Executive Summary
Most lawyers don’t need “more AI”. They need a safe way to use it in real matters without losing control over confidentiality, sources, and accountability.
This checklist is meant to be practical: what to avoid, what to redact, and what to keep inside a controlled workflow, so the lawyer stays in the driver’s seat.
Key Takeaways
- Start with a “data diet”: keep identifiers and sensitive facts out unless you truly need them.
- Separate thinking from filing: use AI for structure, alternatives, and checks — but verify before you send.
- Prefer workflows over free chat: controlled steps, inputs, and checkpoints reduce risk.
1) First decide: what are you asking AI to do?
Before you paste anything, define the output: a short memo, a legal opinion outline, a risk list, a contract clause comparison, or a negotiation plan. “Help me with this case” is too broad.
Clarity makes you safer — because you can limit the inputs to what’s needed.
2) Categorize your inputs (and cut aggressively)
Think in three buckets:
- Public / non-sensitive: statutes, public decisions, templates you already publish, generic fact patterns.
- Sensitive but reusable: anonymized facts, internal playbooks, redacted contract excerpts.
- Highly sensitive: client identities, trade secrets, full document sets, privileged communications.
If you can solve the task with bucket #1 or #2, don’t use bucket #3.
If you would not forward the text to a trusted colleague over email without a second thought, don’t paste it into an AI tool as-is. Redact or restructure first.
3) Redact smart: keep what matters for the legal analysis
Bad redaction removes the legal signal. Good redaction removes the identifiers.
Redact these by default
- Names of people and companies (replace with “Client”, “Supplier”, “Employee”).
- Case numbers, deal codes, invoice numbers.
- Addresses, phone numbers, personal IDs.
- Exact dates when not essential (use relative timelines: “T0”, “T+30 days”).
Keep these when needed for accuracy
- Jurisdiction, forum, and governing law.
- The legally relevant sequence of events.
- The key contractual clause text (only the clause, not the full contract).
4) Use AI for “structure + alternatives” — not as the final authority
AI is useful for:
- Turning messy facts into a clean outline.
- Generating alternative arguments or counterarguments.
- Spotting missing issues (“What else should I check?”).
- Drafting a first-pass structure for a legal opinion, then you fill the verified content.
AI is risky when used as a source of truth. Your safety comes from verification, citations, and checkpoints.
5) Prefer a controlled workflow (Agentic Tasks) over ad-hoc copy/paste
A workflow approach is safer because it encourages a consistent sequence:
- Intake (what the matter is and what output is needed)
- Documents (select only what’s needed)
- Research (sources, citations, and a check step)
- Draft (structured deliverable)
- Verify (lawyer review, edits, and final sign-off)
That’s the direction Wisanna is built around: lawyer-controlled workflows from intake to deliverable, with explicit checkpoints.
6) A simple internal policy (you can adopt today)
One-page AI confidentiality policy
- Default to anonymized facts; never paste identities unless necessary.
- Never paste privileged client communications into generic tools.
- Use AI for structure and options; verify before sending to anyone.
- Keep a short “what we used AI for” note in the file (for internal accountability).
Want a guided, lawyer-controlled workflow?
See Wisanna’s Agentic Tasks (including Draft Legal Opinion) and how teams use them safely, with the lawyer in control.
Learn More