When help comes from a screen: What happens when we trust digital conversational partners?

## Introduction In the span of a few years, digital tools have gone from curiosity to part of many people’s everyday lives. Chatbots offering emotional support, automated advice and conversational therapy based on advanced language models promise scalable, accessible help when human therapists are not available. At the same time, the debate over safety, ethics and effectiveness has intensified. This is not just a technological discussion — it is about whom we trust when we are vulnerable, how we protect privacy, and which social structures change when therapy can be delivered at scale by software. In this article I take an opinionated look at the development: I review what research so far says about the effects of generative, automated therapy, highlight ethical and legal challenges, and place the development in the context of recent findings on how early upbringing shapes the ability to cope with life’s demands. According to a VG article from April 2025, there is reason to view more holistically how both parenting styles and technological "helpers" influence mental health. As I will argue below, we must avoid both technological naivety and premature panic — but the demands for regulation and quality assurance are clear. ## Where does the research on digital conversational therapists stand today? Research on digital therapeutic tools, including chatbots and apps that use cognitive behavioral principles or more advanced generative technology, is growing but still heterogeneous. Several randomized controlled trials (RCTs) have shown that structured digital interventions can reduce symptoms of mild to moderate depression and anxiety, especially when there is a form of guidance or human support alongside. Effects are often moderate; meta-analyses typically point to small to moderate effect sizes (often around d ≈ 0.3–0.5 for guided digital CBT interventions), and the effect of purely automated, unsupervised systems is usually weaker. Examples from the field show both potential and limitations. Commercial and clinical use of tools like Woebot, Wysa and others has been examined in several studies reporting improvements in self-reported symptoms and high user engagement in the short to medium term. At the same time, researchers note that long-term data, robust comparisons with traditional psychotherapy, and studies in vulnerable groups (e.g., people with severe mental illness, suicidal patients, or people with comorbidity) are still lacking or sparse. When tools begin to use large language models for free-text conversation, complexity increases: generative models can be highly persuasive, but they can also hallucinate facts, give inappropriate advice or misjudge risk. Researchers are also clear that context matters. Digital tools work best as part of a larger treatment ecosystem — for some they can be a valuable low-threshold offering or supplement; for others they may be insufficient or even risky if they replace necessary human assessment. This ambivalent picture should guide both clinical implementation and policy-making. ## Ethics and safety: The debate that won’t quiet down When we discuss digital therapists, we must take ethical and safety-related issues seriously. Three main areas recur in the debate: privacy and data security, responsibility and accountability for errors or harm, and quality assurance of the generated content. Privacy: Therapeutic conversations contain highly sensitive information. When such data are collected and processed by commercial actors or stored in the cloud, questions arise about who owns the data, how they can be anonymized, and which third parties may gain access. Breaches of confidentiality or poorly secured data stores can have serious consequences for users. Accountability: Who is responsible if a chatbot gives harmful advice or misdiagnoses a serious condition? In traditional healthcare practice, responsible parties and legal frameworks can point to practitioners or institutions. When algorithms make decisions or suggest actions, lines of responsibility become murkier. This is especially the case with generative AI that does not necessarily follow predefined rules and that can produce unexpected responses. Quality and safety: Generative models can produce believable but factually incorrect advice — what researchers call hallucinations. In a psychotherapeutic context this can lead to incorrect risk assessments (for example around suicide), departures from established treatment approaches, or suggestions that are not tailored to the user's clinical needs. Additionally, there is a danger that users develop excessive trust in an algorithm that cannot provide human empathy or clinical insight into ethics and complex diagnosis. The debate about regulation is therefore intense. Some professional communities demand clear labeling of which model is used, requirements for clinical validation before use in health services, and mechanisms for escalation in cases of serious risk. At the same time, there are important questions about access: strict regulation may slow the scale-up of useful low-threshold services. The solution likely lies in differentiated requirements — higher standards for tools marketed as diagnostic or that replace clinical judgment, and lower thresholds for support tools with clearly limited functions. ## When parenting style meets digital therapists The discussion about digital therapists gains an extra dimension when we consider how early care and parenting style shape mental health over time. According to the VG article "Experts: The parenting style that makes children handle life worse" by Anders Ihle Tovan (published 12.04.25, updated 13.04.25), a parenting style many consider normal can in reality harm children in deeper ways. As the article points out: "A parenting style some see as completely normal can in reality harm children in deeper ways than one might think. An expert points to two particularly worrying features." The article links these traits to challenges like weakened coping ability and increased vulnerability to mental health problems. This insight is relevant when we think about who uses digital therapists