The development of digital tools for complex, domain-specific information retrieval presents a significant challenge in the Digital Humanities (DH). Many DH projects excel in data aggregation and technical innovation while falling short with regards to user engagement and accessibility. Tools are designed to fit a certain research objective without full consideration for the end users or their needs. This paper presents a case study in user-centred design (UCD) applied to the development of a semantic search tool for a digital case law collection, arguing that the primary contribution of this work is a replicable UCD methodology for bridging domain-specific language gaps in tools designed for non-expert users.

The UK National Archives’ Find Case Law (FCL) service was developed in response to needs articulated by the Access to Justice movement, which aims to increase open justice measures to ensure that legal processes and decisions in the UK are transparent and accessible (UK MoJ 2023). Services like FCL have advanced this cause by providing free access to court judgments and tribunal decisions made in England and Wales. While there have been many advancements in the area of Legal AI, most tools are marketed to law firms and legal professionals. However, users without a legal background often struggle to locate relevant judgments as legal language frequently diverges from everyday usage (Charrow / Charrow 1979). The tools presented in this paper aim to bridge this so-called language gap between legal and “everyday” English, and to make case law more accessible to lay users. Specifically, I present the user research (UR) undertaken between the first and second iterations of a semantic search prototype, and how data collected from usability tests and user interviews shaped the development of the second prototype.

The targeted user-base for this project is primarily non-legal professionals, that is to say people who actively search for case law news or judgments but are not directly involved in a legal profession. The first step in conducting user research was to gain a better understanding of the various personas that fall into this target audience (Miaskiewicz / Kozar 2011). I identified three relevant personas: (1) users seeking legal information in relation to news articles or current events; (2) users searching for cases relevant to their own legal circumstances (potentially to bring a case to court themselves); and (3) non-legal professionals with some engagement with the law, such as journalists, scholars, and advice sector workers who research case law for professional reasons.

Based on these personas, usability testing was undertaken on the first prototype, referred to here as the Alpha prototype. Briefly, the Alpha is a search suggestion tool aimed at helping users refine their initial query and guiding them towards more accurate legal terminology (Wilson 2025). Drawing on the distributional hypothesis — the idea that words derive meaning from the contexts in which they appear (Firth, 1957) — lexical meaning is represented using word embeddings encoded as mathematical vectors. This allowed for word usage to be compared between legal and non-legal contexts. For the legal context, I used Word2Vec embeddings (Mikolov et al., 2013) trained on the FCL corpus and, for the non-legal context, the British National Corpus (BNC, 2007).

Figure 1. Screenshot of prototype Alpha.

To evaluate the Alpha prototype’s usability and inform the next development phase, I conducted a round of generative and evaluative user interviews with seven participants representing all three personas (two from persona group 1, two from persona group 2, and three from persona group 3, including a journalist, a legal scholar, and a worker in the advice sector). Interviews employed a think-aloud protocol during scenario-based search tasks, encouraging participants to verbalise their thoughts and actions while navigating the tool (Baxter et al. 2015). Tasks ranged from specific case searches to broader scenarios in which participants were encouraged to follow their own search path.

The user research revealed three areas in which the Alpha prototype failed to meet user expectations, directly informing the redesign of the Beta prototype. First, it revealed a disconnect between lexical and semantic search suggestions: despite the prototype’s technical accuracy, users consistently struggled with the legal terminology present in search results and case summaries. The think-aloud protocol highlighted a significant lexical gap, where users could not map their everyday language to the necessary legal concepts, a difficulty compounded by a lack of clarity on the structure and meaning of legal judgments. Second, the post-task questions revealed pervasive concern regarding the validity and provenance of AI-generated content, with users expressing a desire for clear demarcations between original case text and AI-generated summaries. Third, specific interface usability issues emerged, including confusion with the initial keyword selection process and a preference for natural language queries over keyword selection.

These findings drove a set of targeted design changes in the Beta prototype. Most significantly, the Beta introduces a system of plain-English legal concept tags, derived from guidance articles from Citizens Advice, which are applied automatically to case results. These tags are designed primarily to support persona groups 1 and 2: users with little or no familiarity with the legal system who cannot reliably identify relevant cases from legal terminology alone. By providing plain-English labels for the legal concepts present in a judgment (e.g., “parental responsibility” or “domestic abuse”) the tags aim to create an accessible intermediary layer between the user’s everyday language and the technical vocabulary of case law. Each tag links directly to the relevant Citizens Advice page, providing further context and supporting the access to justice goals of the project. In addition, the Beta shifts from keyword-based to natural language search. A second round of user interviews with the same seven participants (where possible) is planned to evaluate whether these changes have addressed the issues identified in the first round and what further work is required.

Figure 2. Screenshot of prototype Beta.

The transition from the Alpha to the Beta prototype of this semantic search system provides a replicable model for DH projects seeking to maximise user engagement in complex, specialised domains. By treating UCD not as an afterthought but as the core driver of development, I demonstrate how empirical user research can transform a technically sound prototype into a trustworthy, accessible, and usable tool. The case study is particularly instructive for projects at the intersection of computational methods and public-facing access, where the gap between technical capability and genuine user utility is most acute.