How a Small Law Firm Fixed Client Intake Without Risking Confidentiality

Daniel's first reaction was the one most lawyers have: the caller would have been a difficult client anyway, and the firm was busy enough. His second reaction, a week later, was less comfortable. He went back through the firm's voicemail log for the past month and counted. There were more of those calls than he had thought, and he had no idea how many of them had hired someone else.

He is not alone. In its 2024 Legal Trends Report, Clio described a secret-shopper study of 500 U.S. law firms. Only 40% of the firms answered the phone, down from 56% in a similar 2019 study, and 48% were essentially unreachable by phone even after the shoppers left messages. Just 33% replied to emails. Clio sells legal software, so it has a reason to highlight the problem. Even so, the numbers matched what Daniel found in his own voicemail log.

The reason was not laziness. The receptionist worked mornings. The paralegal was usually in the middle of a filing. Daniel and his partner were in court, in meetings or on the phone with existing clients. Prospective clients called in the evenings and on weekends, when they finally had a moment to deal with the worst thing happening in their lives. Nobody was there.

Daniel's first instinct was to buy the cheapest answering tool he could find. His partner, who had sat on a state bar ethics committee, stopped him. Intake for a law firm is not like intake for a plumber, and the differences matter.

Prospective clients are owed confidentiality too

Under ABA Model Rule 1.18, a lawyer owes duties to a prospective client even if the person never hires the firm. Information learned in an initial consultation generally has to be kept confidential, and in some cases it can disqualify the firm from representing someone else in the same matter. A careless intake system that invites a caller to tell her whole story to a machine creates two problems at once: confidential information in a vendor's system, and a possible conflict the firm did not know it had.

AI tools fall under the same duties

In July 2024 the ABA issued Formal Opinion 512, its first ethics guidance on generative AI. It says the existing duties of competence, confidentiality, communication, supervision and reasonable fees all apply when lawyers use these tools. In plain terms, a lawyer who uses an AI assistant has to understand roughly how it works, protect what goes into it and supervise what comes out of it, the same way they would supervise a new paralegal.

The assistant cannot give legal advice

Nothing the assistant says can sound like legal advice, a case evaluation or a promise of representation. "You have a strong case" is out. So is "We can help you with that," if it suggests the firm has accepted the matter. Daniel's partner put it simply: "The machine can take a message. It cannot be a lawyer."

We started by writing two lists with Daniel and his partner. The first was the work that could be handled by an automated assistant. The second was the work that had to stay with a person. The second list mattered more.

What the assistant handles

• Answering the main line after hours and when the receptionist is busy, with a greeting that says it is an automated assistant and that nothing said creates an attorney-client relationship.
• Asking a short, approved set of questions: the caller's name and contact details, the general practice area, the names of the other parties for a conflict check, and any court date or deadline the caller mentions.
• Telling the caller, in plain words, not to share the details of their situation yet, and explaining that a person from the firm will follow up.
• Sending the intake owner a short summary, with a flag on anything that mentions a hearing date in the next seven days.
• Reminding the intake owner if no one has replied within two working hours.

What stays with people

• Running the conflict check and deciding whether the firm can take the matter.
• Every conversation about the facts of the case, fees and strategy.
• Any call where the caller mentions violence, a threat to a child or an arrest. Those go straight to the attorney on call, or the caller is given emergency information right away.
• The decision to accept or decline a matter, and the letter that says so.

That last group is where the firm's reputation lives. The assistant's job is to make sure those conversations happen sooner, not to have them.

The build took about a month. Most of that time went into words, not technology.

First, Daniel's partner wrote the intake notice: the exact sentences the assistant says about automation, confidentiality and the absence of an attorney-client relationship. She reviewed every approved answer the same way she would review a letter going out under the firm's name.

