AI Chatbot for Law Firms

Who actually writes to a law firm’s website at 11 p.m.

When we read through a firm’s recent web inquiries and voicemail transcripts during discovery, genuine new matters are usually a minority. The rest is a steady mix of people the intake team has to sort by hand, and every one of them needs a different next step.

• Prospective clients with a new problem, some of whom fit your practice and many of whom do not.
• Current clients who want to know whether the other side responded, when the hearing is, or why their settlement check has not arrived.
• Opposing parties and opposing counsel trying to reach the attorney on a matter, sometimes without realizing they should not be talking to your staff at all.
• Referral sources, such as another lawyer sending over a matter outside their practice.
• Vendors and the court ecosystem: court reporters, process servers, expert witnesses, records companies and software sales reps.
• Job applicants asking about paralegal or associate openings.

A bot that only runs intake forces all of them through a new-client questionnaire: current clients become duplicate CRM contacts, vendors get marketing follow-up, and the visitor who mattered waits behind them. So the first design decision is classification, a short opening exchange that places the visitor in one of these groups and then branches. The branches share almost nothing. A prospective client gets a conflict screen and a consultation offer. A current client gets identity verification and a message to their paralegal. Opposing counsel gets the responsible attorney’s published contact route and nothing else.

Routing across practice areas, offices and attorneys

In a firm with several practice groups, the bot’s job after classification is to figure out which group owns the inquiry and what that group needs to know before it spends attorney time. Each group writes its own short screen. The questions below are the kind we start from; your attorneys edit them.

Keep each screen to four or five questions; anything beyond what the group needs to decide on a callback belongs in the consultation. Multi-office firms also route on the visitor’s location, because a Houston family matter and a Dallas family matter may belong to different people, and on language, so a Spanish conversation reaches the coordinator who handles Spanish intake instead of being translated for someone who must call back in Spanish anyway.

Matters you do not handle deserve a real exit: a plain statement that the firm does not take that kind of case, plus whatever referral path the firm has approved, such as a state bar lawyer referral service.

Why the bot asks for names before it asks what happened

A prospective client who pours out the full story in a chat window may have handed your firm information that creates obligations to them, even if you never take the case. Many states have adopted some version of ABA Model Rule 1.18 on duties to prospective clients, and the practical risk is familiar to every managing partner: learning too much from someone whose opponent you already represent can create a disqualification problem. A chatbot that cheerfully invites long narratives multiplies that risk across every night and weekend.

So the bot’s screen runs in a deliberate order. It collects the visitor’s name and contact details, then the names of the other people and companies involved, then only a category-level description of the problem. It tells the visitor directly not to share details yet. The wording your attorneys approve might read something like this:

Those names land in dedicated fields so staff can run the conflict search in your practice management system before a detailed intake call. Where the software’s API allows it, the bot can run a preliminary name search and flag possible hits for a person to review; it never decides a conflict on its own and never tells the visitor a result. Free-text fields get a length cap and a reminder, because some visitors will type the whole story anyway.

The trade-off is friction. A few prospects will leave rather than answer a question about other parties before they get to explain themselves. Firms that care about conversion sometimes ask us to allow a one-line summary; that is your call, made with your ethics counsel, and the bot will enforce whatever limit you set.

Holding the line between legal information and legal advice

Visitors will ask the bot for legal advice constantly, usually without noticing. The bot has to answer helpfully while never evaluating a particular person’s situation. These are real question patterns and the kind of response we script for each:

• “Do I have a case?” The bot explains it cannot evaluate claims, that only an attorney can after a consultation, and offers to set one up.
• “How long do I have to file?” It says that deadlines depend on the facts and the jurisdiction, that some are short, and that the visitor should speak with an attorney promptly. It never names a number of days or years.
• “What does a consultation cost?” It repeats the firm’s published policy word for word, or says a staff member will confirm. It never improvises a fee.
• “What should I bring to my appointment?” This is fine to answer from a checklist your practice group wrote.
• “Can you send my landlord a letter?” It explains that representation starts only after the firm agrees in writing and routes the request to intake.

The bot also discloses that it is an automated assistant, not a lawyer, and that chatting does not create an attorney-client relationship. Attorneys reviewing the project may find ABA Formal Opinion 512, which addresses lawyers’ use of generative AI tools, a useful starting point, along with any guidance from their own state bar. We build to the written policies your firm adopts; we do not make ethics determinations.

Mechanically, the bot answers only from firm-approved content such as practice area pages, FAQ text and office information, never from the underlying model’s general knowledge of the law, which nobody at your firm has reviewed. Before launch we run it against a large test set of advice-seeking prompts, phrased every way we can think of, and fix every answer that crosses the line.

Current clients: status questions without exposing the file

Status questions interrupt paralegals all day, and they are where a chatbot can help most, as long as nobody can impersonate a client to learn about a matter. The bot verifies identity before saying anything matter-specific, typically by sending a one-time code to the phone or email already on the client’s record rather than trusting whatever the visitor types.

