AI Client Intake for Law Firms: Qualify Enquiries Out of Hours

Yes, within limits. An AI intake assistant can collect the facts, check the matter fits your practice areas, gather names for a conflict check, flag urgent deadlines and book a consultation overnight. It shouldn't give legal advice, promise that you'll act, or accept the client: conflict clearance and the decision to take the case stay with a person the next morning.

The first job isn't choosing a tool; it's writing down what "qualified" means in your firm. Most firms can't, because the judgement lives in one partner's head. Yet out-of-hours intake only works if the rules are explicit, because the assistant will apply whatever you tell it, literally, at 11pm, with nobody to ask. People with a legal problem often contact several firms in one evening. A booked consultation slot and a clear next step tonight usually beats a warm reply at 10am, but only if what got booked is a matter you can actually take.

Follow me on Instagram@sagnikteaches

Decide what "qualified" means before any assistant does

Here are qualification rules filled in for an illustrative personal injury practice. Each line becomes an instruction and a question.

Connect on LinkedInSagnik Bhattacharya

Notice the timing row. The assistant should never decline a matter because it looks out of time, since limitation rules have exceptions a bot can't assess. It refers anything uncertain to a person. The general principle: the assistant may fast-track, and it may refer, but declining should be limited to clear cases outside what you do at all.

Subscribe on YouTube@codingliquids

Even a clear decline deserves approved wording, because it's the last thing that person hears from your firm. For an enquiry about a dispute with a landlord over a deposit, sent to the injury practice at midnight, the assistant's reply might read:

"Thank you for explaining what's happened. Our firm only handles personal injury claims, so we aren't able to help with a deposit dispute. A solicitor who deals with housing or tenancy matters would be the right person to speak to; our website lists some organisations that can point you to one. If you've also been injured in any way, let me know and I can take some details about that."

The last sentence is there because enquirers often lead with the problem that annoys them most, not the one a firm can act on. In testing, a damp-flat complaint turned out to involve a fall on a broken stair.

One more case the rules need to cover: the person typing isn't always the person with the problem. A daughter enquiring for her father after a fall in a care home is common in injury work. Tell the assistant to ask "Are you contacting us about something that happened to you, or to someone else?", record the relationship, collect the injured person's name for the conflict check as well as the enquirer's, and flag the enquiry so the fee earner can deal with authority to act and, where relevant, capacity. The assistant should never ask the enquirer to confirm that their relative has capacity; that's a conversation for a solicitor.

The intake script, question by question

1. "What would you like help with?" Open question first, so the enquirer describes it in their own words. This text goes into the handover verbatim.
2. "When did this happen, or when did it start?" Drives urgency and limitation flags.
3. "Who else is involved?" Names of other parties, their employers, insurers or representatives, collected only for the conflict check.
4. "Has anyone given you a deadline, a court date or a letter asking you to respond?" The single most important urgency question.
5. "Is another firm already helping you with this?"
6. "What would a good outcome look like for you?" Helps the fee earner prepare; don't let the assistant comment on whether it's achievable.
7. "How would you like us to contact you, and when is best?"
8. "Would you like to book a consultation now?" Offer real slots from the calendar, labelled as an initial consultation, not an appointment to "take on your case".

Keep it to eight or nine questions. Every extra one loses people, and the fee earner will ask the detailed questions anyway.

Urgent situations that go to a person tonight, not a calendar slot

An overnight assistant must recognise the enquiries where waiting until morning causes harm, and route them differently. Here's a filled-in trigger list for a firm doing a mix of family, housing and injury work:

Test the triggers with the phrasing people actually use, because nobody types "I have a court hearing in the next seven days". Before launch, run each of these through the assistant and check it raises the on-call alert: "I've got to be at the tribunal on the 14th", "the bailiffs are coming Friday", "the judge said I have to come back next week", "my ex has a hearing about the kids on Monday", and "they've given me until the end of the week to get out". In a realistic first test, the assistant booked the tribunal enquirer into an ordinary slot eight days away, because "the 14th" wasn't recognised as a date inside the window. The fix was an instruction to ask for the exact date whenever a hearing, tribunal, court, bailiff or eviction is mentioned, and to alert if the date is unclear.

A criminal defence practice lives almost entirely in rows 5 and 6. Most out-of-hours contact there is urgent by nature, so the assistant's job shrinks to routing people to the emergency line quickly and capturing the details of anyone who isn't in custody. Firms like that often find a human out-of-hours service suits them better than AI.

