Can AI Help With Law Firm Intake? What it can and can’t do

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

Can AI actually help with law firm intake?

Yes, for one job: answering fast and writing down what was said. Software can acknowledge an inquiry within seconds at any hour, ask the same qualifying questions every time, and hand a written record to a person. It cannot assess a case, quote a fee, or say anything a lawyer would need a licence to say. Treat it as the front desk, not the lawyer.

The reason firms look at this at all is not enthusiasm for software. It is the gap in the published research. When Clio’s researchers contacted 500 United States law firms as prospective clients, only 33% replied to the email, only 40% answered the phone, and 48% did neither.¹ The problem being solved is not sophistication. It is that nobody picked up.

So the useful question is not whether the software is clever. It is whether it covers the hours and the volume where a firm currently answers nothing, without saying a single thing it should not.

What is AI genuinely good at in intake?

Three things: speed, consistency and volume. It replies in seconds regardless of the hour. It asks identical questions of every inquiry, so nothing gets skipped at five o’clock on a Friday. It handles ten conversations at once as easily as one. Those are exactly the three places human intake reliably fails, and none of them require judgement.

Sorting the work by whether it needs a licensed opinion makes the buying decision much simpler than most vendor comparisons do.

What can AI not do in law firm intake?

It cannot exercise judgement, and every real limit follows from that. It must not tell a caller whether they have a claim, estimate what a matter is worth, name a deadline, or decide to turn someone away. Those are lawyer decisions under the professional conduct rules, and a vendor who blurs that line is selling you a compliance problem with a chat window on it.

The rule underneath this is short. ABA Model Rule 5.5(a) says a lawyer “shall not practice law in a jurisdiction in violation of the regulation of the legal profession in that jurisdiction, or assist another in doing so.”⁴ The operative words for anyone buying intake software are assist another. A firm that lets a nonlawyer system hand out legal conclusions is doing the assisting.

We wrote the longer version of that argument, with the line drawn message by message, in the Model Rule 5.5 guide.

Will an automated first reply cost you the client?

Only if there is no way out of it. In Clio’s 2024 research, 51% of prospective clients agreed a chatbot can be a helpful starting point, and three in five said they would use one only if they had the option to switch to a human. People accept an automated first touch. What they refuse is being trapped in one with no exit.

The same research found only 7% of firms were using one at all.² So the market position is unusual: more prospective clients are willing to start in a chat than there are firms offering it, and the objection they raise is not the technology but the dead end.

That is a script decision, not a technology decision. Any tool can meet it. Most default scripts do not, because they are written to keep the conversation going rather than to end it well.

Who is responsible when the software gets it wrong?

The firm is. ABA Model Rule 5.3 requires lawyers with managerial authority to make reasonable efforts to ensure the firm has measures giving reasonable assurance that a nonlawyer’s conduct is compatible with the lawyer’s own professional obligations, and it requires the same of anyone with direct supervisory authority. Buying software does not move that duty to the vendor.

Rule 5.3(c) goes further: a lawyer is responsible for the conduct if they ordered or ratified it, or if they had managerial or supervisory authority, knew of it while the consequences could still be avoided, and failed to take reasonable remedial action.³

In practice this means one person at the firm reads the script, approves it in writing, and owns it. If nobody at your firm can produce the exact text the system sends, nobody is supervising it.

The ABA addressed the AI-specific version of this in Formal Opinion 512, issued on 29 July 2024, which walks through competence, confidentiality, communication and fees as they apply to generative tools.⁵⁶

How should a firm scope an intake tool before buying it?

Ask to see the script, in writing, before signing anything. If a vendor cannot show you the exact words the system will send and let you change them, the tool is not scoped for a law firm. Everything else follows from that one request, and four short questions cover the rest.

• Can we edit every message and approve it before it goes live? If approval happens after launch, it is not approval.
• Does every conversation offer an obvious route to a human, and an obvious way to stop? This is the one thing prospective clients actually asked for.
• Where does the conversation data live, who can read it, and is it used to train a shared model? Intake conversations are confidential from the first message.
• Can we switch it off in one click, without opening a support ticket? A tool you cannot stop is a tool you cannot supervise.

None of those questions are about accuracy benchmarks or model names, and that is deliberate. For intake, the risk is not that the software is not smart enough. It is that it is allowed to say more than it should.