What a verified client may see is a firm decision. Some firms allow the bot to show the next scheduled event on the matter calendar and the name of the assigned paralegal. Others prefer the bot to say only that the message has been passed along. Settlement figures, strategy and anything a lawyer would want to explain in context stay off the table in every configuration we build. The bot’s real value is turning “any update?” into a tidy task on the matter, assigned to the right person, instead of a phone call that interrupts drafting.

Payments are handled by link, not by conversation. If the firm uses LawPay, Clio Payments or a similar portal, the bot sends the client to it; it never collects a card number in chat. Billing disputes go straight to a person.

One more branch deserves care. If a visitor identifies as a party on the other side of a matter, the bot does not engage on substance. Lawyers are generally barred from communicating about a matter with a person they know is represented by other counsel, and a firm’s automated assistant should not be the loophole. The bot provides the responsible attorney’s office line, ends the exchange politely and logs it for that attorney.

Handing conversations to Clio, MyCase, Lawmatics and the rest

Output matters only if it lands where your team already works. Depending on your stack, we connect to tools such as Clio Manage and Clio Grow, MyCase, PracticePanther, Smokeball, Filevine or Lawmatics, plus Microsoft 365 or Google Workspace calendars for consultation booking. For each, the questions are the same: what record gets created, which fields get filled, and what happens when the person already exists.

1. Match before create. The bot searches by email and phone before creating a contact, so a returning client does not become a second record.
2. Structured fields, not a wall of text. Practice area, office, source, language and the conflict names go into their own fields so your reports and conflict searches work.
3. Summary versus transcript. Attach the full transcript, or store only a summary and keep the transcript in a restricted system, as your policy says.
4. Booking rules. Consultations go only into slots each attorney has opened for them, with buffers and daily caps you define.

API access differs between products and sometimes between subscription tiers, so we confirm what your account allows during discovery and name any fallback in the written quote. Our AI integration page covers the plumbing in more depth.

After-hours coverage that knows when to wake someone

The hard part of nights and weekends is not volume; it is deciding which conversations justify a phone call to a lawyer at home. A family member writing at 2 a.m. about an arrest is the classic case for a criminal defense practice. If the firm keeps an on-call attorney, the bot collects the person’s name, the facility, the county and a callback number, then pages the on-call phone by text with a short summary. If the firm does not keep anyone on call, the bot says honestly when someone will call back and does not imply that help is coming sooner. Firms without criminal work rarely need to page anyone; the bot books consultations into open slots and leaves a morning queue sorted by practice group.

Some situations are not legal ones. When a visitor describes immediate danger, a medical emergency or a threat of self-harm, the bot stops the screen and shows emergency information, including 911, before anything else. It is not a crisis service and it says so.

Where the transcripts go and who can read them

People write some of the most sensitive things in their lives into a law firm’s chat window. The design treats transcripts accordingly, while data handling decisions and any compliance judgments stay with your firm and its advisers.

• We choose model providers and account settings under which your conversation data is not used to train the provider’s models, and we show you the relevant terms rather than paraphrasing them.
• Transcripts are encrypted in transit and at rest, access is limited to named roles, and retention follows a period your firm sets.
• Marketing and analytics tools receive counts and categories only, never conversation content.
• The bot’s knowledge base contains only public firm content. No client files, no matter documents.

We claim no certifications or attestations on your behalf and will not tell you a configuration satisfies a particular rule. We document every place data flows so your firm can make that judgment itself.

How EVOTECH builds a firm’s chatbot, step by step

1. Free consultation. A phone or video call with the managing partner and whoever runs intake about practice groups, offices, software and the inquiries that hurt most.
2. Inquiry inventory. Your staff exports a redacted sample of recent web forms and voicemail notes so we count visitor types from your real traffic.
3. Routing tree and scripts. We draft the classification step, each practice group’s screen, the advice-refusal patterns and the after-hours rules as plain documents your attorneys can mark up.
4. Attorney review. Nothing goes live until the responsible lawyers have approved the wording.
5. Build and connect. The bot is built on firm-approved content and wired to your practice management system, calendars and client verification flow.
6. Adversarial testing. We try to make it give advice, reveal matter information, accept a card number and skip the conflict step. Every failure gets fixed and retested.
7. Staged launch. One practice group first, with transcripts reviewed weekly, then the rest of the firm.

Scope and price are set in a written, fixed-scope quote before build work starts. EVOTECH has operated for over 20 years from the Houston area, holds a 5.0 Google rating, and serves firms anywhere in the United States by phone and video. Our AI chatbots page covers the general approach, and firms that want a narrower tool focused only on new-client screening should read about our AI intake chatbot for attorneys.

Related services

Frequently asked questions