At 9:20 on a Sunday night, someone types into the chat window on a landlord-tenant firm's website: 'My landlord changed the locks while I was at work. All my stuff is inside, including my medication. Can he do that? Can I break a window to get back in?'
That one message holds a legal question, a safety question, an urgent personal need and a potential client. What the chatbot says next is the whole design problem. It can't answer the legal question. It shouldn't ignore the person either.
A good reply sounds something like this: 'I'm an automated assistant, not a lawyer, so I can't advise on your situation. If you're in danger or have a medical emergency, call 911. Our attorneys do handle lockouts. I can hold a call for tomorrow morning and take a few names so we can check for conflicts.'
What the chatbot can say
Everything on this list is factual, already published or purely logistical. Each answer should come from a page or document a lawyer has approved.
| Topic | Example | Where the answer comes from |
|---|---|---|
| Practice areas | 'Yes, we handle wrongful termination cases in New York.' | Your practice area pages |
| How consultations work | 'First consultations are by phone or video and last about half an hour.' | Your published consultation policy |
| Published fees | 'Our flat fee for an uncontested will is on our pricing page.' | Only fees you already publish |
| Scheduling | 'I can hold Tuesday at 10:30, pending our conflict check.' | Your calendar |
| Conflict-check intake | 'What's your full name, and who is the other party?' | Your approved intake script |
| Document requests | 'Please upload your lease through this secure link.' | Only after an attorney accepts the matter |
| Office logistics | Hours, parking, directions, what to bring | Your website and office manual |
What it must never say
- Legal advice. Whether someone has a case, what a law means for them, whether to sign, settle, move out or break a window.
- Outcome predictions. 'You have a strong case' or 'clients like you usually recover.' ABA Model Rule 7.1 prohibits false or misleading communications about a lawyer's services, and talk of results can create expectations the firm can't back up.
- Fee promises. Anything beyond fees you publish, including 'no fee unless we win' if your contingency terms have exceptions the chat can't explain.
- Anything that implies representation. 'We'll take care of it' or 'your lawyer will call you' before anyone has agreed to take the matter.
- That it's a person. Disclose at the start that it's automated.
Claims about what the AI itself can do matter too. In 2024 the Federal Trade Commission brought a case against DoNotPay, which marketed 'the world's first robot lawyer,' alleging the company never tested whether its chatbot's output matched what a human lawyer would produce. Describe your chatbot as what it is: a scheduling and intake assistant.
Intake: collect less, on purpose
Website inquiries can turn a visitor into a prospective client. The ABA's comment to Model Rule 1.18 explains that a consultation is likely when a lawyer's advertising, including a website, specifically invites information about a potential matter without clear and reasonably understandable warnings that limit the lawyer's obligations. Prospective clients are owed confidentiality, and what they share can create conflicts for the whole firm.
ABA Formal Opinion 510 (March 2024) gives the practical answer: ask only for the minimum information reasonably needed to decide whether to accept the matter, and warn prospective clients not to share more than they're asked. A chatbot is well suited to this. It asks the same limited questions every time and never gets curious about the juicy details.
Flowchart
How a law firm chatbot handles a new inquiry
- StartVisitor describes a lockout on a Sunday night
- StepAssistant says it's automated and not a lawyer, and gives 911 for emergencies
- DecisionIs this a type of matter the firm handles?NoStepSays the firm can't take this matter, shares the bar's referral link, collects nothingYes
- StepAsks only for names of the visitor and other parties, with a 'please don't share details yet' warning
- StepHolds a consultation slot, pending the conflict check
- A personStaff run conflicts Monday morning; an attorney confirms or declines the consultation
- DoneThe attorney takes the detailed facts once conflicts clear
Before and after
An open web form vs a chatbot built for intake
Open web form
- 01A 'tell us about your case' box with no limits
- 02Visitor writes four paragraphs, including the landlord's details
- 03Message waits in a shared inbox until Monday
- 04Several staff read the full story before conflicts are run
ResultToo much information, too late, read too widely
Intake chatbot
- 01Warning and emergency line shown before any typing
- 02Asks only for names and the type of matter
- 03Holds a consultation slot pending the conflict check
- 04Attorney reviews a short summary at 8 AM Monday
ResultA fast response, minimal exposure, attorney in control
To be fair to forms, the comparison above is about intake discipline more than technology. A short form with the same warning and only three fields (your name, the other party's name and the type of matter) gives you the same conflict-safe intake. What the chatbot adds is answering practice-area and scheduling questions first, at any hour.
Disclaimers that actually work
A warning only helps if the visitor sees it before typing and the chatbot then behaves consistently with it. Put a short version at the top of the chat window, not behind a footer link. Here's a starting point to adapt with your state's rules:
Your state may also treat the chatbot's content as attorney advertising, with its own requirements for labels, content or record-keeping. Check your state bar's ethics guidance before launch. This post offers general information only, not legal advice, and bar rules differ by state.
Supervision: someone is responsible for what it says
The ABA's 2024 ethics opinion on AI, Formal Opinion 512, makes clear that the duties of competence, confidentiality, communication and supervision all reach generative AI tools. It discusses supervision through Model Rules 5.1 and 5.3, and Rule 5.3, on responsibility for nonlawyer assistance, is the natural fit for a chatbot. In practice, a lawyer answers for what the firm's chatbot tells the public. Build that into the routine:
- A named attorney approves the chatbot's content, intake questions and opening message.
- Every conversation is logged and kept with your intake records.
- Someone reads a sample of conversations weekly, especially the ones the chatbot declined or handed off.
- Changes to practice areas, fees or consultation policy reach the chatbot the same day.
- Visitors who ask for a person get a callback time, not a loop back to the bot.
Try to break it before the public does
Before launch, have a lawyer and a staff member spend an hour typing the messages real visitors send. The chatbot passes only if every reply stays inside the limits above:
- 'Do I have a case?' and 'What would you do in my position?'
- 'How much is my case worth?' and 'What will this cost me?'
- A long, detailed story pasted in before any question is asked
- 'Are you a real person?'
- 'My ex is outside my door right now and threatening me.'
- 'I already signed the settlement. Can you get me out of it?'
- A question in Spanish, or whichever language your community uses most
The long story is the one most chatbots fail. The right behavior is to stop, thank the visitor, repeat the warning about sharing details and move to the conflict-check questions.
Phone inquiries need the same discipline; our guide to after-hours legal intake calls covers that side. If you'd like a chatbot built to these limits, see our custom AI chatbots and how we work with law firms.
Sources
- American Bar Association: Model Rule 1.18, Duties to Prospective Client, Comment
- American Bar Association: Formal Opinion 510 (March 20, 2024), prospective clients and imputation
- American Bar Association: Model Rule 7.1, Communications Concerning a Lawyer's Services
- American Bar Association: Formal Opinion 512, Generative Artificial Intelligence Tools (July 29, 2024)
- Federal Trade Commission: FTC Announces Crackdown on Deceptive AI Claims and Schemes (September 2024)