NETHERLANDSResearch
arXiv: Maastricht study finds most legal chatbot conversations seek personal guidance
Maastricht University researchers built a taxonomy of how people seek legal help from chatbots and applied it to 2,122 conversations from the public WildChat-1M dataset. Of the 1,367 conversations with legal content, 71.1% sought personalized or practical guidance, most often advice on the user's own rights and options. Because the sample was enriched for legal content, the authors say it cannot measure how common such requests are.
Key points
- Advice on the user's own rights and options was the main request in 52.1% of legal-content conversations; information requests were 21.4%.arXiv
- Among guidance seekers, immigration and citizenship was the top topic at 14.1%, followed by criminal law at 11.4%.arXiv
- About a quarter of guidance-seeking conversations, 25.3%, involved more than one legal topic.arXiv
- Labels were generated with an OpenAI model and checked against a human-reviewed gold set of 200 conversations.arXiv
- The paper is accepted at JURIX 2026, and its annotations, model checkpoints and code are public.arXiv