Yes, when the lead asked first. An AI intake agent that calls back a person who filled in your form is answering a request, and the ABA’s own comment says a communication made “in response to a request for information” is not a solicitation. The same agent dialing an accident victim from a police report is a solicitation by real-time voice, which Model Rule 7.3(b) bans and which Texas Rule 7.03 names outright as “electronically interactive” contact made “on behalf of a lawyer”.
I read both texts and their comments on 28 September 2026 because every AI intake build for a law firm starts with the same request: call every new lead back inside a minute. Whether the agent may do that is decided by a rule written for humans, and the rule is clearer than the debate around it.
The rule never asks whether a person or a program made the call. It asks who started the conversation.
The mistake: buying intake as a receptionist decision
Most intake conversations I have start like a receptionist purchase: hours covered, price per call, whether it books the consult. Those are the right questions for inbound. They are the wrong questions for the second thing firms ask the agent to do, which is reach out. The moment the agent places a call to someone who has not contacted you, you have left the answering-service question and entered the advertising rules of your state bar, and the duty in those rules is addressed to the lawyer, not to the software.
The rule turns on who started the conversation
Model Rule 7.3(a) defines a solicitation as a communication “initiated by or on behalf of a lawyer or law firm” that is “directed to a specific person the lawyer knows or reasonably should know needs legal services in a particular matter” and offers to provide them. Three elements: you started it, you picked the person, you offered. Paragraph (b) then bans solicitation “by live person-to-person contact” when a significant motive is money, with three exceptions: another lawyer, someone with a family, close personal or prior business or professional relationship with the firm, and a person who “routinely uses for business purposes the type of legal services offered”.
Comment [1] is where the inbound agent lives. A communication is not a solicitation if it is “in response to a request for information or is automatically generated in response to electronic searches”. A web form, a chat, a missed call and a text that says “call me” are all requests. The callback answers them, so the element “initiated by” fails and Rule 7.3 never engages. Texas puts the same idea in its definition: a solicitation communication goes to “a specific person who has not sought the lawyer’s advice or services” (Rule 7.01(b)(2)). Someone who submitted your form has sought them.
Two limits still apply to the callback. Paragraph (c) bars solicitation of anyone who “has made known to the lawyer a desire not to be solicited”, and any contact that “involves coercion, duress or harassment”. An agent that keeps calling after “stop calling me” is on the wrong side of both, and it is your name on the bar complaint, not the vendor’s.
Is an AI voice call “live person-to-person contact”?
For the outbound case, the Model Rule bans only “live person-to-person contact”, and Comment [2] defines it: “in-person, face-to-face, live telephone and other real-time visual or auditory person-to-person communications where the person is subject to a direct personal encounter without time for reflection”. It then carves out “chat rooms, text messages or other written communications that recipients may easily disregard”.
An AI voice call is real-time and auditory, and the person cannot put it down and think. The only word that gives a firm any room is “person-to-person”, and the text does not say whether a synthetic voice acting for the firm counts as the second person. The comment’s reason for the ban is the “private importuning of the trained advocate in a direct interpersonal encounter” and “insistence upon an immediate response”, which describes an agent scripted to book the consult on this call rather well. I searched the rule and its comments for “artificial intelligence”: zero hits, on a page the ABA dates April 2019. Nothing in it contemplates the agent, so nothing in it excuses the agent.
Speed to lead: the agent calls every lead in under sixty seconds, including the ones we source.
True for the leads who asked. For the leads you sourced, it is a solicitation by real-time voice.
A callback to a request is not a solicitation. A first call to a stranger who needs a lawyer is, and by live contact it is banned.
ABA Model Rule 7.3(a), (b), (c), Comments [1] and [2]; Texas Rules 7.01(b)(2) and 7.03(a)(1), read 28 September 2026.
Texas already answered it
Texas’s current rule closes the gap the Model Rule leaves open. Rule 7.03(a)(1) defines “regulated telephone, social media, or other electronic contact” as any “telephone, social media, or electronic communication initiated by a lawyer, or by a person acting on behalf of a lawyer, that involves communication in a live or electronically interactive manner”. Paragraph (b) bans soliciting employment from a non-client through that contact, with three exceptions that track the Model Rule’s. The question of whether the voice is a person falls away: the firm initiated the contact through whatever tool it chose, and the test that follows is whether the contact is interactive. A conversation is.
Comment 6 explains why written channels survive: regular mail and email “can easily be ignored, set aside, or reconsidered”, and “there is evidence in tangible or electronic form of what was communicated”, citing Shapero v. Kentucky Bar Association, 486 U.S. 466 (1988). That same written route carries a duty. Under 7.03(d)(2), a solicitation communication to a prospective client must be “plainly marked or clearly designated an ‘ADVERTISEMENT’” unless the recipient falls in an exception. So in Texas, as I read it, the outbound agent may text or email a sourced lead with the label, and may speak only once the person has replied and asked for the call. The ABA’s Comment [2] reaches the same split for text messages, and Rule 7.3 itself carries no label requirement.
What the agent says is the other half
Getting the routing right is half the exposure. The other half is the script. Texas Rule 7.01(a) calls a communication misleading when it “omits a fact necessary to make the statement considered as a whole not materially misleading”. On my reading, an intake agent that lets a caller believe they are speaking to a lawyer, or to a human, omits a fact. The agents I build open by naming themselves as the firm’s intake assistant, and I keep that line whether or not a state disclosure statute applies, because the ethics rule reaches the omission on its own.
The agent also must not advise. It can collect the facts of the matter, the dates, the other party and the contact details, and it can book the consult. It cannot say whether the caller has a case or what it is worth, because that is the lawyer’s work and a wrong answer is the firm’s statement. Texas Comment 5 adds the practical reason the intake calls I build are recorded, with consent: the contents of live contact “can be disputed and may not be subject to third-party scrutiny”. A transcript makes them subject to it. The conflict check and prospective-client duties under Rule 1.18 are covered in my intake software review, so I will not repeat them here.
Three moves before your agent dials anyone
Split the agent’s call list by who started it
Inbound leads (form, chat, missed call, a text asking for a call) get the voice callback. Sourced leads (purchased lists, court records, accident reports) never get a first call from the agent; they get a labelled written message, and the agent speaks only when they reply.
Build the do-not-contact list into the agent, not the CRM
Rule 7.3(c) makes “do not contact me” binding from the moment it is said. The agent must hear it, log it and stop, on the same call, and the list must be shared with every other channel the firm uses.
Fix the opening line and the boundary
The agent names itself as the firm’s intake assistant in its first sentence, says the call is recorded, and answers every “do I have a case” with the consult booking rather than an opinion. Read your own state’s version of Rules 7.1 and 7.3 before launch; Texas is stricter than the Model Rule and yours may be too.
The bottom line
An AI intake agent is allowed to do the job firms actually need, which is answer and call back fast. The line is drawn at the first call to someone who never asked, and it is drawn by who started the conversation, not by what made the call.
I have built outbound and intake systems for more than 200 businesses, and the law-firm builds are where I route by that question before I write a single line of the script. If your vendor’s demo showed the agent “reaching out”, send me the call list and I will show you which rows it may call. I am not a lawyer and this is not legal advice; it is what the two texts say today.