Structured prompts turn deposition transcripts and case files into tables and timelines you can actually work from — here's the exact format that works, and where human verification still matters.
A 300-page deposition transcript used to mean an afternoon of highlighting and margin notes. It still requires a lawyer's read for anything that matters, but Claude has changed what that afternoon looks like — you're reviewing a structured summary and checking it against the source, not building the structure from nothing.
Start with structure, not summary
The instinct is to ask for "a summary of this deposition." That produces a vague paragraph that's hard to use. Asking for a specific structure produces something you can actually work from:
I'm going to paste a section of a deposition transcript. Produce a table with these columns: Page/Line, Topic, Summary of Testimony, and Flag (mark anything that seems inconsistent with earlier testimony, evasive, or worth a follow-up question).
Do not omit any topic covered, even briefly. Do not add interpretation beyond what's flagged — just summarize and flag.
Transcript section:
[paste transcript]Because transcripts are long, I process them in chunks of roughly 15-20 pages and ask Claude to keep a running list of open threads — topics raised but not resolved — that I carry into the next chunk's prompt. This catches contradictions that span dozens of pages, which is exactly the kind of thing that's easy to miss reading linearly but obvious once it's laid out in a table.
Case files: pulling the thread that matters
For case files — medical records, financial documents, correspondence threads — the highest-value prompt isn't a summary at all, it's a targeted extraction:
From the document below, extract only information relevant to [specific issue, e.g. "the plaintiff's employment history and any gaps"]. Present it as a chronological timeline with dates, source document, and a one-line note on relevance. Ignore everything not relevant to this issue.
Document:
[paste document text]The verification step that isn't optional
Every summary gets checked against the source before it goes into a work product. Claude is very good at structuring information that's in front of it — it's much less reliable when a document is long enough that context gets lost, and it will occasionally smooth over a genuine inconsistency instead of flagging it. Treat the summary as a map, not the territory.
A good summary tells you where to look. It doesn't replace looking.
Kumail Mehdi
ISO 42001 Lead Auditor and AI strategist. 11 years in corporate leadership, 14 years running a digital agency. I help professionals, consultants and AI startups turn expertise into governed, AI-powered systems.
