Public-facing generative-AI legal guidance systems: applied legal-ethics, reliability and governance limits – a multi-case study

Purpose This study aims to examine how public-facing generative-AI legal guidance systems are designed and governed, evaluating three contrasting implementations: DoNotPay, GOV.UK Chat and Insolvency Bot. It analyses the technical and institutional conditions under which such systems can deliver legally accurate, jurisdiction-specific and context-appropriate guidance to non-lawyers while complying with consumer protection and accountability standards. Design/methodology/approach Adopting Yin’s (2009) qualitative multiple-case study methodology, the research draws on regulatory enforcement materials, government documentation, peer-reviewed developer publications and structured author-conducted user simulations. Each case is analysed individually across defined evaluative dimensions, followed by a cross-case synthesis to identify recurring design patterns. Findings The analysis reveals points of convergence and divergence in how public-facing generative-AI legal guidance systems operationalise applied legal-ethics considerations through system architecture and oversight. More reliable systems are characterised by jurisdiction-specific grounding through retrieval-augmented generation (RAG), curated legal corpora, lawyer-in-the-loop validation and structured user education through explicit disclosures and onboarding. The findings demonstrate that accountability in AI-mediated legal services must be embedded in design rather than treated solely as an external regulatory constraint. Originality/value This research provides an empirical cross-case evaluation of deployed public-facing generative-AI legal guidance systems, integrating accountability theory with technical design analysis to inform governance-by-design in AI-supported legal services.