and why. Children and young people who grow up in overprotective or performance-oriented environments can develop low tolerance for discomfort and a strong preference for immediate consolation or solutions. Digital therapists offering quick emotional relief or simple tools can become attractive; they may solve an acute need, but they do not always address the deeper skills for emotion regulation and coping. In other words, technology can reinforce short-term coping strategies rather than build resilience — unless services are designed with this in mind. Furthermore, the combination of early vulnerability and easily accessible digital support creates a risk that some groups will receive only automated solutions instead of qualified human help. For example, young people with ADHD, who often experience emotional regulation challenges and barriers to accessing care, may end up in a parallel "digital-solution" that does not capture complex needs or comorbidity. This is precisely the kind of system failure we must avoid: technological solutions that appear helpful but in reality confirm and cement underlying problems. ## Practical consequences and recommendations for implementation What should clinicians, decision-makers and parents do now? Here are concrete recommendations, based on current research limitations and the ethics debate. 1) Use digital therapists as a supplement, not a replacement. For many, chatbots and apps are a useful low-threshold offering that can provide short-term symptom relief or support between consultations. Prioritize hybrid models where human assessment is always available when needed. 2) Clear labeling and informed consent. Users must receive clear information about what the tool can and cannot do, about data handling, and about contact options in case of crisis. This is especially important for adolescents and groups with lower health literacy. 3) Requirements for clinical validation and ongoing quality control. Apps and chatbots that aim to give therapeutic advice should undergo independent studies, have documentation of effectiveness and safety, and be subject to continuous monitoring for unexpected negative consequences. 4) Implement escalation mechanisms. The system must be able to identify warning signs (for example suicidal thoughts) and have automatic, safe routes to connect the user with human professionals. 5) Protect data. Strong encryption, minimal data retention practices and clear rules for third-party sharing must be standard. No commercial use of health data should occur without explicit and informed consent. 6) Invest in digital health literacy. Both clinicians and parents need training in what such tools can do, how they should be used, and when to refer to traditional treatment. 7) Pay special attention to young users. Parents and schools must learn to distinguish between useful support and replacement for human care. According to the VG article, there are two particularly worrying features of certain parenting styles that can make children less resilient; technology cannot compensate for lack of coping training and emotional support from close caregivers. ## A responsibility for technologists, healthcare and parents Technologists developing digital therapists must shoulder responsibility in design choices. This means building in safety mechanisms, ensuring transparent model behavior and collaborating with clinical experts in development and validation phases. At the same time, healthcare systems must take ownership of how such services are integrated into the care chain — including setting clear criteria for what can be accepted as clinical support and what requires human treatment. Parents and caregivers also have responsibility. As the VG article notes, parenting style itself can lead to reduced coping in children. When digital aids become part of daily life, adults must still provide boundaries, model coping strategies and seek qualified professional support when a child’s needs exceed what apps can handle. Ensuring fair access is also a moral obligation. Digital therapy promises to democratize access to help, but without proper regulation it can create a divide: resource-rich users get tailored, validated care while vulnerable groups are offered automated solutions without sufficient follow-up. This is a development we must counter through policy, subsidizing quality services and demanding inclusive design. ## Conclusion Digital therapists and generative conversational technology represent a powerful tool that can increase access to mental health help. Research shows potential for symptom reduction, especially in low-threshold or supplementary roles, but it also documents clear limitations — particularly regarding severe mental illness, long-term outcomes and risk management. Ethical and safety challenges require that we as a society demand transparency, data protection, accountability and clinical validation. At the same time, findings about the importance of parenting style remind us of a crucial point: technology cannot replace the need for human care and the development of resilience. As the VG article reminds us, "A parenting style some see as completely normal can in reality harm children in deeper ways than one might think." This statement should serve as a warning: in the pursuit of quick technological fixes we must not forget the fundamental social and relational factors that shape mental health. My message is simple and twofold: embrace the potential, but do so wisely. Build regulations, quality standards and routines that ensure human oversight. Invest in prevention and caregiving in childhood so that digital tools become the supplements that reinforce human help — not substitutes that cement vulnerability. ## Sources - According to the VG article "Experts: The parenting style that makes children handle life worse" by Anders Ihle Tovan (published 12.04.25, updated 13.04.25): "A parenting style some see as completely normal can in reality harm children in deeper ways than one might think. An expert points to two particularly worrying features." (VG, 2025) - Professional analyses and meta-analyses of digital psychotherapeutic interventions (compiled evidence from RCTs and systematic reviews up to 2024).