Is Your Website Practicing Law Without a License?

A visitor lands on your firm’s site at 11 PM and tells the chat widget the whole story: a contractor took a deposit, never finished the work, and stopped answering calls. "Do I have a case?"

The assistant rises to it. It explains breach of contract, mentions the state’s statute of limitations, suggests the dispute looks like a small claims matter, and offers a rough sense of what the visitor might recover.

Helpful, fast, and exactly the problem. An AI chatbot on a law firm site just analyzed a stranger’s legal dispute, applied law to their facts, and recommended a course of action. That is the practice of law. The thing doing it holds no license, carries no malpractice coverage, and answers to no bar. The firm whose name sits at the top of the page owns every word of it.

Unauthorized Practice Is the First Exposure

Every state restricts the practice of law to licensed attorneys, and the Model Rules make a lawyer responsible for unauthorized practice that happens through them. When a tool on your site applies law to a person’s facts and tells them what to do, it has crossed from information into advice. The line is not subtle in practice. "What is the statute of limitations for breach of contract in Texas?" is information. "Based on what you have told me, you should file in small claims court" is advice, and the second one is what these tools reach for, because answering fully is what they are built to do.

Rule 5.3 closes the escape hatch. A lawyer is responsible for the conduct of nonlawyer assistance, and a chatbot is nonlawyer assistance. You cannot delegate to software what you could not delegate to an unlicensed clerk, and you cannot disclaim your way out of being the supervising attorney.

The Confidence Problem Has a Court Record

There is a failure mode unique to these tools that should end the conversation on its own. Ask one for the case that supports a position and it will, on a bad day, invent one. A plausible name, a plausible citation, a plausible holding, none of it real.

That is not hypothetical. Lawyers have been sanctioned for filing briefs built on citations a chatbot fabricated, after the tool produced cases that did not exist and the filers trusted them. Now move that same machine to the public side of your website, where it answers visitors with the same fluent confidence and no human reads the output first. A prospective client who relies on a confidently wrong answer about a deadline or a right is a malpractice claim wearing a friendly interface.

What a Visitor Tells Your Website, the Vendor Now Holds

A person describing their legal problem to your chat widget is handing over exactly the kind of confidential, sometimes privileged, information Rule 1.6 exists to protect. Where does it go? Through a third-party AI vendor’s servers, very likely, under terms the firm never negotiated and a security posture no one at the firm has audited. The duty of confidentiality does not pause because the intake happened in a chat box instead of a conference room.

The conflict problem rides in on the same intake. A real firm runs a conflicts check before it takes information from a prospective client. A chatbot takes the story from anyone, including the opposing party in a matter you already handle, with no check and no record that protects you. The intake you cannot see is the one that hurts later.

Marketing Rules Do Not Stop at the Chat Box

Attorney advertising is regulated, and Rule 7.1 bars false or misleading communication about a lawyer’s services. A tool that hints at outcomes, suggests how strong a case looks, or sets an expectation about recovery is making exactly the kind of statement the rule governs, and it is doing it in writing, at scale, without anyone reviewing the claims before they go out.

The Disclaimer You Are Counting On

"This chat does not constitute legal advice and does not create an attorney-client relationship." Then the same widget analyzes the visitor’s facts, names the statute, and recommends a filing. A disciplinary board looks at what the tool did, not at the sentence under it. So does the person who relied on it.

Reliance is what matters, and a detailed, specific, authoritative-sounding answer creates reliance no matter what the gray text claims. You cannot tell someone "this is not advice" in one breath and give them advice in the next and expect the first breath to win.

Why a Better Prompt Will Not Save It

The vendor’s answer is always the same: instruct the model not to give legal advice. Tell it to stay general.

It will follow that instruction right up until a question does not look like the thing it was told to avoid, which is most of the time. "Is a verbal agreement enforceable in my state?" reads like a general question, so the model answers it with a mini-analysis that applies law to the asker’s situation. It believed it was being general. It practiced law. The instruction governs the requests that announce themselves and does nothing about the ones that arrive in ordinary words.

The deeper issue is the category. An instruction is a request the model can set aside; it does not change what the model is able to say. A real boundary lives in the system around the model and decides what may leave before anything is generated, so an analysis of a visitor’s facts never reaches them regardless of phrasing. "Will not" is a suggestion. "Cannot" is an architecture.

What AI Chatbot Liability Costs a Law Firm

Bar discipline is the headline risk, and it runs from a private reprimand through suspension, with the firm’s name in the public record either way. Unauthorized-practice statutes carry their own civil and sometimes criminal exposure. A malpractice claim built on advice a visitor relied on starts its defense in the tens of thousands of dollars before anyone argues the merits, and the carrier notices.

The slower cost is reputational. A firm sells judgment and discretion. A public tool that gives sloppy legal analysis or leaks what a prospective client confided says the opposite about how the firm handles trust, and that impression is hard to walk back.

What a Law Firm Actually Needs

Chat earns its place on a firm’s site. Prospective clients want to know your practice areas, request a consultation, find the office, and ask plainly whether you handle their kind of matter. Every one of those is useful and none of them requires a tool that gives legal advice.

The question is not whether the assistant can be coached to be careful about advice. It is whether the assistant is structurally incapable of analyzing a visitor’s facts, predicting an outcome, or taking confidential intake into an unvetted system, no matter how the question is framed. If the safeguard is the quality of the prompt, it fails the first time a question arrives in unexpected words.

Fred is built for that constraint. It answers from your firm’s own content, books the consultation, captures the lead, and routes anything that asks for analysis, a prediction, or a recommendation to an actual attorney. It runs more than 50 industry guardrail packs, and the legal pack is built around unauthorized practice, fabricated authority, and confidential intake. Fred does not weigh your visitor’s case or guess at their odds. It cannot. It tells people what you do and lets a lawyer do the lawyering.

Frequently asked questions