Most legal AI demos answer a question and stop there. The hard part isn’t producing an answer — it’s knowing whether that answer is general legal information or specific legal advice. That distinction is the line that the unauthorized-practice-of-law (UPL) rules are built around, and it is the line a generic chatbot has no way to see.
General information describes the law: what a statute says, how a doctrine generally works, what a clause typically does. Advice applies the law to a particular person’s facts and tells them what to do. The same sentence can fall on either side of the line depending on who is asking and why.
Why generic AI can’t hold the line
A general-purpose model optimizes for a helpful-sounding answer. It does not distinguish between explaining a concept and counseling a specific client, because nothing in its objective rewards that distinction. The result is fluent text that drifts across the boundary without flagging it — which is exactly the failure mode a firm cannot afford in front of a client.
What a two-lane guard does
A governance layer treats the boundary as a first-class control rather than a hope. Every output is classified into one of two lanes before it ships:
- Information — describes the law in general terms. Ships, with the standard disclaimer attached.
- Advice — applies law to specific facts. Blocked from auto-delivery and routed to a supervising attorney.
Enforce, don’t promise
The difference between a marketing claim and a defensible posture is enforcement. Classification, the enforcement decision, and the disclaimer should all be logged per matter, so the firm can show — not assert — that the line was respected on every response. That audit trail is what turns AI from a liability question into a supervised tool.
Information and workflow assistance — not legal advice. Does not create an attorney–client relationship.
See this applied to your firm.
A 30-minute walkthrough on your real use cases.