Practical Applications
Chapter 6: Prompt Engineering for Litigation
When Should
When Should
I Prompt?
In This Lesson
- check_circle Identifying the reliability ceiling of current LLMs.
- check_circle Applying the "30-Minute Rule" to task delegation.
- check_circle Technical constraints: Context windows and token limits.
The 30-Minute Rule
Framework for AI reliability and professional risk management.
A general heuristic for technical risk: If a task takes a competent human more than 30 minutes to complete, the probability of LLM "hallucination" or reasoning failure increases exponentially.
verified
Reliable Zone
Standardizing boilerplate, summarizing short emails, drafting simple cover letters, or reformatting data structures.
warning
High-Risk Zone
Deep synthesis of trial transcripts, identifying subtle inconsistencies in 100+ page depositions, or complex jurisdictional analysis.
Pro Tip
"Always prompt the AI to include direct citations from the source text. If it cannot find a quote, it is less likely to invent a fact."
Execution Strategy
Strategic Input:
Strategic Input:
Bite-Sized Curation
For litigation professionals, the "dump and run" approach—uploading 500 pages and asking for a summary—is a recipe for technical failure. The AI’s "attention" is diluted across the broad context.
Instead, utilize Manual Curation:
- keyboard_double_arrow_right Break large records into 10-15 page "chunks" for focused analysis.
- keyboard_double_arrow_right Process witnesses individually before asking the AI to compare testimonies.
- keyboard_double_arrow_right Verify the "Critical Path"—manually check the AI's output against the 3 most important pages.
Tokens & Context Windows
Understanding the physical limits of artificial intelligence through the lens of legal documentation.
The Unit of Measure
600
Average tokens per single page of standard legal 8.5" x 11" paper.
GPT-3 Era
4K
Context limit (~7 pages). The AI would "forget" the beginning by the end of a short brief.
GPT-5 Target
1M+
Theoretical limit (~1,600+ pages). Greater capacity, but higher risk of "lost in the middle" phenomena.
The Printer Paper Analogy
Think of the context window as a physical desk. GPT-3 had a desk the size of a clipboard. Modern models have desks the size of a boardroom table. However, even on a large table, if you pile 1,000 pages high, the AI can only "see" the top sheet and the few it's currently holding. Curation is the act of putting the right page on top.
Action Required
Validate Your Knowledge
To receive CLE credit for this module, you must complete the Technical Risk assessment. This ensures you understand the "30-Minute Rule" and token constraints in practice.
BEGIN KNOWLEDGE CHECK open_in_new