Law firm chatbots and ABA Formal Opinion 512.

What Opinion 512 is

The ABA's Standing Committee on Ethics and Professional Responsibility issued Formal Opinion 512, Generative Artificial Intelligence Tools, on 29 July 2024. It was the committee's first formal opinion on generative AI, and it runs to 15 pages.

It interprets the ABA Model Rules of Professional Conduct. The Model Rules bind nobody by themselves: each state adopts its own rules, usually close to the model, and its own bar or courts enforce them. Several state bars have published their own AI guidance as well, among them Florida's, and Texas's Opinion 705 of February 2025.

So read 512 as the profession's common starting point, then check your state's rules and opinions. This guide is general information about those rules, not legal or ethics advice.

Opinion 512 is guidance. Your state's rules of professional conduct are what bind you, so check them and your bar's own AI opinions before you launch.

The duties it applies

The opinion does not create new duties. It applies existing ones to AI tools. In the ABA's own summary:

• Competence, Rule 1.1. Lawyers need to understand the benefits and risks of the technology they use.
• Confidentiality, Rule 1.6. Information relating to a client's representation stays confidential unless the client gives informed consent, and other rules extend the same protection to former and prospective clients.
• Communication, Rule 1.4. Lawyers must reasonably consult clients about the means used to reach their goals.
• Fees, Rule 1.5. A lawyer may bill the time spent putting information into a tool and checking what comes out, but in most cases not the time spent learning to use it.

On confidentiality the opinion goes a step further than general care. Texas's ethics committee sums up the ABA and Florida opinions as suggesting that lawyers get informed consent before using generative AI tools with confidential client information, and asks Texas lawyers to apply at least the same careful analysis.

Where an intake chatbot fits

Opinion 512 is mostly about lawyers using AI in their own work: research, drafting, reviewing documents. A chatbot on the firm's website differs in one way that matters. It talks to the public, before anyone has checked for conflicts or agreed to act.

That brings in rules that sit beside the opinion's duties. Two matter most for intake: Rule 1.18 on prospective clients, and Rule 7.1, which bans false or misleading communications about the lawyer or the lawyer's services. Add Rule 1.6(c), which requires reasonable efforts to prevent unauthorised disclosure of, or access to, client information, and Rule 5.3 on work done by people and services outside the firm.

Read together, they describe what a firm should expect from its chatbot. The next sections take them one at a time.

What a prospective client types is protected

Under Rule 1.18, someone who consults a lawyer about possibly forming a client-lawyer relationship is a prospective client. Even if the firm never takes the matter, it may not use or reveal what that person told it, beyond what the rules allow for former clients. The comment to the rule adds that the duty applies however brief the first contact was.

Two points in the comment fit a chatbot closely:

• A consultation is likely to have happened when a lawyer, including through advertising in any medium, invites information about a possible matter without clear warnings that limit the lawyer's obligations, and the person sends information in response. A chatbot that asks "Tell us what happened" is an invitation.
• To avoid taking in information that could later disqualify the firm, a lawyer deciding whether to take a matter should limit the first consultation to what is reasonably necessary for that decision.

So an intake chatbot should ask for what the conflict check and the first call need: the names of the people and companies involved, the type of matter, key dates, contact details and how urgent it is. It should ask the visitor to keep the story for the lawyer, and show the firm's approved warning before the visitor types anything substantial. The example conversation on our law firms page works this way: the chatbot takes the employer's and the manager's names, then asks the visitor to save the details for the call.

Why it never gives legal advice

No rule says "a chatbot may not give legal advice". The profession expects it anyway, for reasons that follow from the rules above:

• Advice needs a lawyer's judgement about one person's facts, and the only facts a chatbot has are whatever the visitor typed.
• A confident wrong answer about a deadline or a claim can cost the visitor their case, and they will say the firm told them.
• What the chatbot says about the firm is a communication about the firm's services under Rule 7.1. "You have a strong case" or "we usually win these" can mislead.
• A visitor who gets an answer about their own matter may reasonably think the firm is now acting for them.

The line runs between the firm and the visitor's matter. "Do you handle tenancy disputes?" gets an answer from your practice areas. "Can my landlord keep my deposit?" gets no view at all, and an offer of a call-back. Time limits do the most harm, so the chatbot never says whether one has passed or how long someone has. What a chatbot should hand over draws the same line for other trades.