Second, we mapped the data. We wrote down every field the assistant would collect, where it would be stored, who could see it and how long it would be kept. The firm chose a setup where the summary goes into its existing practice-management system and nowhere else. The voice platform keeps call records only as long as the firm's retention policy allows, and staff access is limited to the people who do intake. Daniel's partner reviewed the vendor terms for confidentiality and data use before anything was signed.

Third, we tested the hard calls. A caller who starts telling the whole story despite being asked not to. A caller who says she is afraid to go home. A caller who asks, "Can you just tell me if I'll get custody?" A caller from the other side of an existing matter. A caller who speaks mostly Spanish. Daniel and his partner listened to each test call and marked what to change. The assistant now interrupts politely when a caller starts sharing details, and it moves any safety concern to a person immediately.

Finally, we tested the boring parts too. Does the summary arrive? Does the two-hour reminder fire? Does a deadline flag stand out in the inbox? These are the parts that quietly fail in many systems, so we checked each one with the firm watching before go-live.

We do not promise more clients or more revenue, and the firm did not get a miracle. What it got was a process it could see and trust.

Evening and weekend callers now reach a polite assistant instead of a voicemail box. Each call becomes a short, structured summary with a conflict-check list attached. The paralegal runs the conflict check before anyone calls back, so Daniel no longer has to take a call and then discover the firm cannot help. The two-hour reminder means that a Friday morning message gets a reply on Friday morning.

A few weeks after launch, a caller left a message at 9:20 on a Sunday night about a protective order hearing on Wednesday. The summary was flagged. Daniel's partner called her at 8:05 on Monday morning, ran the conflict check on the way to the office and met with her that afternoon. Daniel told us that was the call he kept thinking about. Not because the firm won anything, but because this time someone answered.

Not everything worked on the first try. Some callers found the "please do not share details yet" message cold, so the firm softened the wording: "We want to hear your story, and an attorney will do that with you directly. For now, I just need a few details so the right person can call you back." And the firm decided that calls about existing matters should skip the assistant entirely and go to the paralegal's voicemail, because those clients expect to reach someone they know.

It depends on where your calls are lost. If your receptionist answers nearly every call during business hours and few calls come in after hours, a better voicemail greeting and a same-day callback rule may be all you need. If calls pile up in the evenings and on weekends, or your team is regularly in court when the phone rings, structured intake can help.

Before you decide, look at a few things:

• How many prospective-client calls went to voicemail last month, and how many got a reply the same day?
• Who owns intake at your firm, and what happens when that person is out?
• Have you written down the questions intake is allowed to ask, and the ones it is not?
• Which vendors would see prospective client information, and have you reviewed their confidentiality and data-use terms?
• What should happen when a caller mentions safety, an arrest or a hearing in the next few days?

If you cannot answer the first one, start there. Most phone systems will show you missed calls by hour of the day, and that one report usually settles the question.

We are not a law firm, and we do not interpret professional rules for you. Your attorneys and ethics counsel make those calls. What we do is build the intake workflow so that the decisions your firm made are the ones the system follows, and we test it with you until you trust it. You can read more on our AI automation for law firms page.

• A written scope. The approved questions, the intake notice, the handoff rules, the deadline flags and what will be working at handover.
• Confidentiality by design. We map every field, every vendor and every person with access, and we keep case details out of the automated path.
• Hard test calls first. You hear how it handles the difficult callers before any real prospective client does.
• Accounts in your name. Phone numbers, platforms and records belong to the firm.
• Honest limits. It does not give advice, accept matters or replace your intake staff. It makes sure the right person hears about a call in time to do something about it.

Daniel still thinks about the woman who called at 6:47 on a Thursday. His firm did nothing wrong in the narrow sense. The phone rang after hours, and nobody was there. Now someone is, or at least something that knows its limits, takes down what matters, tells the caller what happens next and makes sure a person follows up first thing.

If your firm has its own version of that voicemail, book a short intro call. Please do not include client names or case details when you reach out. We will look at how calls reach your firm today, show you what careful intake could look like and tell you plainly if a simpler fix would do.