What a law firm can safely hand to the machines

Yes, with a firm line. AI can answer the phone, ask the intake questions, and capture a new matter at 2am on a Saturday. What it should not do is tell a caller whether they have a case, or anything else that sounds like legal advice. Triage and capture belong to the machine. Judgment and advice stay with a lawyer.

That line matters. Cross it, and a tool meant to help your firm can create a client relationship instead. It can also create a malpractice exposure or a bar complaint before an attorney has even seen the file. This article covers what AI intake actually does and where that line sits. It also covers the one detail that shows whether a system was built for a law firm, or bolted on from a generic call tool.

What "AI legal intake" actually means

Intake is what a firm does with every new inquiry, before it becomes a matter. Who is calling, what happened, does the firm handle this, and is there a conflict of interest. Every firm runs some version of it, whether that is a receptionist with a script or a partner taking notes by hand.

An AI receptionist or intake tool is software that answers the phone or a web chat, asks a fixed set of questions, and writes down the answers. It runs on a script the firm approves in advance. It should say plainly, at the start, that it is automated. For the general version of these tools, what they cost, and where they fit against a live answering service or a hire, see AI receptionists: do they work.

Unauthorized practice of law, usually shortened to UPL, means doing a lawyer's work, such as giving legal advice, without a license to practice. Every state defines it a little differently. This article is not legal advice about your state's rule; confirm the exact line with your bar. What follows is the distinction most state definitions share.

Can AI actually do intake

Yes, and it handles the mechanical part well: taking a name and number, asking what type of matter this is, gathering the basic facts, checking availability, and booking a consultation. That is information gathering, not advice. It is the same work a well-trained receptionist already does under a script.

Where it earns its keep is speed. Legal-consumer surveys consistently find that most people shop before hiring a lawyer, contacting more than one firm first, with some putting the share as high as eight in ten. Separate research on lead response reports that firms answering within five minutes convert far better than firms that take half an hour. The odds of reaching a caller at all also drop fast once an hour has passed. Put together, the case is simple: the firm that answers first is often the firm that gets hired. A caller with an urgent problem does not wait for office hours.

Is it unauthorized practice of law

It can be, if built wrong. It is not, if the line is respected. This is not settled law. A New York bill introduced in 2026 would make chatbot operators liable for bot responses that count as UPL when a person says them. At least one lawsuit is underway over a general-purpose AI chatbot allegedly walking someone through legal strategy. That is the world an intake tool operates in, and it is a reason to take the line seriously, not a reason to skip AI intake altogether.

The distinction most states share is between information and advice. Asking "what happened" and recording the answer is information gathering. Telling someone "you have a strong case" is advice, because it applies legal judgment to their specific facts. A well-built intake tool stays on the information side. It asks questions and never answers the legal question back. If a caller asks "do I have a case", the right response is some version of "an attorney will review this and follow up." Never a guess.

What it must never say

A short list, because vague guidance is how firms end up improvising this under pressure:

• Never state or imply whether the caller has a strong case or a weak one.
• Never state a deadline or a statute of limitations. Deadline and calendaring errors are consistently among the most common causes of legal malpractice claims. That is not a place for a script to guess.
• Never estimate a settlement value, a fee, or an outcome.
• Never confirm the firm will take the case. That call is the firm's, and it usually depends on the conflict check.
• Never keep going once a caller needs a real legal answer. Hand off, do not improvise.

The conflict check detail

Here is the piece that separates a tool built for law firms from a generic call-answering product with a legal skin on it: capturing the opposing party's name.

A conflict check is the review every firm runs before taking a new matter, to confirm it does not already represent someone on the other side of the dispute. Running one means knowing who the other side is, not just who is calling. A generic script asks the caller's name and problem, then stops. A script built for a law firm also asks who else is involved, the other party, the other driver, and passes that name along before anyone books a consultation.

The AI tool does not clear the conflict. It should never decide one does not exist. It captures the name, checks it against the firm's matter list if that lookup exists, and flags anything uncertain for a person. The decision stays human. The capture does not have to.

Is after-hours coverage worth it

For intake-heavy practices, usually yes. Personal injury, family law, and criminal defense all see a real share of first contact land outside office hours, often right after the event that made someone need a lawyer. Every call that hits voicemail is likely a caller phoning the next firm on the list within minutes, per the response-time research above.

It is a weaker case for firms running on scheduled, referral-driven work, like estate planning, where a missed after-hours call rarely costs a client. The same question applies here as with any AI receptionist decision: does the arithmetic actually work for your call volume.

Client data and confidentiality

Anything a caller says during intake may touch privileged or confidential information, even before a formal attorney-client relationship exists. That raises the bar on what an intake tool can do with the data it collects, including whether a vendor can use it to train a general model. That question deserves its own answer: see is AI safe with your customers' data for the checklist to run before any intake tool touches a caller's information. The short version: an intake tool for a firm should run on infrastructure the firm controls. The vendor's data use should be spelled out in writing, not a public chatbot with an open-ended privacy policy.

The same discipline covers what happens after intake. A message back to a prospective client, confirming an appointment or requesting a document, should follow the approval pattern in should an AI ever talk to your customers without a human approving. Routine confirmations can run on their own. Anything that reads like legal guidance waits for a person. If a script does slip, when the AI gets it wrong covers who holds that risk.

What we build

We build the intake layer that answers the call or the web chat. It asks the questions a firm actually wants asked, and captures the opposing party's name for the conflict check. It hands everything to a person before a word of advice gets said. Every capture lands in an audit trail your office can review, and any action beyond gathering information waits behind an approval gate.

If your firm's intake is mostly scheduled consultations from referrals, this may not move the needle much for you. The ROI calculator takes about two minutes with your own call volume and tells you plainly either way.