In 2024, a research company acting for Clio rang 500 law firms posing as prospective clients. Forty per cent of those calls were answered by a person. Including callbacks, the shoppers reached 52% of the firms, meaning nearly half were effectively unreachable by phone even after being given the chance to respond to a message. Of the firms that missed a call, one in five rang back.
In 2019 the same exercise found 56% of calls answered. Answering rates went down.
This guide covers what an AI receptionist does for a law firm, which intake calls it should and should not handle, and the professional obligations that make legal call handling different from a plumber's or a dentist's.
Why legal intake breaks
A prospective client with a legal problem is rarely calm and rarely well informed. They usually do not know which area of law their problem falls under, whether they have a case, or what it will cost. They frequently ring several firms in one sitting.
Fee earners cannot answer the phone. Someone in court, in a client meeting or drafting cannot break off, and the billing model gives them no reason to. So the call goes to a shared line, a receptionist juggling three other things, or voicemail.
Clio's study measured what happens next. Among prospective clients who did reach a firm, 73% said they were unlikely to recommend it. But among those who spoke to someone directly by phone, 39% said they would recommend the firm. Reaching a person changed the outcome a great deal.
The same study found firms are poor at the conversation even when they have it. Only 41% offered any rate information, 12% could estimate a total cost, and 36% explained the process and next steps. On websites, only 30% of shoppers could easily understand how to hire the firm and 14% could find pricing.
What software handles well in a law firm
Answering every call immediately, including out of hours. Legal problems surface in the evening and at weekends: after an arrest, after a letter arrives, after an accident. A firm answering at nine on a Friday evening is competing against firms that are not.
Capturing intake information consistently. Name, contact details, area of law, brief description, how they found you, urgency. Recorded the same way every time, which is more than most manual intake achieves.
Initial conflict screening. Taking the names of other parties so a conflict check can be run before anyone speaks substantively to the caller. This is one of the strongest arguments for structured intake: it stops fee earners having conversations they should not be having.
Booking consultations straight into the fee earner's diary, with confirmation sent while the caller is on the line.
Answering procedural questions. Opening hours, location, parking, which documents to bring, which areas of law you cover, whether you offer fixed fees. Factual, repetitive, and safe.
Filtering. Sales calls, recruiters and recorded messages, which take up a disproportionate share of a law firm's inbound volume.
What it must not handle
This is where legal work departs from every other sector, and where a supplier's standard product needs constraining.
Anything resembling legal advice. An AI receptionist must not tell a caller whether they have a case, what their chances are, what the limitation period is, or what they should do next. In England and Wales, reserved legal activities are restricted, and a caller who acts on something your phone system told them is a problem regardless of the disclaimers on your website. Configure the system to refuse, not to hedge.
Fee quotes beyond published ranges. Publishing a fixed-fee range is fine. Estimating what a particular matter will cost is not something software should attempt.
Distressed and vulnerable callers. Family breakdown, bereavement, criminal allegations, immigration detention, safeguarding. These need a person quickly. The system has to recognise the signals and escalate, and you have to decide in advance what those signals are.
Anything with a deadline attached. A caller in police custody, facing a hearing tomorrow, or with a limitation date expiring cannot be put in a queue.
Confidential detail beyond the intake minimum. Take enough to run a conflict check and route the matter. There is no reason for a phone system to be collecting a detailed account of the facts, and every reason not to.
Professional obligations to work through first
Confidentiality. Your supplier will hold call recordings, transcripts and intake notes containing privileged information. Establish where that data sits, how long it is kept, who at the supplier can see it, and whether it is used to train models. Get the answers in the contract.
Data protection. Recording calls and storing transcripts brings UK GDPR duties. You need a lawful basis, a retention period, and a data processing agreement. Legal enquiries frequently include health information and criminal allegations, which are special category data with a higher bar.
Client care duties. SRA requirements on service and information apply to how enquiries are handled, not only to retained clients. If your intake process misleads a caller about cost or process, that is your exposure.
Insurers. Some professional indemnity insurers ask about automated client intake. Check before deploying rather than after.
Supplier security. Ask for certifications and access logs rather than accepting general reassurance about encryption. Encryption at rest is table stakes. Who inside the supplier can listen to a recording, under what circumstances, and whether that access is logged, is the harder question.
Where AI-only falls short
Clio's data makes the case better than any argument. Fifty-one per cent of clients found automated tools useful for working out their legal options. Sixty-seven per cent still wanted the ability to speak to a human when they needed one. And the recommendation gap, 73% unlikely to recommend overall against 39% willing to recommend after speaking to someone by phone, points the same way.
A law firm's inbound calls split roughly into three groups. Procedural questions, which software handles well. Straightforward new enquiries, which software can capture and book. And calls where the person on the line is frightened, angry or facing something urgent, where the handling determines whether they instruct you, and where software should be doing nothing more than getting a person on the line fast.
Firms should be wary of two sales pitches: that AI can run legal intake unsupervised, and that AI has no place in a law firm's call handling. Clio's answering rates suggest the second position is being held by firms answering 40% of their calls.
Setting it up
Start with your own recordings. Listen to a fortnight of intake calls: what gets asked, what gets missed, where callers give up. That is your specification, and it beats any supplier's template.
Write the escalation triggers down. Which words, subjects and tones of voice put a call straight to a person, and to whom. Be specific: family law, police, court date, limitation, audible distress. Ambiguity here causes the failures that end up in complaints.
Restrict the system explicitly. Tell it what it must not answer, not only what it should, and test that it holds the line when pushed. Penelope works within set guardrails and states plainly when something falls outside them rather than improvising, which is the behaviour you want on a legal line.
Then test with awkward calls before you point your main number at it. Have colleagues ring in as a distressed family client, a caller with a hearing tomorrow, and someone asking whether they have a case. How the system behaves on those three tells you whether it is safe to go live. Review transcripts weekly for the first month: every failure is a configuration instruction you had not thought of yet.