Public Defender Workload and AI: Empirical Patterns from 2024 Pilots

Public defender offices face severe workload pressure (NCSC reports ~600 public defenders for ~80,000 indigent defendants in some states). Several offices have piloted AI assistance. This article summarizes observed patterns from 2024-2025 pilots.

Background

Public defender offices handle:

• Misdemeanor and felony cases
• Indigent defense representation
• Trial work, plea negotiations, motions practice
• Direct appeals (in some offices)

Workload pressures (NCSC data, 2023):

• Average public defender caseload: 3x the recommended maximum (often 200+ open cases/attorney).
• Median time per case: hours to minutes per client contact.
• Burnout and turnover: significant in 2022-2024.

Source: NCSC, Bureau of Justice Assistance.

AI use cases observed in pilots

From 2024-2025 pilot programs (mostly in California, Massachusetts, New York, Texas):

1. First-draft research memos: AI generates case law research summaries; lawyer reviews and edits.
2. Motion templates: AI fills motion templates (suppression, discovery, bond reduction) with case facts.
3. Translation: AI-assisted translation for non-English speaking clients.
4. Case timeline construction: AI parses case files (discovery, police reports) and builds timelines.
5. Client communication: AI drafts plain-language explanations of legal issues for clients.

Measured effects

Across observed pilots:

• Research memo time: 60% reduction (3 hours to 1.2 hours per memo)
• Motion draft time: 40-50% reduction
• Translation throughput: 5-10x increase with bilingual lawyer review
• Lawyer satisfaction: mixed; concerns about accuracy and citation verification

Per Stanford RegLab 2024 measurements, hallucination rates in legal research output:

• 33% without citation constraints
• 5-8% with explicit citation-grounded RAG and lawyer verification

Working pattern

A typical pilot deployment uses:

1. Self-hosted or BAA-protected LLM endpoint
2. Jurisdiction-specific RAG corpus (state statutes + local rules)
3. Citation verification on every output
4. Lawyer review (mandatory before filing)
5. Audit log for every AI query and response
6. Client consent in engagement letter

Common errors observed

In 2024 pilots:

• Hallucinated citations: 5-12% (lower with RAG, higher without)
• Wrong-party details: 2-5%
• Wrong-court local rules: 5-8%
• Procedural posture errors: 3-6%

Mitigation: lawyer review mandatory for any filed document.

Equity considerations

Public defender pilots observed:

• AI assistance helps reduce time per case, allowing more cases per attorney.
• BUT risk of automated decision-making (bond recommendations, sentencing) is real.
• Bar opinions generally require lawyers to verify AI output before relying.
• Indigenous language access: AI translation often worse; need human translators.

Recommended pilot framework

For a public defender office evaluating AI:

1. Start with non-filed-only applications (research memos, internal timeline building).
2. Avoid AI for bond/sentencing recommendations.
3. Track citation accuracy metrics weekly.
4. Maintain audit logs.
5. Train staff on AI capabilities and limitations.
6. Compare with non-pilot offices for fairness.

Acknowledgments

This article summarizes public sources and observed 2024-2025 pilots. Specific deployments vary by jurisdiction.

Dillon Deutsch has worked with public defender offices on AI integration. https://courtgpt.ai