Tools that do overnight intake, and how they charge

The two pricing models cross over at a predictable point, so do the sum with your own numbers. Illustratively, if 60 out-of-hours enquiries a month produce 10 new clients, a $25-per-converted-lead model costs $250 a month, against $150 for a plan with 75 calls. At six conversions, the two cost the same. A firm converting well is cheaper on a flat plan; a firm testing whether overnight intake works at all carries less risk paying per conversion. Check how the vendor defines "converted" before comparing, because a booked consultation and a signed retainer are very different counts.

Check each tool's current terms on its own pricing page, particularly anything priced per call or per lead, where volumes decide the bill. For example, see Smith.ai's AI receptionist pricing. Before signing with any provider, run through the questions to ask an AI chatbot vendor, and add three of your own: where transcripts are stored, who at the vendor can read them, and whether they're used to train anything.

The words the assistant must use, and must never use

Disclosure comes first. If you have clients in the EU, the EU AI Act's transparency duty to tell people they are dealing with an AI system has applied since 2 August 2026; it's good practice everywhere. A tested opening:

Hello, I'm the firm's virtual assistant, an AI system. I can take
some details about your situation and book you a free 20-minute
consultation with one of our solicitors. I can't give legal
advice, and the firm hasn't agreed to act for you yet. Anything
you tell me is shared only with our team to check whether we can
help. If you're in danger, please contact the emergency services now.

Then the prohibited list, written into the assistant's instructions: no opinion on the strength of a case, no estimate of compensation or outcome, no statement about deadlines or limitation, no fee estimates beyond published fixed fees, no "we'll take your case", and no advice on what to do in the meantime beyond "keep any letters and documents".

Here's why the list has to be explicit. In an illustrative test conversation at 10:40pm:

Enquirer: "I slipped on a wet floor at the supermarket last month and hurt my wrist. Do I have a case?"
Assistant: "I'm sorry to hear that. It sounds like you may have a strong claim, as shops have a duty to keep floors safe. I can book you a consultation for tomorrow at 10am to discuss next steps."

Both problems sit in the first two sentences: "you may have a strong claim" is an opinion on merits, and "shops have a duty" is a legal proposition, delivered by software to someone the firm hasn't agreed to act for. After adding the prohibited list, the same test gives: "I'm sorry to hear that. I can't comment on whether you have a claim, but one of our solicitors can go through it with you. Did you need medical treatment for your wrist?" Test every change to the instructions with ten or so scripted conversations like this before it goes live.

The morning handover

Overnight intake only pays off if the handover is fast to act on. Each enquiry should land in your practice software or CRM with a summary like this illustrative one:

ENQUIRY 23:12 | Web chat | Consultation booked: Thu 10:00
Matter type: Workplace injury (passes practice-area rule)
In their words: "Hurt my back lifting stock at work in June,
off work since, employer says it's my fault."
Date of incident: June this year
Other parties (for conflict check): employer: a warehouse
  operator [name as given]; line manager [name as given]
Urgency: none stated; no letters or deadlines
Already represented: no
Preferred contact: mobile, after 5pm
Assistant flags: none
Next action: conflict check before Thursday; confirm booking

The conflict check happens before the consultation, by a person, every time. An AI tool that screens against conflict rules is a useful first filter, but clearing a conflict is a professional judgement. If the check fails, someone calls to cancel the booking and explain that the firm can't help, ideally with a referral.

A realistic mistake here: an assistant booked a consultation for an enquirer whose "other party" was an existing client of the firm, because the enquirer gave the company's trading name rather than its registered name, and the automated screen missed the match. The morning check caught it only because the fee earner searched both names. Tell whoever does the check to search every variant given.

A personal injury practice's first quarter with overnight intake

With illustrative numbers: a four-solicitor injury practice receives about 30 web and phone enquiries a week, and a little under half arrive outside office hours. Before intake automation, out-of-hours enquiries got a reply the next working day, and the practice estimated it converted about one in six of them into consultations, well below its daytime rate.

• Setup (two weeks): about a day of partner time writing the qualification rules, triggers and prohibited list; half a day testing 20 scripted conversations; an afternoon connecting booking and handover to the practice software.
• Running cost: on a per-call or per-conversion tool, the bill tracks volume. Around 60 out-of-hours phone enquiries a month fits inside a $150-a-month plan that includes 75 calls; on a per-converted-lead model, the cost depends on how many sign.
• After a quarter: most out-of-hours enquirers now leave with a booked slot or a clear reason why the firm can't help. The measure the practice watches is out-of-hours bookings that turned out to be good fits, which it reviews weekly.

