Feb 16 2026
Regulation is built, justified, contested, and revised through text: draft rules, technical analyses, public comments, and agency responses. That written record is a gold mine for scholars—but it is also a practical barrier. When the evidence is mostly presented in the form of text documents, measurement is slow, difficult to standardize, and hard to scale across many proceedings or years. In my PhD research, I studied responsiveness to public input by ANATEL (Brazil’s telecommunications regulator), using thousands of consultation submissions linked to the agency’s responses, which indicate whether each suggestion was accepted, partially accepted, rejected, or treated as out of scope. Doing that work forced me to confront three recurring research challenges: identifying who is in the record, measuring what participants are asking for and how they justify it, and connecting that content to agency responses in a meaningful, rigorous way. This methods note tells that story, emphasizing a methodological tool that can greatly improve and expand efforts to study regulation: using a general-purpose AI model as a research assistant. While acknowledging multiple AI-based approaches to text analysis (text representations, predictive modeling, and training specialized models), in this piece I focus primarily on the most intuitive workflow—one that can be useful for a wide variety of research applications and research teams: LLM-assisted coding, that is, using general-purpose AI models to read text and extract specific, auditable measures that can be validated through traditional methods and used in social science research
Full paper to download at
