GENERATIVE ARTIFICIAL INTELLIGENCE IN LEGAL RESEARCH AND PARAMETERS FOR ESTABLISHING ETHICAL AND PROCEDURAL CRITERIA

INTRODUCTION: The introduction of Generative Artificial Intelligence into legal research and practice faces significant resistance, motivated by concerns over algorithmic hallucinations, the proliferation of plagiarism, and the consequent loss of critical skills among researchers. However, this unrestricted rejection ignores the reality that legal research already suffers from the phenomenon of the semantic parrot, characterized by the mere uncritical repetition of doctrines without effective factual innovation. Given this problem and the imminent digital transformation of academic routines, there is a justified need to investigate how this technology can be used ethically and methodologically appropriately within the legal field. AIMS: The central problem of this study is to define how generative artificial intelligence can be validly incorporated into the investigative stages, seeking to establish procedural parameters that act as a safe bridge between technological innovation and non-negotiable scientific integrity. MATERIALS AND METHODS: The method adopted is based on the systemic approach of Maturana and Varela, which repositions the researcher as a participant observer actively integrated into the phenomenon under study. This perspective was articulated with documentary research techniques and literature review. In addition to the theoretical survey, the study encompassed an empirical experiment based on the documented use of the GenAI Claude, by Anthropic, intended to assist in structured prompt engineering, and the GenAI NotebookLM, by Google, chosen for allowing operation over a strictly closed and predefined documentary corpus, a methodological strategy essential to mitigating the invention of false precedents. Practical applicability was tested across three sequential, delimited rounds: data extraction and systematization, assistance in cross-referencing the literature review, and spelling review of the manuscript. All interactions invariably occurred under rigorous human intellectual curation, without delegation of value judgment. RESULTS: The main results include the taxonomic mapping of the emerging models most suitable for academic rigor and the demonstration of the inseparable relationship between the traditional stages of legal investigation and the computational functions pertinent to each moment. Empirical validation categorically confirmed that the machine must operate under a model of assistive cognitive amplification, completely refuting the delegation of argumentative capacity and logical-normative subsumption to concrete cases. As an outcome of the investigation, the work consolidated the proposal of an expanded four-step, non-negotiable procedural protocol: the mandatory requirement of full transparency and declaration of use, the requirement of continuous critical validation of results to ward off the pernicious bias of automation, the absolute understanding of algorithmic limitations in hermeneutic reasoning, and the responsible management of the methodological records generated. FINAL CONSIDERATIONS: It is concluded that the use of this technology proves feasible and beneficial when structured under intentional guidelines, consolidating the jurist’s role as intellectual orchestrator and protecting the Law from the serious risks of a rigid, standardized monoculture of interpretations.

PALAVRAS-CHAVE: Generative Artificial Intelligence; Legal research; Prompt engineering; Scientific integrity; Research protocol.