Confidentiality: follow the conversation

Rule 1.6(c) asks for reasonable efforts to stop client information being disclosed or accessed without authority. For a chatbot, that means knowing every place a conversation goes:

• The chatbot company's servers, and which country they are in.
• The AI service that writes the replies: whether it keeps the text, for how long, and whether it uses it to improve its product. The business plans of the major AI services do not use customer content for that by default. Personal accounts often do.
• The email or chat message that alerts your intake team, and who can read that inbox.
• Your case management system or CRM, if requests go there.
• Anyone at the chatbot company who can open conversations.

Enquiries that never become matters are the awkward part. The firm still holds names and a description of someone else's dispute. Set a retention period for them and limit who can open them, for example to the intake team, with transcripts kept in your case management system or a restricted inbox rather than a shared one. For the same questions about AI tools your staff use, see Is it safe to put customer data into ChatGPT?

Vet the vendor as Rule 5.3 expects

The comment to Rule 5.3 gives using an internet-based service to store client information as an example of work done by non-lawyers outside the firm. A lawyer using such a service must make reasonable efforts to see that it is provided in a way that fits the lawyer's professional obligations. How much effort depends on the provider's experience and reputation, the nature of the service, the terms that protect client information and the law where the work is done, especially on confidentiality.

Lawyers with managerial authority also need measures in place that give reasonable assurance that outside services working on firm matters act in line with those obligations. In plain terms: a written agreement, read by someone who understands it, and a named person at the firm who owns the chatbot after launch.

What to ask whoever builds it

Ask these before you sign, and keep the answers with the agreement:

• Where are conversations stored, in which country, and for how long by default?
• Which AI service writes the replies, and does it keep or reuse what visitors type?
• Who on your side can read our conversations, and is that access logged?
• Can we set our own retention period and delete an enquiry on request?
• Will you tell us promptly about a security incident, or about a demand for our data such as a subpoena?
• Can the chatbot be limited to answers we have approved, and does it say when it does not know?
• Can it collect conflict check names in the format our system needs?
• Can our approved disclaimer and urgent rules be used word for word, and changed only by us?
• How do handovers reach our intake team, and what happens if that connection fails?
• Can we read every conversation, and export them all if we leave?

The warning and the first message

The comment to Rule 1.18 talks about clear and reasonably understandable warnings. The firm writes its own, approved by a partner or whoever handles ethics, and the chatbot shows it before the visitor types anything substantial. It might cover:

• This is an automated assistant, not a lawyer.
• It cannot give legal advice.
• Using it does not make you a client of the firm.
• Please do not send the details of your case here. A lawyer will ask for them after a conflict check.
• If something is urgent, such as an arrest or a hearing this week, call the number shown.

Keep it short and show it once, at the start. Then let the chatbot be useful. A warning repeated in every reply loses the enquiry without protecting anyone. Some US states and the EU now also expect a chatbot to say it is an AI, which the first line covers: see chatbot disclosure laws.

Supervising it after launch

A chatbot is not something to set up and forget. A named person at the firm owns it:

• Reads a sample of conversations each week, and every one flagged as urgent or as a complaint.
• Tests it after each change with awkward real enquiries: "Do I have a case?", "Is there a time limit?", the other side in one of your matters, a journalist, someone in danger.
• Approves changes to the disclaimer, the practice areas and the urgent rules.
• Keeps a short log of what changed and why.
• Makes sure the intake team knows what the chatbot says, so a caller who says "your chatbot told me" gets a sensible answer.

The same care applies to AI tools your own team uses. If lawyers or staff paste client emails into chat tools on personal accounts, that is a confidentiality problem under Rule 1.6 before it is anything else. An AI policy for staff gives a one-page rule, and an AI assistant for your documents gives the team an approved tool with agreed access, where each answer shows its source.

How our law firm intake works

Our law firm intake chatbot never gives legal advice, and tells the visitor so. It takes the type of matter, contact details and how urgent it is, collects conflict check names in the format your team needs and asks for a call time. It answers only from information the firm approves, says when it does not know and passes anything about the visitor's own situation to a person. The disclaimer and regulatory wording come from a partner or your compliance officer, and we use them word for word.

Before launch we test it with your past enquiries, names removed, and with the awkward ones above. Your team tries it before any visitor does, and afterwards you can read every conversation. Our AI chatbots page explains how a build runs, the prices page lists the setup, and the lead capture chatbot shows the same intake pattern outside law.