Turn a pile of documents into something your audience can explore.
This site shows how to use AI to gather information, sort it into a clear structure, display it, and package it as handouts, PDFs and slides in different lengths, and, just as important, how to check the work before anyone relies on it. It teaches through one real project you can click through.
One mark for each of the 176 opinions. Circles were affirmed, squares were error that was harmless, dashed rings were raised but not reached, and diamonds were reversed.
One year of court opinions, one question.
“Reversed” counts opinions where an evidence ruling contributed to reversal. Other opinions were reversed in part on non-evidence grounds such as double jeopardy or fees.
The question was simple: where do trial-court evidence rulings get appealed, and where do they actually get reversed? We used AI to sort the opinions into a fixed set of categories, run on full-text opinions from July 1, 2025 to June 30, 2026, then built an interactive map where every opinion is one mark you can hover, click and trace to the official court document.
Four ways in
The process
Six steps from question to finished materials, with who did what, and what went wrong.
Walk through it →The heat map
Every opinion as a mark. Filter, search, hover, click through to the court’s own document.
Open the map →Prompts to copy
A simple structure for asking well, plus ready-to-use prompts for each step.
Get the prompts →Handouts and slides
A discussion card, poster and field guide, plus slide decks for 5-, 10- and 15-minute slots, all from the same data.
See them all →Three habits that made it work
- Treat AI output as a fast first draft, never as a source. It is fast at reading, sorting and drafting a first pass. It is not an authority on whether something is true, and it will sometimes be confidently wrong.
- Ask it to organize before you ask it to conclude. Fix your categories first, then let it sort. Conclusions come from the numbers, not from the model’s confidence.
- Build in a check, then check the check. We had the AI audit its own work, and the audit found real mistakes. The audit also showed what is still undone: no person has yet read every record.
A starter prompt
Turn a topic into a teachable question
I teach [audience, e.g. new paralegals] and want a 15-minute session on [topic]. Suggest 5 specific questions about this topic that: 1. can be answered from public information, 2. would surprise or interest this audience, and 3. could be shown in one picture. For each, tell me what data I would need and the biggest way it could mislead people. Then recommend one and explain why.
Judicial educators, and the people who help them prepare
Faculty, program planners, court education staff, instructional designers, librarians, and anyone asked to “put together something on this” by Friday. No programming is needed to follow the process or use the prompts.
It is a teaching example of a method. The court-opinion data is a screening-level dataset: useful for seeing patterns, not for citing as findings. The About page says what has and has not been checked. It is an Exploratory Project by Danielle M. Teagarden. Not legal advice, and not an official statement of any court or organization.