Legal AI should route, not advise
Sorting a case to the right practice area and lawyer is logistics. Telling a client what the case is worth is legal advice, and only a licensed lawyer can give it. The two look similar in a demo and are nothing alike in front of a bar association.
In short
5 things that decide this
- 01Matching a case to a practice area and an available lawyer is a classification problem. It does not require a law license to perform.
- 02Estimating a case's value, likely outcome, or settlement range is legal advice. An AI system that does this is practicing law without a license.
- 03The safe design keeps AI on the sorting side of that line and routes anything closer to it to a licensed human, every time.
- 04Guided intake forms produce cleaner routing data than a chat interface, because the questions are fixed instead of improvised.
- 05An admin who can see every case's status and every routing decision is what makes the line defensible after the fact, not just in the design document.
Sorting and advising look alike from outside
A prospective client fills out an intake form describing a car accident. The system reads it and does two things that feel like the same task. First, it sorts the case as personal injury and forwards it to a lawyer with an open slot. Then, if it goes one step further, it might say something like: cases like this typically settle for a certain range.
The first task is classification. Giving the second without a license is unauthorized practice of law in every US state. Both can be built by the same engineer in the same sprint. That is exactly why teams cross the line without deciding to.
The rule exists to protect people who cannot judge the advice
Unauthorized practice of law rules are not a formality. A person asking about their accident case cannot spot a wrong answer on their own. That is exactly why the advice has to come from someone licensed and accountable. A chatbot with no bar number and no malpractice insurance carries none of that accountability, however confident its output sounds.
Routing carries a different kind of risk. Send a case to the wrong practice area and a lawyer redirects it. Give a wrong value estimate and a client may reject a real offer, or accept a bad one, based on a number nobody stood behind.
- Guided intake formFixed questions about the facts of the case, not open-ended chat.
- Practice area classificationPersonal injury, family law, immigration. A category, not a judgment.
- Lawyer matchingRoutes to an available, verified lawyer in that practice area.
- Value or outcome estimateLegal advice. Reserved for a licensed lawyer, never the system.
Everything left of the line is classification. Everything right of it needs a license.
Structured intake keeps the system on the right side
Lexpair, a legal lead-matching platform we built, runs on guided forms rather than open chat. A client answers a fixed set of questions about the facts of their case. AI-assisted logic reads those answers and sorts the case into a practice area, then matches it to a verified lawyer with capacity. Admins see every lead and every match from one dashboard.
The form structure does real work here. An open chat window can be led into a value estimate if a user keeps asking. A fixed set of questions cannot. The system has no field for a settlement range, because the product never asked it to have one.
Visibility is what makes the line hold up later
A design decision only counts if someone can check that it held. Every lead on Lexpair carries a visible status and a match history. An admin can see exactly what the system decided, and why, case by case.
That record is what a compliance review wants to see: proof the system routed and did not opine. Building that visibility in after a complaint arrives is much harder than building it from the first sprint.
Questions this raises
01Can AI be used for law firm intake?
Yes, for classification and routing. AI can sort a case into a practice area and match it to an available lawyer without practicing law. The line is crossed only when the system starts estimating outcomes or value.
02Is AI-generated legal information considered legal advice?
It counts as legal advice when it predicts an outcome, not when it classifies one. A description of what a type of case generally involves is information. A statement about what this specific case is likely worth is advice, and only a licensed lawyer can give it.
03How do law firms qualify leads with AI without practicing law?
By keeping the AI's job to classification and matching: which practice area, which available lawyer, verified against fixed intake questions rather than open-ended chat. Anything that predicts an outcome or a value gets routed to a lawyer, never answered by the system.
Related
- Legal software development →Our work building intake, matching and case management systems for legal services firms.
- Does sending client data to an LLM waive privilege? →A related line legal AI systems have to hold: what a tool's terms do to privileged data.
- Legal AI: build vs. buy →Where a custom intake and matching system earns its cost against an off-the-shelf tool.