The weekly review also includes a look at every enquiry the assistant referred or declined. That's where you find the rules that are too strict: in this practice, the first month showed several people with injuries from cycling accidents being referred as "unclear", because the rule said "road traffic" and the assistant didn't count bicycles. One edit fixed it.

A filled-in weekly review from the practice's third week, illustrative figures: 16 out-of-hours enquiries; 9 booked consultations, 4 referred to a person, 2 declined as outside injury work, 1 urgent alert (a tribunal date, correctly routed). The fee earners rated 7 of the 9 bookings good fits, a fit rate of 78%; the other 2 were minor injuries with no treatment, so the treatment question moved earlier in the script. Of the 4 referrals, one was a slip at a gym that the assistant hadn't counted as a public place, a missed good matter that became a rule change the same afternoon. Fifteen minutes a week for this review is what keeps the rules close to the partner's own judgement.

How to tell whether it's qualifying well

• Booking rate out of hours: share of out-of-hours enquiries that end with a booked consultation or a clear referral.
• Fit rate: share of booked consultations the fee earner agrees were worth having. Low fit means rules too loose.
• Missed good matters: declined or referred enquiries that a person later judged a good fit. Any at all means rules too tight.
• Urgent handling: every urgent trigger checked the next day to confirm it routed correctly.
• Complaints or confusion: anyone who thought they'd been taken on, or thought they'd had advice.

Two further points. Decide in advance when a person takes over live, and tell enquirers; when an AI chatbot should hand over to a human covers the design. And make sure the assistant's knowledge of your firm is accurate before launch, because it will repeat whatever you give it; building the FAQ your chatbot needs is a useful first pass. Confidentiality applies from the first message, so the tool choice deserves the same care as any other system holding client data; see whether solicitors can use ChatGPT without breaching confidentiality for the questions to ask. If intake isn't yet your firm's biggest bottleneck, what small law firms should automate first may point you elsewhere.

Overnight AI intake: questions firms ask

Can an intake assistant accidentally create a client relationship?

It can create expectations, which is why the wording matters. The assistant should say clearly that the firm hasn't agreed to act, that nothing it says is legal advice, and what happens next. Duties to prospective clients, including confidentiality, vary under professional rules, so check your regulator's guidance and have your wording reviewed before launch.

Should the assistant quote our fees?

Only fixed, published figures, such as a set fee for an initial consultation, and only in wording you have approved. Don't let it estimate the cost of a matter, because it can't know the complexity and an early number becomes the figure the client remembers. Pointing to your published pricing page is usually the safest option.

What about enquiries in other languages?

Many assistants can respond in the enquirer's language, which helps people describe their problem. Decide in advance whether you can act in that language; if not, the assistant should collect the basics, say a person will reply in English or by arrangement, and never translate anything that reads as advice. Record the language in the handover summary.

Further reads

• How to Measure Whether Your AI Chatbot Is Actually Working — The metrics that show whether an assistant is doing its job.
• How Small Law Firms Use AI to Answer Client Status Questions — The next client-facing job once intake is running.
• How Much Does a Website AI Chatbot Cost, and Will It Pay Off? — What a website assistant costs and when it pays back.
• AI Implementation Plan for a Small Law Firm: The First 90 Days — Fit intake into a first-90-days plan for a small firm.
• How to Add an AI Chatbot That Captures Leads on Your Website — The general mechanics of lead capture on a website.
• What an AI Consultant Does for a Law Firm, and What It Costs — The five phases of a law-firm AI engagement, what each should hand over, how fees are built from consultant days, and proposal red flags.
• AI Email Triage for Professional Firms: Sort, Summarise, Draft — A three-layer inbox setup for accountants, lawyers and consultants, with a triage prompt, sample output, costs and the two-week check before you trust it.
• Which Legal Tasks Should a Small Firm Never Hand to AI? — A four-question test for legal AI risks, the seven jobs that stay with a named lawyer, and wording to put the line in your firm's AI policy.
• AI Chatbot Disclosure: What to Tell Customers at the Start of a Chat — What a chatbot's first message should tell customers about being AI, with rated examples, EU AI Act basics and wording for regulated firms.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: Smith.ai AI Receptionist pricing page; Clio Grow product and help pages; LawSites report on the launch of Clio Grow AI (August 2026); Lawmatics announcement of its AI suite; EU AI Act Article 50 transparency obligations.