INTRODUCTION: Artificial intelligence-supported personalisation is accelerating in mental health services, yet it can reclassify relationship, biography, and judgement as optional. AIM/QUESTION: To explain how "personalisation" drifts into substitution and to specify minimum conditions for safe, workable implementation in mental health nursing. METHOD: A debate essay using purposive critical synthesis of empirical and conceptual literature to organise three recurring tensions and derive auditable safeguards. RESULTS: Three tensions recur across current deployments: alliance-like chatbot support can be mistaken for therapeutic presence; automation can narrow clinical reasoning and contribute to deskilling; and trace-based personalisation can override care biography and erode trust through extractive data practices. DISCUSSION: The common failure mechanism is optimisation drift under weak governance, where simulated alliance replaces therapeutic presence, metric authority displaces professional judgement, and proxy traces override care biography. IMPLICATIONS FOR PRACTICE: Service governance and procurement should treat therapeutic presence, narrative authority, and moral agency protection as non-negotiable clinical outcomes at the point of care. RECOMMENDATIONS: Implement relationally accountable augmentation with explicit red lines, escalation pathways, protected relational time, biography-first documentation, and evaluation that tests accountability and safety under routine conditions.
Erman Yıldız (2026) studied this question.