6.3 Outlook
Early psychoanalysis defined psychotherapy as the ’talking cure’: healing takes place through talking. This is precisely where the machine comes in. It will take over everything that consists of words (Harari, 2026; cf. Chapter 5.1), and with it those forms of therapy that place their healing power in talking.
Yet the cure never lay in talking alone. The objection is not directed against speaking; it merely shifts the locus of healing to the embodied, face-to-face encounter that underpins the conversation (Chapter 3.3; Fuchs, 2023). AI can simulate verbal exchange, the appropriate response, the mirroring. Anyone who ties healing to language may therefore regard a talking machine as therapeutic. Anyone who ties it to the embodied encounter realises that the machine lacks precisely that which sustains it. This is what the term ‘Embodied Cure’ stands for: healing through embodied encounter.
What applies to therapy applies to human beings in general. Fuchs traces this insight back to the ‘feeling of being alive’ (2026b), to the experience of inhabiting one’s own body, from which all meaning springs. For him, this leads to a conclusion that extends beyond psychotherapy:
‘Therapy would thus lie in a new, embodied humanism, … which enables us, as living beings, to distinguish ourselves from our machines without having to enter into competition with them […]’ (Fuchs, 2026a, pp. 8–9)
AI is prompting people to re-identify what is specifically human and to consciously nurture it. It is precisely the confrontation with AI that highlights the value of physical interaction. What matters is not whether one believes in or rejects the technology, but whether one distinguishes between the tool and the interaction.
The current social climate makes this distinction urgent. The Artificial Intelligence Index Report 2026 (Stanford HAI, 2026) documents a further acceleration in the spread of AI and its penetration into medicine and mental health applications, alongside growing social concern (cf. Maslej et al., 2025). This intensifies the risks described in Chapter 5.3: AI systems can reinforce delusional thinking and do not replace safe care (Moore et al., 2025). Intensive use is also associated with a decline in personal agency (Kosmyna et al., 2025, preprint).
The contribution this work can make to this situation is limited. It is a theoretical and conceptual foundational study, not an empirical study. It locates the boundary at the point of interaction rather than in the tool itself. Consequently, its findings remain valid regardless of how capable future systems may become. However, it also says nothing about what these systems will be capable of in the future. This limitation affects the two levels of the study differently: the ontological argument (Chapter 4) does not become outdated, whereas the empirical findings and the legal framework (Chapter 5) do become outdated at the pace of technological development. This study will therefore be continued beyond the submitted version as a digital edition, which will be continuously updated with new studies, legal developments and revisions.Footnote 1 The submitted version (as of July 2026) is decisive for the assessment. The updated digital version is available at: https://ki-und-gestalttherapie.netlify.app/
This limitation gives rise to the following follow-up questions, which would require empirical investigation:
Firstly, this concerns the triad (Chapter 5.1). Whether the use of AI as a third element can actually be structured in such a way that it alleviates the burden of the encounter rather than supplanting it is a question of practice, not theory. It would be worth investigating how clients experience the presence of an AI tool in the process and whether they perceive the therapist as more present or more absent.
It also concerns the distinction between echo and resonance (Chapters 2.2, 4.2). So far, this distinction has been grounded in phenomenology. Whether it can also be measured (for example, in the client’s bodily experience, in physiological correlates or over time) remains an open question. The finding that AI effects were no longer detectable after three months (Zhong et al., 2024) points in this direction, but does not explain the reason for this.
A third area is multimodal simulation. Systems incorporating voice, facial expressions and biometric feedback could significantly enhance the illusion of inter-corporeality. This does not affect the thesis of this work, as even a sensor does nothing to alter the lack of a physical boundary. In practice, however, the risk of deception is heightened, and it would be worth investigating whether users’ ability to distinguish between reality and simulation is eroded as a result.
A fourth area concerns the long-term consequences (Chapter 5.3). When people become accustomed to an interlocutor who never contradicts them and makes no demands, it remains unclear what effect this has on their ability to endure real relationships (with their friction, disappointments and the need for reconciliation). It is precisely these ‘repairs’ that lead back to the kintsugi argument (Chapters 2.3, 4.2). So far, this has been justified on phenomenological grounds: AI cannot break, so there is nothing to repair. Whether this becomes apparent in the process would require comparative investigation: for example, by examining how a breakdown in interaction unfolds in the case of a human versus an AI system – whether it is noticed, named and worked through together, or whether it goes unnoticed whilst the surface of the conversation continues smoothly. This would turn one of the central arguments of this thesis into a testable hypothesis.
Finally, Gestalt therapy has its own tool for making its core competencies testable: the Gestalt Therapy Fidelity Scale (Chapter 3). This could be used to investigate how the quality of the process in physical chair work differs from its virtual implementation (Yoo & Kim, 2025; Chapter 5.1) or to what extent an AI system fails to capture the eight key concepts that form the international consensus (Fogarty et al., 2016). The irreplaceability that this thesis theoretically substantiates would thus evolve from a finding into a research task for Gestalt therapy itself.
This brings this work full circle. Therapy requires a body, a field and a present-moment process. Healing occurs where two people subject themselves to what no machine can subject them to: contradiction, shame, rupture and repair. AI plays no part in this process. It is a powerful tool, but an impossible ’you’.
For Gestalt therapy, this presents not a cause for concern, but a call for self-reflection. It need not reinvent itself to survive in the age of AI. It may calmly reflect on its core competencies, on what has always defined it: working with the body, with contact, in the moment. Reflection does not, however, mean standing still. The task also involves further refining its own phenomenological terminology, taking the initiative to demonstrate its own effectiveness, and thus carving out a position within the scientific discourse that corresponds to its clinical substance. The follow-up questions mentioned point the way forward. Self-confidence and self-examination are not mutually exclusive, but rather mutually dependent: only that which stands up to scrutiny can be defended with conviction.
What was long regarded as a scientific weakness, because it cannot be forced into categories, is now proving to be precisely what makes Gestalt therapy irreplaceable:
A person who is concerned with the other, a body that resonates rather than merely reflecting, a shared moment that simply happens and cannot be calculated, a response that can surprise, contradict and transform. None of this can be manufactured, scaled or delegated.
This is the Embodied Cure: healing as a physically present encounter between two people who endure one another. This work is written out of the conviction that, particularly in the age of machines, this counts not less but more.
‘Only if we truly inhabit our bodies will we be able to preserve the Earth as a habitable place’ (Fuchs, 2026a, p. 13).