ABSTRACT
Automated contract generation has been a goal of computing and legal researchers for more than half a century, promising to increase accuracy, reduce delays and improve transactional efficiency across the legal industry. Yet, unlike other legal tasks such as document searching and case analysis, true automated contract generation has never been achieved. Early rule-based systems produced documents modestly tailored to user questionnaire responses, but were highly limited in their scope and range. Today’s GenAI tools can generate plausible short contracts and contractual clauses, often reflecting a user’s preferences, but are plagued by inconsistency, irreproducibility, hallucination, hidden bias, the inability to anticipate future events, and a general lack of transparency.
Surprisingly, over the past fifty years comparatively little scholarly attention has been paid to the degree to which computer technology can best produce coherent and enforceable contractual text, paling in comparison to the study of algorithmic capabilities and deficiencies in other areas of legal practice and decision making. This article, for the first time, deeply analyzes the underlying contextual and linguistic differences between the text of contracts and other private legal documents, such as briefs and memoranda, and finds that these differences present unexpected challenges for the automated generation of contractual text. Adopting a historical approach, it then proposes that GenAI contracting processes be augmented by greater reliance on industry-adopted consensus template agreements, with clear marking of deviations from such templates. It is hoped that this approach will reduce issues arising from the lack of transparency endemic to current GenAI output and engender greater trust in AI-generated contractual text, thereby enabling the efficiency and other benefits that this technology has long promised to deliver.
Contreras, Jorge L, Solving for Agreement (August 12, 2025).
Leave a Reply