What Your AI Intake Tool Can and Can’t Say to a potential client

Last reviewed 31 July 2026  ·  Regulation and rule text checked against the linked sources on the date shown.

What can an automated intake message safely say?

Four things. That the message arrived, who it arrived at, one or two questions that move the matter forward, and how to reach a human or stop the messages. Everything past those four starts drifting toward advice, advertising or a promise, and each of those is regulated separately and by a different body.

Written out, a compliant first message is unglamorous, and that is the point:

Name of the firm, confirmation, one question, an exit. No assessment, no timeline, no number. It takes about eight seconds to read and it does not require anyone to defend it later.

What must an automated intake message never say?

Anything that contains a conclusion or a promise. No view on whether there is a claim, no estimate of value, no deadline, no prediction of outcome, and nothing that implies an attorney-client relationship has formed. If a sentence would need a licence to say out loud, it does not belong in a script that runs unattended at three in the morning.

• “You have a case.” A legal conclusion, and the firm has not seen a single document yet.
• “This is worth about…” A conclusion and an advertising claim in one sentence.
• “You have two years to file.” Legal advice, and wrong answers here end careers.
• “We will get you compensated.” A prediction of result, restricted in every state.
• “We are now representing you.” No system should be able to form a relationship.
• “We cannot take your case.” Declining is a lawyer’s decision, not a script’s.

The clean way to enforce that list is not a filter that catches bad sentences after the fact. It is a script the system cannot depart from, approved in writing under Model Rule 5.3 by a lawyer at the firm.³ Model Rule 5.5(a) is the reason it matters: a lawyer may not practise law in violation of a jurisdiction’s regulation of the profession, “or assist another in doing so”, and a system handing out conclusions is the other.⁴

Does texting back a web form count as solicitation?

It depends on the state, which is exactly why it needs a state-by-state answer rather than a general one. Under the federal rules an inquiry or application creates an established business relationship for the following three months, which affects the telemarketing definitions in 47 CFR § 64.1200. State bar advertising rules are separate, and usually stricter.

The federal regulation defines a telephone solicitation as a call or message initiated to encourage the purchase or rental of, or investment in, property, goods or services, and excludes calls made with the person’s prior express invitation or permission and calls to a person with an established business relationship.¹ A person who filled in your form and asked to be contacted is not in the same position as a stranger on a purchased list.

That does not settle the bar question. The state rules we have written up in detail are California, Texas and Florida and Arizona, and they do not agree with each other.

Does the first message need an advertising label?

In some states, yes, and the label has to be in the message itself rather than somewhere on your website. Texas requires the word ADVERTISEMENT on covered written communications under its Rule 7.03(c). California has its own labelling requirement for solicitations. Neither is satisfied by a disclaimer the recipient never sees.

The traps are practical rather than conceptual. A label that only appears in the English script and not the Spanish one is a label that does not apply to half your messages. A label that lives in a signature block your SMS provider strips is not present at all. And a state filing requirement does not care that the message was generated automatically.

Each of the three state guides on this site walks the actual rule numbers, the filing obligations and the exemptions. Start with the state your firm is licensed in, not the state the lead is in.

What about confidentiality before anyone is a client?

It attaches from the first message. Under ABA Model Rule 1.18, a person who consults a lawyer about the possibility of forming a client-lawyer relationship is a prospective client, and a lawyer who has learned information from a prospective client may not use or reveal it, even when no representation ever follows.

That is quoted straight from the rule.² It has a direct consequence for script design: every question the system asks creates protected information, so the script should ask for the minimum needed to route the matter and leave the rest to the human.

Name, callback number, language, what happened, roughly when. Not a social security number, not a date of birth, not an insurance policy number, not photographs. The full version of that argument, including what to require of a vendor, is in the confidentiality guide. ABA Formal Opinion 512, issued 29 July 2024, is the AI-specific reading of the same duty.⁵

How should a firm write and approve the script?

In a document, signed off by a named lawyer, before the tool is switched on. Not in a vendor dashboard, not verbally on a call. The approved text is the thing you will be asked to produce if anyone ever queries a message, and a screenshot of a chat window is not it.

• Write every message the system can send, in every language, in one document.
• Have one lawyer approve it in writing and date the approval.
• Add the state label to every covered message, in both languages.
• Test it from an outside number, after hours, before launch.
• Re-read it every quarter, and re-date the approval when it changes.

Firms that do this find the script gets shorter each time rather than longer. That is the correct direction. The measure of a good intake message is not how much it accomplished; it is how quickly a real person took over — which is what prospective clients told Clio’s researchers they wanted, three in five saying they would use an automated first touch only with the option to switch to a human.⁶