AI in Administrative Law: Synthesizing Public Comments

Administrative law agencies regularly receive hundreds of thousands of citizen submissions on proposed regulations, creating a massive bottleneck in the rulemaking process. This article provides an overview of how agencies deploy artificial intelligence—specifically public-comment clustering and duplicate-campaign deduplication—to rapidly synthesize, organize, and analyze this high volume of public feedback while remaining compliant with legal mandates requiring agencies to address every substantive argument.

The Challenge of Mass Rulemaking

Under administrative law frameworks like the U.S. Administrative Procedure Act, agencies must issue a notice of proposed rulemaking and allow the public to comment. Agencies are legally required to review these submissions and respond to all significant, substantive arguments in the final rule. However, advocacy groups and automated bots frequently generate hundreds of thousands of nearly identical comments. Manually sorting through this volume to find unique, substantive feedback is nearly impossible under strict regulatory deadlines.

Duplicate-Campaign Deduplication

To manage identical or near-identical submissions generated by mass-mailing campaigns, agencies use AI-driven deduplication.

• Exact Matching: Basic algorithms first strip away formatting and punctuation to identify 100% identical text submissions.
• Near-Duplicate Detection: Citizens often take a form letter provided by an advocacy group and add a personal sentence at the beginning or end. AI uses "shingling" (breaking text into overlapping word sequences) and MinHash algorithms to calculate the mathematical similarity between documents.
• Isolating Unique Additions: When a comment is flagged as a 95% match to a known campaign template, the AI groups it with that campaign. Crucially, the software highlights the remaining 5% of modified text. This allows human analysts to read only the unique additions made by the citizen without re-reading the boilerplate text thousands of times.

Public-Comment Clustering

Once duplicates are grouped, agencies are still left with tens of thousands of unique comments. Public-comment clustering AI organizes these remaining submissions by topic and sentiment.

• Natural Language Processing (NLP): The AI converts the text of each comment into numerical vectors (embeddings) that represent the semantic meaning of the words.
• Topic Modeling: Using unsupervised machine learning algorithms, the AI identifies recurring themes across the dataset. For example, in an environmental regulation, the AI might automatically generate clusters for "economic impact on small businesses," "water quality data," and "enforcement timelines."
• Routing to Experts: Instead of reading comments in chronological order, agency staff can view them by cluster. Economists are routed directly to the financial impact cluster, while scientists review the technical data cluster.

Synthesis and Human Review

AI does not replace the legal requirement for human decision-making; rather, it acts as a triage system. After clustering and deduplication are complete, generative AI and extractive summarization tools are often used to draft preliminary summaries of the main arguments within each cluster.

Agency officials then review these clusters and summaries to draft the "concise general statement of basis and purpose" required for the final rule. By using AI to filter out noise and organize arguments, agencies ensure that a single, well-researched comment from an individual citizen is not buried beneath half a million automated form letters, thereby protecting the integrity of the administrative process.