AI Lawyer Chatbot Achieves 89% Success Rate for New Website Lead Inquiries
For a personal injury law firm's website, we built an AI intake assistant that triages new visitors by case type, states the attorney-client disclaimer up front, and collects the details the legal team needs before handing off a conversation.
A contact form does not triage a legal question
Most law firm websites still run on a plain contact form or a phone number. A visitor who lands on the site outside business hours gets two options: fill out a generic form and wait, or leave. Neither tells them what happens next, and neither sorts a workplace injury from a traffic ticket from a general question. We cover this same gap in our guide on turning missed calls into revenue with chatbot and live chat setup.
The firm wanted a structured first conversation instead: one that could ask the right follow-up questions, state in plain terms what the conversation was and was not, and hand off an organized record to the intake team.
A structured intake flow, not a single script
The assistant runs as a set of connected flows inside SendPulse's automation builder. A filter step reads the visitor's language, a menu narrows the case type, and a shared contact capture flow closes every conversation the same way.
Language routing before the first real question
The moment a new visitor opens the chat, a filter checks their stated language and branches into English or Spanish before any case-type content loads. This is the same pattern we use across our AI assistant chatbot builds for multilingual service businesses.
A separate menu in Spanish, not a translation layer
The Spanish flow runs its own welcome message and its own case-type menu (injury, accident, traffic ticket, or another legal issue) and the platform tracks each option's click count on its own.
One shared finish line for every case type
However a conversation starts, once flagged as an injury, an accident, or a traffic ticket, it funnels into a single flow called Contact Capture, Consent and Scheduling. That flow asks for a name, then a phone number, then explicit consent to text appointment details.
It is also the source of the 89% figure in this case study, which the Results and Analysis sections below cover in detail.
The numbers, from the platform itself
Every figure below comes straight from the SendPulse dashboard for this bot, covering the ten-month window from November 1, 2025 through August 31, 2026.
What the assistant says to visitors
These are unedited exchanges from the live chat platform. Names, the firm's domain, and other identifying details are removed or blurred.
A person stays in charge
Every conversation carries controls that staff can use from the same screen: pause the automation for an hour, review the pause history, or restart a specific flow. The assistant runs the first conversation. People stay in charge of the rest.
This is the same approach we build into every chatbot and live chat setup: automation handles the structured first response, and a human can step in at any point.
Where the 89% figure comes from
The number traces back to one flow: Contact Capture, Consent and Scheduling. It is the step every qualified conversation reaches, regardless of case type, and it runs three questions in a fixed order.
37 conversations reached the first question. 36 continued to the phone number question. Of those 36, 32 replied yes to text consent and 2 replied no, accounting for 34 of the 36. 32 divided by 36 equals 88.9%, which rounds to the 89% reported for this case study.
What "success" measures here. It is the completion rate of that last step, among conversations that already made it deep enough to be asked for a phone number. It is not the share of all website visitors who become a lead. Across the same window, 37 of 203 total conversations reached this stage. Both numbers are real. They answer different questions, and the 89% reports the first one.
The flow has been refined, not left alone
The bot structure includes earlier versions of the injury-severity and contact-capture flows, now marked inactive and replaced by the versions measured here. The 89% reflects the current flow, not its first draft.
Why the disclaimer runs first
This is not just a compliance formality. The American Bar Association's Formal Opinion 512 (2024) lays out how a lawyer's existing duties, including competence, confidentiality, and communication, apply once a firm brings AI tools into its practice. Model Rule 1.18 already covers what a firm owes a prospective client from the first conversation onward. That is why the disclaimer runs before the first question, not after.
What this case study shows
- Disclaimer first, always: every visitor hears the attorney-client statement before being asked for a single detail, in English or Spanish.
- One shared finish line: every case type (injury, accident, traffic ticket) funnels into the same contact capture flow, so intake gets a consistent record no matter how the conversation started.
- A measured result, not an estimate: the 89% figure traces to a specific flow and a specific pair of numbers (32 and 36), not a marketing estimate.
- Human control at every step: staff can pause, take over, or restart any conversation from the same dashboard the bot runs on.
Want a system like this for your firm?
We build AI chat assistants for law firms and other local service businesses that need a real first response, not just a contact form.
Not a law firm? The same approach runs through our general AI assistant chatbots service. If you also need more of these visitors finding the site, our legal and law firm SEO services cover that side of it. More client work is in our case studies.
