Every judge profile on CaliforniaCourtIntel includes an AI-generated narrative — a plain-English summary of the judge's courtroom style, ruling tendencies, and procedural preferences. Here is exactly how that narrative is created.
Data inputs. The narrative AI draws on three categories of source material. First, tentative ruling text: for courts that publish tentative rulings publicly, the AI ingests the full text of every available ruling. It analyzes recurring language patterns, the depth of reasoning, and the factors the judge emphasizes when deciding motions. Second, attorney observations: practitioner-submitted notes that have cleared AI moderation are incorporated as qualitative input. The AI synthesizes recurring themes across multiple observations rather than reproducing any single attorney's words. Third, structured statistics: grant rates, ruling frequency, appointment year, and courthouse assignment provide quantitative context that shapes the narrative's framing.
How the AI writes the narrative. A large language model processes all three input types and generates a first-draft narrative. The model is instructed to write in plain English accessible to any attorney, stick to patterns supported by multiple data points rather than extrapolating from a single event, explicitly flag uncertainty when data is limited, and avoid language that could be interpreted as defamatory or legally problematic.
Editorial review. Narratives for judges with 50 or more data points go through an automated quality check before publication. Narratives for high-profile judges or judges who have received accuracy flags from users are reviewed by a member of CaliforniaCourtIntel's research team before going live.
Refresh schedule. Narratives are regenerated each time a meaningful new data point is added — typically when 5 or more new tentative rulings are ingested, or when 3 or more new attorney observations are approved. The last updated timestamp on the profile header reflects the most recent regeneration.
Limitations and responsible use. Narratives cannot capture sealed proceedings, judicial behavior outside public tentative rulings, or first-hand personality observations. They reflect statistical patterns in available public data, not a holistic assessment of a judge's character or jurisprudence. Use the narrative as a starting point for your preparation, not as a complete or definitive picture. If you believe a narrative contains a material inaccuracy, click Flag as Inaccurate at the bottom of the AI Insights panel and our data team will review it within 48 hours.