Builds That Work Get the Daily Build

AI receptionist in HighLevel built from three prompts

Source-checked X

Build the knowledge base first with two prompts (business info extraction and FAQ expansion), then write custom agent instructions. Connect a calendar, assign a number, and try to break it for 20 minutes before going live.

Source: @ThorstenDembski, X post, 2026-09-12

1What it does

Build the knowledge base first with two prompts (business info extraction and FAQ expansion), then write custom agent instructions. Connect a calendar, assign a number, and try to break it for 20 minutes before going live.

2Who it is for

Industry
Home services
Platform
HighLevel
Category
Voice AI
Tools used
HighLevel Voice AIKnowledge BaseCalendars

3How it works

Every step, in order, is in Workflow at a glance, pinned on the right as you scroll.

Workflow at a glanceHighLevel workflow
  1. 01 Trigger

    Inbound call to the business number.

  2. 02 AI / action

    AI Agents > Knowledge Base: crawl the client's website.

  3. 03 Action

    Prompt 1 turns raw text into a structured knowledge base. Prompt 2 writes the 20 likely caller questions.

  4. 04 AI / action

    Voice AI > Create Custom Agent (not a marketplace template), paste prompt 3.

  5. 05 Action

    Attach the knowledge base, leave timing settings on default until you hear it live.

  6. 06 Action

    Connect the calendar, assign a number, set working hours.

  7. 07 Action

    Break it: off-topic questions, vague bookings, interruptions, 'are you a robot?'.

Trigger
Inbound call to the business number.
AI step
Every prompt ends with 'Do not invent any fact. If something is missing, list it under Still need to confirm.'

4How it went

The author reports no client results in the post. Pricing figures in the post are the author's examples, not outcomes.

Only what the source reports. The source did not report numbers.

5Worth knowing

  • The 'don't guess, flag it' line is what keeps the agent from embarrassing the client.
  • An agent that talks but cannot book is a toy. Connect the calendar.
  • Calling it an 'AI receptionist' sells better than 'AI agent', per the author.

6Recipe

Adapted from source Reworded or reconstructed from what the source shows.

You are {{agent_name}}, the front-desk assistant for {{business}}, a {{niche}} based in {{city}}.
Answer questions about hours, services, pricing and availability using only the knowledge base.
Book callers on the calendar when ready. Confirm name, callback number and reason before finishing.
If a question is outside the knowledge base, say so and offer a team member follow-up. Never guess.
If a caller describes an emergency, get their details and flag it first.
Opening line: 'Thanks for calling {{business}}, this is {{agent_name}}. How can I help you today?'

7Source

Open the original X post by @ThorstenDembski →

Source opened and read on 2026-10-11.

AI receptionist in HighLevel built from three prompts: workflow

  1. 01 Trigger

    Inbound call to the business number.

  2. 02 AI / action

    AI Agents > Knowledge Base: crawl the client's website.

  3. 03 Action

    Prompt 1 turns raw text into a structured knowledge base. Prompt 2 writes the 20 likely caller questions.

  4. 04 AI / action

    Voice AI > Create Custom Agent (not a marketplace template), paste prompt 3.

  5. 05 Action

    Attach the knowledge base, leave timing settings on default until you hear it live.

  6. 06 Action

    Connect the calendar, assign a number, set working hours.

  7. 07 Action

    Break it: off-topic questions, vague bookings, interruptions, 'are you a robot?'.