Most AI receptionists don’t fail because the technology is bad. They fail because somebody pasted the firm’s old voicemail greeting and a list of practice areas into a prompt box and called it done. The voice sounds fine, the system answers every call, and yet the booking rate is worse than it was with a part-time human. Nine times out of ten, the script is the problem.
Key Takeaways
- Open with the firm name and an offer to help in under five seconds; long greetings cause hang-ups.
- Ask the fewest questions needed to qualify and route the call, and ask them in the order a good receptionist would.
- Write out what the AI should say when it doesn’t know something, instead of letting it improvise.
- Define clear triggers for transferring to a human, especially for emergencies, upset callers, and existing clients.
- Review call recordings weekly and rewrite the parts where callers get confused or drop off.
What a Script Actually Is (For an AI)
With a traditional answering service, a script is a set of lines a person reads. With an AI receptionist, it’s closer to a job description plus a set of guardrails. You’re telling the system who it is, what it’s trying to accomplish, what information it needs, what it’s allowed to say, and what it should never do.
That’s a different kind of writing. You aren’t scripting every sentence, because the AI will handle natural conversation on its own. You’re scripting the decisions. Where most firms go wrong is spending all their effort on tone (“warm, professional, friendly”) and almost none on logic.
The Greeting: Get to the Point
Callers decide within a few seconds whether they’re going to stay on the line. A greeting like “Thank you for calling the Law Offices of Smith, Jones and Partners, where we’ve been proudly serving the greater metro area for over 30 years. Please note that this call may be recorded…” loses people before anything useful happens.
Try this instead: “Thanks for calling Smith and Jones. This is the firm’s virtual assistant. How can I help?”
A few things are going on there. The firm name comes first, so callers know they reached the right place. It’s honest about being an AI, which avoids the awkward moment when a caller figures it out on their own (and some states now require that disclosure). And it hands the conversation to the caller right away. If you need a recording disclosure, keep it to one short sentence after the greeting.
Qualifying Questions: Fewer Than You Think
Every question you add is another chance for the caller to get frustrated. Write down the minimum information you need to either book the appointment or route the call correctly, then stop.
For a personal injury firm, that might be:
- What happened, in the caller’s own words
- When it happened
- Whether they were hurt and got medical treatment
- Whether they already have a lawyer
- Their name and best callback number
For a medical practice, it’s usually new or existing patient, reason for the visit, insurance, and preferred appointment times. For an accounting firm, it’s individual or business, what service they need, and whether it’s time sensitive (an IRS notice, for example, is a different call than “I’m thinking about switching CPAs”).
Order matters
Ask the question that determines routing first. If you only take certain case types, find that out before you collect someone’s full contact information and insurance details. There’s nothing worse for a caller than answering eight questions and then hearing “Sorry, we don’t handle that.”
Get the callback number early, too. If the call drops, you want to be able to reach them.
Write the “I Don’t Know” Answers
This is the section most firms skip, and it’s the one that causes the most trouble. Callers will ask things the AI shouldn’t answer: “Do I have a case?” “How much will this cost?” “Is this mole cancerous?” “Can I deduct my boat?”
If you don’t tell the AI what to do with those questions, it’ll try to be helpful, which in a legal or medical setting can mean saying something it shouldn’t. Write explicit responses for the predictable ones:
- Case evaluation questions: “That’s exactly what the attorney will go over with you. I can set up a free consultation. Would tomorrow morning work?”
- Fee questions: Give the answer you’re comfortable with (free consultations, contingency, flat fee ranges), and nothing beyond that.
- Medical questions: “I’m not able to give medical advice, but I can get you in with the doctor. If this feels urgent, please call 911 or go to the nearest emergency room.”
- Anything else outside scope: A polite deferral plus an offer to schedule or take a message.
Also tell it what not to do. Don’t guess at office policies. Don’t promise outcomes. Don’t quote case results. Don’t confirm whether someone is a client to a third party caller.
Handoffs: Decide When a Human Takes Over
An AI receptionist should know its limits. Define the triggers that move a call to a person (or to an urgent callback queue after hours).
Common triggers include:
- Emergencies or anyone describing immediate danger
- A caller who’s upset, confused, or asks for a person more than once
- Existing clients or patients calling about an active matter
- Opposing counsel, courts, hospitals, or other professionals
- High-value leads you want a partner to talk to right away
Write down what the AI says during the transfer (“Let me connect you with someone on our team right now”), and what happens if nobody picks up. A warm transfer that rings into a voicemail box undoes all the goodwill the AI just built.
Booking: Close the Loop on the Call
If the goal is a consultation or appointment, the AI should offer specific times, not ask “When works for you?” Two or three options are plenty. Confirm the date, time, and location (or video link) out loud, then send a text confirmation right away.
It’s also worth scripting a fallback for callers who won’t commit. Something like, “No problem. I’ll have someone from our team follow up with you tomorrow. Is this still the best number?” keeps the lead alive without pushing too hard.
Tone Is Last, Not First
Once the logic works, then adjust the voice. Match the personality of your firm. A boutique estate planning practice and a high-volume PI firm shouldn’t sound the same. Tell the AI to use plain language, keep answers short, avoid legal or medical jargon, and mirror the caller’s pace. Short sentences sound more natural out loud than long ones.
Review Calls Every Week
No script is right on the first try. Pull ten or fifteen recordings a week, especially the ones where callers hung up early or asked for a person. You’ll spot the same problems quickly: a question that confuses people, an answer that’s too long, a transfer that didn’t work.
Fix one or two things at a time and compare the next week’s numbers. Over a couple of months, those small edits add up to a much better booking rate. If you’d like help writing or auditing your AI receptionist setup, Clevr Marketing builds and tunes these systems for law firms, medical practices, and accounting firms.
