AI Intake for Law Firms: How It Works and What to Test Before You Buy
AI intake is a chat on your website that talks with a new visitor, collects what your team needs and sends you a record. The AI writes the conversation. Everything around it, from what it may never say to where the record goes, is set by rules your vendor wrote and you approve. Before you buy, test the parts a demo does not show: the record your team actually receives, what happens when delivery fails, where the data goes and how you supervise it after launch.
How AI intake works, step by step
- A visitor opens the chat. It is usually a small window on your website. The first message says it is an AI assistant and not a lawyer.
- The AI writes each reply. A large language model reads the conversation so far and writes the next message, within instructions the vendor configured for your firm: your practice areas, your service area, your office details and what it must never discuss.
- Some sentences are fixed. For the moments that matter most, a well-built tool does not improvise. The AI disclosure, the “this is not legal advice” line, emergency resources and your fee sentence should come out word for word, every time.
- The visitor confirms their details. Before anything is sent, the tool reads back what it collected and asks the visitor to correct it. A mistyped phone number is a lost client.
- Your team receives a record. Usually by email, sometimes also in your case-management system or as a text alert. It should be short, the same shape every time and in your team’s working language.
What the AI does not do is decide. It does not decide whether the firm can take the matter, what it is worth or what happens next. Your attorney does.
What to test before you buy
A demo shows the conversation. These eight tests show the rest. Most take a few minutes, and a vendor who has built the product carefully will welcome them.
1. Run a real test intake and read the record. Start a conversation as a new client, finish it and then open what your firm received. Does it match what you typed? Is the other party’s name there for the conflict check? Is urgency marked, with a reason? If you cannot use the record in a morning callback, the conversation does not matter.
2. Ask what happens when delivery fails. Email fails more often than people think, and it often fails silently: the chat looks fine, and nothing arrives. Ask the vendor how a failed send is detected, who is told and whether every send is logged. “We’d notice” is not an answer.
3. Test the hard moments. Ask what your case is worth. Ask whether you have a case. Say you are in danger. A good tool hands the first two to the attorney and puts emergency resources first on the third. Our legal chatbot checklist has eight messages to type and what a good answer looks like.
4. Test your languages end to end. If you serve Spanish speakers, start a conversation in Spanish and stay in Spanish. A greeting in Spanish followed by English questions is not Spanish intake. Then check that your team’s record arrives in the language your team works in.
5. Ask where the data goes. Ask for the list of outside companies that process your visitors’ information (often called subprocessors), whether any of it is used to train AI models, how long it is kept and what happens to it when you leave. Ask for the data processing agreement. A visitor may tell your chat about a divorce or an injury; that information deserves the care you give your own files.
6. Ask what you can change, and how fast. Your practice areas, your service area, the person who receives the leads, your fee wording. Who makes the change, who approves it and how long does it take? Ask what is fixed on purpose, too. Some things, like the safety answers, should not change from firm to firm.
7. Read the pricing terms. What counts as a conversation or a lead? What happens in a busy month? Is there a minimum term, and how much notice does canceling take? The monthly figure matters less than what it covers.
8. Ask how you will supervise it. Will someone at your firm be told about every conversation? Can you review what the chat said? Will the vendor tell you before the tool’s wording or underlying AI changes? Plan to run your own tests again after every change and on a regular schedule.
What bar guidance says
Bar guidance on chatbots keeps returning to the same points. The Florida Bar’s Ethics Opinion 24-1 says a firm using a chatbot must tell visitors they are talking to AI and that the lawyer is responsible for what it says. ABA Formal Opinion 512 applies the duties of competence, confidentiality and supervision to generative AI tools. The Oregon State Bar’s Formal Opinion 2026-208, approved in February 2026, addresses chatbots used for client intake directly. Among its points: the chatbot should identify itself as automated, should not give legal advice, a responsible lawyer should be notified of conversations and review them promptly, prospective clients reached through a chatbot belong in the firm’s conflict system, and the firm’s due diligence should reach past the vendor to the AI and other software underneath. It also says the evaluation should happen at purchase and again from time to time, because the technology changes.
Underneath all of it is ABA Model Rule 1.18 and its state versions: someone who shares information about a possible matter can become a prospective client, with duties that follow even if the firm never takes the case. Rules differ by state, so read your own.
The bottom line
AI intake is not magic and not a black box. It is a conversation written by a language model inside rules someone wrote, ending in a record someone has to read. Buy the one whose record you would use, whose delivery you can trust and whose rules you can see and supervise.
Common questions
Is AI intake allowed for law firms?
Generally, yes, with clear disclosure that it is AI, no legal advice and a lawyer supervising it. Bar guidance varies by state, so check the rules where you practice.
Does AI intake replace intake staff?
It covers the hours and the volume your staff cannot, especially nights and weekends. Your team still calls the person back, and the attorney still decides.
Can AI intake give legal information?
The line between general information and advice is easy to cross in a live conversation. The safer design keeps the chat to intake and next steps and leaves legal questions to the attorney.
How long does it take to set up?
It depends on the vendor. Ask what they need from you, such as your practice areas, service area and logo, and how long the build takes once they have it.
Lexara AI’s intake assistant was built and tested by an attorney. It says it is AI in its first message, gives emergency resources first when someone is in danger, never estimates case value or deadlines and quotes only your firm’s fee wording. It reads the details back before sending, holds the conversation in English, Spanish or Haitian Creole and delivers the record to your team in English, through our own monitored delivery with every send logged. Client data is never used to train AI models. Try the live demo on our homepage.
Related: Legal Intake, Explained — our complete guide · Legal chatbot checklist · Legal intake software for small firms · Legal intake compliance guide · AI for personal injury lawyers · all guides.
Florida Bar Ethics Opinion 24-1 (January 19, 2024); ABA Formal Opinion 512 (July 29, 2024); Oregon State Bar Formal Opinion 2026-208, Chatbots and AI Agent Communications (February 2026); ABA Model Rules 1.18 (prospective clients) and 5.3 (nonlawyer assistance). All paraphrased. Rules differ by state; nothing on this page is legal advice. Checked November 6, 2026.