Skip to main content
An AI receptionist is four decisions, not one product: how she answers, what she finds out, what happens to the caller next, and who she hands off to. Get those right and the build takes about ten minutes. Get them wrong and you have an expensive voicemail. This page is the whole picture. Follow it top to bottom and you’ll have a receptionist answering your firm’s calls today.

Build it yourself, no code

Lawtte Studio walks you through all four decisions in a browser.

Build it on the API

Wire the same pieces into your own product or intake stack.

​
What you’re actually building

Every call runs the same four stages. An AI receptionist worth having does all four — most “AI answering services” only do the first.
1

Answer

She picks up in your firm’s name, 24/7, and identifies herself as an automated assistant. No hold music, no “press 1.”
2

Intake

She learns what happened, in the caller’s own words, then fills in the specific facts your practice area needs — the date of the accident, whether they’ve been served, whether an insurer already called.
3

Outcome

One thing happens at the end of every call: the intake is emailed to your team, a consultation is booked on a real calendar, or the caller is transferred to a person.
4

Handoff

Everything she learned reaches a human in a form they can act on — a summary in an inbox, a calendar invite with the case on it, or a warm transfer with an introduction.

​
Decision 1 — How she answers

Three things a caller hears in the first five seconds: your firm’s name, her name, and whether she sounds like your firm.
  • Firm name as spoken, not as registered. “Reyes and Cole” beats “Reyes & Cole, LLP, A Professional Corporation.”
  • Her name is the one your clients will repeat back. Keep it easy to say.
  • Voice and languages. Pick the voice, then add every language your callers actually use — she opens in your first language and switches when the caller does.
Say your greeting out loud before you commit to it. Anything that’s awkward to say is awkward to hear forty times a day.

​
Decision 2 — What she finds out

This is where most intake systems fail, and it’s not for lack of questions. It’s the opposite: a list so long the caller feels processed. A good intake has six or seven things she must learn and a dozen she’ll ask only if the conversation goes there. The difference between “always” and “when it comes up” is what makes a call feel like a conversation rather than a form. Four behaviors that matter more than the list itself:
She asks what’s going on, listens, and pulls the facts out of what they already said. Re-asking something a caller just told you is the fastest way to sound like a robot.
Someone detained, a hearing tomorrow, a protective order, a safety concern — she gets just enough to route the call and acts. She does not finish the questionnaire first.
Right after the name, with the reason said out loud: “in case we get cut off.” A dropped call after a five-minute intake with no number is a lost client.
Numbers and emails get confirmed out loud. Transcription is good; it isn’t perfect.
In Lawtte you don’t write this from scratch — each practice area comes with a real intake list you edit. See her intake questions.

​
Decision 3 — What happens to the caller next

Pick one. Firms that try to do all three end up with calls that go nowhere.
If you pick email, point it at a monitored inbox. The most common failure in a new AI intake setup isn’t the AI — it’s perfect summaries landing in a shared mailbox nobody opens until Monday.

​
Decision 4 — Who she hands off to, and when

Two separate things people conflate:
  • Office hours tell her when a human is available to take a transfer. They do not control when she answers — she answers every call, always.
  • Urgent routing overrides your hours. A caller in custody at 11pm either reaches your on-call number or generates an urgent-flagged message, depending on what you chose.
Decide what you want to happen at 2am before you go live, not after the first 2am call. See after-hours coverage.

​
The hard calls

Three calls every receptionist gets, and every one of them is a way to lose a client or embarrass the firm:
  1. “What do you charge?” — the safe default is that she doesn’t discuss fees. If your firm quotes a consultation price or works on contingency, tell her once and she’ll say it once.
  2. “I have a hearing tomorrow.” — flag it urgent, or transfer it. Your call.
  3. A salesperson pretending to be a client. — she screens and ends the call, or points them at an email address.
Whatever else she does, she won’t give legal advice, predict an outcome, quote a fee you didn’t give her, discuss an existing client’s case, or say anything implying an attorney–client relationship. Those guardrails aren’t configurable, and that’s what makes it safe to let her answer every call.

​
Build it in Studio

1

Open the builder

lawtte.ai/studio — pick your practice areas, name your firm, name her.
2

Call her before you sign up

Once she has a name, a Call button appears. Role-play a real client. Free, no account.
3

Answer the four decisions

Intake list, voice and hours, the outcome, and the hard calls — each one is a screen. Full walkthrough: create your AI receptionist.
4

Give her a phone line

Claim a local number, then forward your office line to it or publish it directly. See her phone number.
5

Test, then go live

Three calls — a good new client, a current client, a salesperson. See test her, then go live.

​
Build it on the API instead

If you’re wiring intake into your own product, the same four stages map onto endpoints: Start at the API overview and authentication.

​
Then keep improving her

The build is the easy part. The firms that get the most out of an AI receptionist all do the same unglamorous thing: they read the transcripts.

Read every call for the first week

You’ll find the missing question faster than any amount of planning would.

Change one thing at a time

Adjust a question, then call her about that question.
A receptionist that’s been tuned for two weeks against real transcripts outperforms a perfectly planned one that nobody ever listened to.