The real problem: response time, not traffic
A firm that invests in its website but takes hours to answer a contact form loses the prospect before ever meeting them. According to Hennessey Digital's 2025 study of more than 1,300 US law firm websites, only 25% of firms respond to an online lead form within five minutes (up from 13% in 2021), and one firm in four never responds at all.
Firms that offer a way to get in touch at any time (a form, online booking, a chat) report meaningfully more new matters and revenue than those relying solely on phone calls during office hours, according to Clio's 2025 Legal Trends Report. This isn't a marketing budget problem: it's an availability problem.
- A call missed during a hearing or a client meeting doesn't always come back as a same-day returned voicemail;
- An enquiry received on a Friday evening at 7pm waits until Monday morning at many firms;
- A prospect who contacts several firms overwhelmingly picks whichever one answers first.
The risk nobody talks about: generic AI also recommends struck-off firms
Before discussing chatbots, one fact is worth knowing: in September 2026, access-to-justice charity Blind Justice UK published a study ("AI's Blind Spot") showing that ChatGPT, Claude and Gemini sometimes list, at the top of their answer, firms the Solicitors Regulation Authority (SRA) has fined or shut down. Asked thirty times for the best conveyancing solicitor in Doncaster, all three assistants named first a firm the SRA had just fined. In Maidenhead, Claude recommended a firm closed seven months earlier, according to the Law Society Gazette.
The reason is simple: a general-purpose AI assistant relies on online reviews and search rankings, not on the regulator's disciplinary register. This doesn't directly concern a firm's own website chatbot, but it illustrates a broader rule: an AI tool speaking on a firm's behalf must be built on verified information, not on what a general-purpose model guesses.
What the regulator actually expects from a chatbot
The SRA doesn't ban chatbots. Its own "Risk Outlook: the use of artificial intelligence in the legal market" report even acknowledges that a well-designed chatbot can answer routine enquiries, triage matters by urgency, and offer 24/7 availability. But the same report flags two recurring risks: inaccurate AI-generated answers ("hallucinations"), and confidential client information being entered into tools that don't guarantee its protection.
In practice, a firm's chatbot has to meet the same client care obligations as any staff member: it must not mislead, must clearly disclose that it's an automated system, and must give visitors the information they need to make an informed decision. None of this bans AI; it defines what it's allowed to say.
What a law firm chatbot should do, and must never do
The line is simple to state, harder to hold in practice: the chatbot qualifies and routes, it doesn't advise.
- It can qualify an enquiry (type of matter, urgency, jurisdiction) and offer a booking;
- It can answer general questions about the firm's practice areas, published fixed fees, and typical timelines;
- It must immediately hand off to a human any question that edges toward legal advice on the visitor's specific case;
- It must identify itself as an automated assistant from the first message, not only in the small print.
A chatbot that oversteps this boundary, even unintentionally, exposes the firm to the same client care obligations as if a staff member had answered instead. This is exactly why a law firm chatbot needs to be built on the firm's own verified content (FAQ, fees, practice areas), never on a general-purpose model left free to answer anything.
Checklist before you deploy
- The chatbot identifies itself as automated in its first message;
- It hands off to a human as soon as a question goes beyond simple routing;
- Personal data exchanged is hosted and processed in the EU or UK, never transferred to a third country without an equivalent safeguard;
- Answers rely solely on the firm's verified content, never on what a general-purpose model guesses;
- Someone at the firm can review and correct the standard answers before they go live.
How to get started
Setting this up generally follows three steps: list the 15 to 20 most frequent enquiries the front desk receives, define precisely where the chatbot's role ends and a staff member's begins, then test on a limited volume before opening the tool to all site visitors.
At H'appi, this is exactly the kind of chatbot we build for firms in the UK and France: based on the firm's own verified content, hosted in Europe, delivered in 2 to 3 weeks. Law firms remain a market we're targeting, not yet a client base: today our only named references are Mobilier de France and INnatural. If you run a firm, the first question isn't "which AI should I install," but "which enquiries go unanswered today before the end of the day."