AI Chatbot for Law Firms: Qualifying Cases Without Touching Attorney-Client Privilege

An AI chatbot for a law firm answers a real need: a prospect who reaches out outside office hours, with something urgent — a custody hearing, a dismissal, a divorce — doesn’t want to wait until tomorrow to know if they’ve found the right firm. But the question that stops most firms isn’t whether the technology works, it’s what that prospect is about to type into a chat window, and where it goes next. France’s National Bar Council (Conseil National des Barreaux, CNB) adopted, on March 17, 2026, an ethics guide on generative artificial intelligence built on one simple principle: AI is tolerated “occasionally and under supervision,” provided the lawyer keeps intellectual control and responsibility over the case at all times. An intake chatbot can keep that promise — if it’s built to qualify, not to listen.

What an intake chatbot can do — and what it must never do

The useful scope is deliberately narrow: which area of law applies (family, criminal, employment, business), how urgent it is, an appointment slot, documents to bring, the price of a first consultation. Nothing on that list resembles legal advice. What the chatbot must never do: phrase a response that reads like a legal position, even a cautious, “informational only” one — and above all, invite the prospect to describe the facts of their case in free text before a lawyer has reviewed the request. An open field like “describe your situation” is exactly where information that qualifies as privileged slips through, the moment it is shared with the intent of obtaining advice.

• →Replace open text fields with closed choices: area of law, perceived urgency, region — never “explain your situation.”
• →Hand off to a call or a secure form the moment a prospect strays from the script, instead of forcing a generated answer.
• →Pick a vendor that doesn’t train its models on conversations and documents a short retention period for exchanges that never turn into a case.
• →Keep a timestamped log of what was asked and answered — useful if the bar association ever audits your intake, as the CNB guide recommends.

The real constraint is privilege — not the technology

Qualifying a request with AI is unremarkable technology in 2026; the risk is specific to the profession. A 2024 study published in Jurimetrics, Exploring the Intersections of Privacy and Generative AI: A Dive Into Attorney-Client Privilege and ChatGPT, by Jordyn C. Tye, shows that the boundary of attorney-client privilege becomes uncertain the moment a client or lawyer enters confidential information into a general-purpose conversational tool, outside any contractual confidentiality framework — exactly the scenario of a chatbot left open to any question. It’s the same compartmentalization principle we applied when hardening a banking WhatsApp chatbot, another regulated sector where every exchange carries a strict confidentiality obligation: splitting the flow into separately validated sub-workflows, a supervisor QA agent ahead of production, and full logging of what passes through, rather than one single flow that sees the entire conversation.

The architecture that works looks a lot like the B2B lead-qualification chatbot for a website that we covered in an earlier article: a few closed questions, a score, a fast handoff to a human — except here, the cost of one question too many is measured not in conversion rate but in ethics risk. Before you connect a chatbot to your firm’s contact form, inventory what it is allowed to ask — not what it is able to answer. That scope, not the sophistication of the model behind it, decides whether the tool saves you time or exposes you.