
This answers the second research question. It was: Why does therapeutic resonance (underpinned by physicality, presence and the dialogical ‘in-between’) remain inaccessible and thus irreplaceable? Resonance requires a second physical subject who truly responds, who can break away and reconnect at their own risk. It cannot be produced on demand. It is, by its very nature, inaccessible (Rosa, Chapter 2.2). AI cannot generate it, but can only throw back an echo. And what cannot be produced cannot be replaced by a tool. Therein lies its irreplaceability.
But if corporeality, presence and the dialogical ‘in-between’ cannot be manufactured, then a method that places precisely these dimensions at its centre takes on particular significance in the age of AI.
6.1.3 FF3 – The special value of Gestalt therapy
This method is Gestalt therapy. Its value in the AI age is evident in three respects: as a method, as a form of training and as a counter-model.
As a method, Gestalt therapy is a form of relational medicine in the sense described by Fuchs (Chapter 3). It works with bodily enactment (3.1), with the contact boundary and the field (3.2), and with what is happening in the here and now (3.3); and it is precisely these three aspects that are inaccessible to AI (Chapter 4). Its effectiveness therefore stems precisely from what the machine lacks.
This gives rise to a connection that appears paradoxical at first glance. Anything that can be captured in rules can be automated. A procedure therefore becomes replaceable to the extent that it can be formalised into a manual. Research into effective factors already shows that it is precisely the elements that can be formalised into a manual that do not make a difference: neither adherence to the manual nor technical competence correlates significantly with therapeutic success (Wampold, 2015; see Chapter 2.1). What Gestalt therapy has long been interpreted as a scientific weakness – namely, that its core cannot be forced into categories – thus proves to be the very reason for its irreplaceability.
As a form of training, Gestalt therapy works on both sides. During their training, therapists develop a reflective, physically felt capacity for resonance through awareness and self-resonance in relation to themselves. And in therapy itself, the aim is to foster precisely this ability in clients: to feel their own bodily experience and to resonate with others (Chapter 3.1). In this way, the very skills that machines structurally lack – and which make the encounter possible in the first place – are cultivated.
Finally, as a counter-model, Gestalt therapy responds to a trend that extends beyond the therapy room. Anyone who becomes accustomed to an interlocutor who never contradicts them and makes no demands forgets what a relationship requires: to tolerate contradiction, to reveal oneself in moments of shame, to experience a rupture and to repair it together. The machine’s unchallenged affirmation deprives the client of precisely these experiences (Chapter 5.3) and thus turns confluence into a societal pattern (Chapter 4.2). Gestalt therapy counters this: it demands of the client precisely what the machine spares them. It also employs the concepts that precisely describe what happens at the human–AI interface: what other analyses describe as parasocial attachment is termed ‘confluence’ here; what appears as the ‘animation’ of the machine is termed ‘projection’; and concern for the boundary of contact proves to be the counter-model to the confluence machine. Drawing on Gestalt therapy practice, Kampmann and Schön (2026, p. 100) reach the same conclusion: ‘Attitude cannot be automated.’
This answers the third research question. Gestalt therapy is valuable in the age of AI not because it opposes technology, but because it cultivates what technology cannot produce: bodily resonance, present-moment contact, and the recognition of difference. The more the machine takes over what is accessible, the more important the space becomes in which the inaccessible occurs.