Legal Experts Inform Interface Design For Skimming And Writing Tools In 22-Professional Study

Legal professionals routinely navigate dense and intricate documentation, but a comprehensive understanding of their reading and writing processes, and their expectations for digital tools, has remained elusive. Chelse Swoopes from Harvard University, Ziwei Gu, and Elena L. Glassman from Harvard University, addressed this gap by directly investigating how legal experts approach these tasks. The team interviewed twenty-two professionals, presenting them with prototypes designed to highlight connections between documents and to allow users to confidently manage potential errors from artificial intelligence systems. This research reveals critical limitations in current workflows, provides specific feedback on designing robust and reliable legal technology, and ultimately offers actionable guidance for both tool developers and legal professionals seeking to improve efficiency and accuracy in their work.

Researchers interviewed 22 legal professionals to investigate workflows, challenges, and technology use. Each session incorporated prototypes designed to reveal relationships across documents and support interaction that anticipates and recovers from errors in AI systems, enabling users to identify, assess, and correct unexpected behaviour. Participants then evaluated how these tools could integrate into their work with different document types and at various stages of a project. The analysis details limitations and challenges within current legal workflows, alongside feedback on AI-resilient interfaces, and expert insights relevant to legal technology design. These findings offer actionable guidance for technology designers.

AI Tools for Legal Information Access

This document details research conducted through interviews with legal professionals regarding their potential use of three AI-powered tools: a skimming and summarization tool, an abstract-based document navigation tool, and a contextual retrieval system. The skimming and summarization tool proved valuable for processing lengthy or formulaic documents such as contracts, complaints, deeds, and regulatory guidelines, allowing professionals to quickly extract key information. It also aided in efficiently identifying main takeaways from formal or academic reading materials like research articles and judicial opinions, and streamlining practice-facing communications. However, participants expressed concerns about the risk of losing nuance, particularly with contracts and statutes where precise wording is essential, and acknowledged an ethical obligation to read sources fully when citing them.

The abstract-based document navigation tool showed promise in navigating large or poorly organized documents, particularly contracts and clause libraries, helping professionals quickly understand their content and identify patterns. It also leveraged the similarity to scientific abstracts in legal journals for academic and scholarly legal writing. However, its effectiveness diminished with unstructured texts like memos or legal reviews with inconsistent writing styles. The contextual retrieval system proved particularly useful for contracts due to their length, repetitive structure, and template-like conventions.

It also supported narrative and court-oriented drafting, such as statements of claim and legal opinions, especially where document structure is conventional. Internal documents like firm playbooks and negotiation parameters could benefit from its use to maintain consistency, and it offered educational benefits for student legal memos and exams, enabling comparative review. Participants emphasized the importance of preserving intra-document context when retrieving smaller sections from contracts. Several overarching themes emerged from the research. The suitability of each tool heavily depends on the type of legal document being used, with highly structured documents proving more amenable than unstructured ones.

Legal professionals are concerned about the risk of losing nuance or making errors due to AI-powered summarization or retrieval, prioritizing precision in many legal contexts. Maintaining context within documents is essential, particularly when retrieving smaller sections from larger corpora. The tools could be valuable for various legal roles, including litigation, transactional law, academic research, and education, offering the potential to significantly improve efficiency by automating tasks such as skimming, summarizing, and retrieving information. However, legal professionals emphasize the need for careful implementation and validation of these tools to ensure accuracy and avoid errors. In conclusion, the research suggests that these AI-powered tools have the potential to be valuable assets for legal professionals, but their successful implementation requires careful consideration of document type, risk tolerance, and the need for maintaining context and accuracy.

Legal Reading Practices and Workflow Challenges

This research delivers detailed insights into the reading and writing practices of legal professionals, revealing the challenges they encounter and their perspectives on potential tools. Interviews with 22 legal experts explored workflows and tool usability, providing a foundation for improved design in legal technology. The study identified a diverse range of documents commonly read and written, spanning case law, statutes, contracts, and internal communications, demonstrating the breadth of legal workflows across various roles. Participants consistently described reading as laborious, citing both document volume and density as persistent challenges, with many engaging with documents containing hundreds or even thousands of pages.

They experienced mental fatigue when parsing long linear text, and reading was particularly time-consuming when searching for small snippets of information buried within lengthy documents. Participants highlighted the difficulty of extracting relevant information, noting that the challenge often lay in sifting through abundant unnecessary information rather than decoding complex text. Understanding specialized terminology and jargon also presented a hurdle, with some pointing to inherent ambiguity and intentional obfuscation within legal language. Structural complexity further compounded reading difficulties, as statutes and regulations often follow nested, code-like structures rather than standard expository formats.

Cross-referencing between documents or laws also presented a significant challenge, with meaning frequently buried in obscure references or dependencies. Participants consistently expressed concern over the risks of skimming, noting that it could lead to serious mistakes by missing hidden clauses or relying on assumptions. The study demonstrates a clear need for tools that address these specific challenges and support more efficient and accurate legal reading and writing.

Legal Workflows, Privacy, and Interface Design

This research details valuable insights into how legal professionals approach complex documents and their expectations for technology designed to support their work. Through interviews with practitioners, the study identifies key challenges in legal workflows and provides domain-specific feedback on the design of AI-resilient interfaces. Participants emphasized the importance of balancing efficiency with maintaining strict confidentiality and accuracy, particularly given the high stakes often involved in legal texts where even minor details can have significant consequences. The findings offer actionable guidance for designers developing reading and writing support tools for legal professionals, highlighting the need for interfaces that are both time-saving and uphold privacy standards. The researchers note that the patterns observed in the legal domain may be transferable to other high-stakes fields such as medicine, aviation, and finance, suggesting potential for broader improvements in AI-assisted support tools across these sectors. Further research could explore the application of these insights in these additional domains to determine the extent of their applicability and refine design principles for critical applications.