Assistive Technology in Dementia Care: Prevent, Detect, Comfort, or Substitute? A Critical Narrative Review of the Evidence, the Alternatives, and the Regulatory Gap

Background. Chatbots and companion robots for loneliness, sensors and cameras for safety, and older low-technology aids such as handrails are being marketed and deployed into dementia care at scale, often faster than the evidence that they help, and to a population among the least able to evaluate, refuse, or contest them. Approach. A critical narrative review reported against the Scale for the Assessment of Narrative Review Articles. We build a study-level evidence map that grades each technology by population, setting, study design, outcome, and effect size on a single labelled statistical basis (controlled between-group versus uncontrolled within-group), and classify each by a four-way frame: does it prevent the harm, detect it, offer comfort without changing the outcomes that matter, or substitute for human care? Benefit and harm are held to the same evidential standard. Key findings. The interventions with high-certainty evidence of preventing harm are almost all low-technology: home modification and exercise for falls, cognitive stimulation therapy and music for psychosocial symptoms. Sensing and surveillance technologies mostly detect rather than prevent. Two large randomised trials and a Cochrane review find bed and chair alarms do not reduce falls; wearable detectors identify a fall well but have no evidence of preventing one; and a government trial of artificial-intelligence cameras was independently judged not to have reached an accuracy acceptable to staff. Companion robots improve engagement, and reduce agitation against usual care, but in the only trial with a sham comparator an identical switched-off plush toy produced the same agitation benefit (p = .68), so the effect does not require the robotics. The causal chatbot evidence comes from healthy adults around 40 years old, and no peer-reviewed efficacy evaluation exists in dementia. Implications. A decision frame follows for buyers, clinicians, and families: fund the evidenced low-technology first, treat companions and monitors as unproven adjuncts, and recognise that in Australia a device that manages behaviour or restricts movement can meet the definition of a restrictive practice. Canonical web version: lightage.ai/assistive-technology-in-dementia-care. How this was made: produced AI-driven and human-directed. A human set the topic, the standard of proof, the angle and the audience, and is accountable for what is published; AI assisted the evidence-gathering, drafting and verification. Independent analysis, not peer reviewed.