Audit your AI prompt before you run it
A local heuristic scan for the patterns in legal AI prompts that show up badly in discovery.
The argument in this earlier essay is that AI prompts run by lawyers in the course of a matter are likely discoverable, and that most lawyers are not thinking about that. The first practical takeaway is prompt hygiene. Treat the prompt like a memo. Avoid the language patterns that read badly when surfaced.
The privilege risk prompt auditor runs a local heuristic scan against a prompt. It flags six pattern categories: strategy disclosure, subjective characterization, privileged content references, adversary characterization, personal opinion language, and tactical framing. For each finding it surfaces the matched text and suggests a rewrite.
The tool runs entirely in your browser. Nothing is sent anywhere. You can paste a draft prompt, see the findings, rewrite, and rescan.
The deeper version
For matters where the prompt is going to drive a meaningful chunk of work, the local heuristic is a first pass. The tool also generates a structured prompt you can paste into Claude, ChatGPT, or Gemini for a more sophisticated analysis. The deeper prompt asks the AI to evaluate the same six categories with model-grade attention rather than keyword matching, and to produce specific rewrite suggestions.
What this is not
It is not a privilege determination. Work product analysis is fact-specific and jurisdiction-specific. The tool catches the obvious patterns; it does not catch subtle ones. The doctrine is being worked out in real time across the country. Until it settles, the conservative move is to keep prompts spare and task-focused, and to negotiate short vendor retention so even mistakes have a short half-life.
Walter Allison is a corporate attorney in Denver. He writes here about M&A, private equity, and venture capital structure.
Follow on LinkedIn ·
Firm bio ·
More posts