6.2 Recommendations for Action

6.2.1 AI as a tool within a relational framework

The triad described in Chapter 5.1 provides the framework: AI tool, therapist, client. The therapist remains the ‘you’; the encounter takes place between people; the AI is the instrument. In practical terms, this means that AI takes on tasks that can be delegated (documentation, processing, structuring) and frees up the therapist for what only they can do: physical presence, primary empathy, and the ability to tolerate uncertainty. The aim is not to simulate humanity as convincingly as possible, but to restore space for the encounter. However, such a reduction in workload does not happen of its own accord. Without conscious application , the burden simply shifts to monitoring the tools (see Chapter 5.3).

It is not only administrative tasks that can be delegated. As a creative medium, AI can make inner images, dreams or feelings visible, which are then processed through shared contemplation (Chapter 5.1). Here, too, the medium remains at the service of the encounter; it is not itself the counterpart.

6.2.2 Distinguishing according to the therapeutic goal

Not every therapeutic goal requires the same approach. When the aim is to reduce symptoms, AI tools may suffice; they are most effective where they remain embedded within human care (Chapter 5.1). The triggering mechanism works even without a human counterpart. When, on the other hand, the aim is relational healing – change at the level of implicit self-processes – the ‘you’ is irreplaceable. The distinction therefore lies not between good and bad use of AI, but between two types of therapeutic work.

6.2.3 Duties of care: transparency and data

Two duties of care arise directly from the above. The first concerns deception: AI is a tool, and its use must not feign a ‘you’ . Where it is used in a therapeutic context, this use must be disclosed to clients as well as in the documentation. The second concerns data. Anyone who has session content processed via cloud-based AI risks not only a data protection breach but also, potentially, a breach of the duty of confidentiality (Chapter 5.3). In practical terms, this means: data minimisation; wherever possible, locally operated models that do not send data to third parties; and informing clients about the risks involved if they use chatbots independently. Both obligations are underpinned by the same principle: responsibility cannot be delegated to an ‘it’ (Chapter 5.1). This includes ensuring that human interaction does not become a premium service (see Chapter 5.3), but remains available to clients in need as standard care in the future.

6.2.4 Addressing clients’ use of AI

More and more people are seeking advice from chatbots (Chapter 5.2). Independent use of these tools should therefore be addressed in the therapeutic conversation: it should be raised, not left to happen by itself. The information provided should not be limited to data risks (Chapter 6.2.3), but should also cover clinical limitations: the incorrect handling of crises, a lack of ability to pause and bring sessions to a close, and the avoidance of shame (Chapter 5.3). If a bond with a system already exists, it should not be dismissed, but rather addressed for what it is: a projection (Chapter 4.2). Gestalt therapy work with projections provides the tools for this. The aim is to restore the ‘as-if’ awareness, which is naturally preserved at the empty chair (Chapters 4.2, 5.1): to know that one is projecting.

6.2.5 Placing resonance competence at the heart of training

Augustin et al. (2026) show that psychotherapists’ AI competence has so far been only moderately developed and that the ability to actively shape the process is practically non-existent. The obvious call is to retrofit technical knowledge. Whilst this is justified, it falls short on its own. Anyone who realises that these systems are trained to elicit agreement (Chapter 4.2) recognises their confluence for what it is: a training arte , not a counterpart. Above all, what is needed is the embodied knowledge of what a tool fundamentally cannot be.

The concept of ’aesthetic relational knowledge’ (Spagnuolo Lobb et al., 2022; Spagnuolo Lobb et al., 2023; cf. Chapter 3.1) demonstrates that this competence can be described and taught. Bodily self-awareness and resonance can be systematically practised and are therefore teachable. However, they can only be taught through lived practice.

As shown in Chapter 3.1, this competence matures over the course of a long professional career and, much like the lost experiential knowledge of organisations (see Chapter 5.3), it cannot be acquired retrospectively, but arises through self-experience and supervision. Awareness and self-resonance work therefore do not belong on the periphery of training, but at its very centre.

6.2.6 Confidently championing one’s own core competencies

Gestalt therapy should confidently champion its bodily-relational strengths, rather than conforming to manualised, symptom-centred evidence of efficacy. Shifting it towards the manualised pole only makes it more replaceable (see Chapter 6.1.3). Self-confidence does not, however, mean foregoing research: the fact that the core competencies of Gestalt therapy can be described and tested is demonstrated by the international consensus and the Gestalt Therapy Fidelity Scale, as well as by the competency study from the integrative training context (Fogarty et al., 2016, 2020; Grillmeier-Rehder, 2020; see Chapter 3), a body of research that defines Gestalt therapy through its relational-embodied approach rather than through technique. Fuchs and Schmidsberger (2024) are working in the same direction: their proposal to establish phenomenology as a common foundation for the humanistic cluster is a step towards an embodied paradigm. Gestalt therapy thus actively contributes to the further development of psychotherapy.

6.2.7 Taking a stance on professional ethics

The legal situation will remain unregulated for years to come: the key provisions of the AI Act will not come into force until the end of 2027 at the earliest, whilst the PThG 2024 makes no mention of AI (Chapter 5.3). As long as the legislator has not regulated ’ ’, the professional associations must not leave this gap to be filled by individual therapists, onto whom the responsibility is currently, in effect, being shifted (Chapter 5.3). Professional ethical guidelines are needed for the use of AI tools: transparency towards clients, handling of sensitive data, and limits on their use in direct therapeutic communication, as already established by law in Illinois (Chapter 5.3). The DVG conference demonstrates that the Gestalt therapy community is already engaged in this debate (Bauer, 2026; cf. Chapter 1.2). This debate must be continued and translated into binding standards.