Your AI Receptionist

Every business owner has the same problem: they can’t answer every call. They’re with a customer, they’re driving, they’re eating lunch, they’re closed for the day. The phone rings, nobody picks up, and that lead calls the next business on Google.

Voice AI and Conversation AI change that. An AI agent answers the phone, greets the caller, asks the right questions, collects their information, and either handles the request or transfers to a human. A chat agent does the same thing on their website, through SMS, and across social channels. It runs 24/7 and never calls in sick.

This playbook walks you through setting up both, voice and chat, from scratch. You’ll build it for yourself first, then deploy it to a client.

Step 1: Decide What the AI Handles

Don’t deploy everything at once. Pick one channel to start with. For most agencies, inbound phone calls are the highest-impact starting point because missed calls are the pain point every business owner understands viscerally.

If the client’s biggest problem is response time on web inquiries or social messages, start with chat instead. Either way, pick one. Get it working. Then add the second channel once the first one is proven.

The goal of the AI receptionist is not to replace a human. It’s to make sure no interaction goes unanswered. The AI handles the first touch, collects what it can, and hands off to a human when the situation requires it.

No element links. This is a scoping decision.

Step 2: Create the Voice AI Agent

Go to Settings > AI Agents in the sub-account. Click Create Agent. You’re configuring four things: the agent’s name, the voice, the initial greeting, and the direction (inbound to start).

Pick a name that makes sense for the business. Not “AI Bot #1.” Something like “Sarah” or whatever fits the brand. Choose a voice from the library that matches the tone the business wants to project. A law firm doesn’t sound like a pizza shop.

Write the initial greeting. This is the first thing the caller hears. Keep it natural: “Hi, thanks for calling [Business Name]. How can I help you today?” Not corporate. Not robotic. The way a real person would answer the phone.

Where this connects:

• Voice AI: creating and configuring voice agents, voice selection, greeting setup

Step 3: Write the Prompt

This is where the agent gets smart or stays dumb. The prompt is the instruction set that tells the AI what to do, what to collect, what to say, and what NOT to say.

GHL offers basic mode (select data fields to collect) and advanced mode (write a full prompt). Start with basic mode if you’ve never done this. Select the information you want collected: name, phone number, reason for calling, whether they want to schedule an appointment. Basic mode handles the conversation flow for you.

When you’re ready for more control, switch to advanced mode and write the prompt yourself. Define the agent’s personality, the questions it asks, the boundaries it respects, and the conditions for transferring to a human. The prompt should answer: What does this agent do? What does it NOT do? When does it hand off?

The biggest mistake is making the prompt too loose. An agent with no boundaries will make promises, give inaccurate information, and book things it shouldn’t. The second biggest mistake is making it too rigid, scripting every response so tightly that it can’t handle a natural conversation. Find the middle ground: clear goals, clear limits, flexible execution.

No element links. This is prompt engineering. The skill, not a platform feature.

Step 4: Configure the Actions

Actions are what happen during and after the call. This is where the AI connects to the rest of the GHL ecosystem.

Update Contact Fields: The agent collects the caller’s name, email, reason for calling. That data should automatically populate the contact record. Configure which fields get updated and from which conversation data.

Trigger Workflows: After the call ends, trigger a workflow based on what happened. New lead? Route to the sales pipeline. Existing customer with a question? Notify the team. Appointment request? Trigger the booking flow.

Send SMS: Configure the agent to send a follow-up text after the call: a thank you, a booking link, a summary of what was discussed.

Transfer to Human: Define the conditions. If the caller asks for a specific person, if the issue is complex, if the caller explicitly asks to talk to a human, the agent should know when to hand off and how to do it gracefully.

Where this connects:

• Workflow Builder: the post-call automations the agent triggers
• Contact Manager: where the collected data lands
• SMS & MMS: follow-up text messages after the call

Step 5: Set Up Working Hours

Decide when the AI answers and when it doesn’t. There are two common models:

After-hours only: The AI picks up when the business is closed. During business hours, real staff answers. This is the easier sell because it positions the AI as a safety net, not a replacement.

24/7: The AI answers every call. It handles what it can and transfers to a human when needed. This works for businesses that are too busy to answer every call even during business hours, or for solo operators who can’t be on the phone while they’re working.

Configure the working hours in the agent settings. Define what happens outside those hours. Does the AI still answer, or does it go to voicemail?

No element links. This is a configuration decision in the agent settings.

Step 6: Assign a Phone Number

Pick the phone number the agent will answer. Assign it in the agent’s phone and availability settings. This can be an existing number in the sub-account (as long as it’s not already assigned to something like an IVR) or a new number purchased for this purpose.

Once assigned, every inbound call to that number during the configured hours gets handled by the AI agent.

Where this connects:

• Phone & Call Tracking: number assignment and call routing

Step 7: Set Up Conversation AI for Chat

Now add the chat side. Conversation AI handles inbound messages across web chat, SMS, Facebook Messenger, and Instagram DM. Same thinking as the voice agent (what does it answer, what does it collect, when does it hand off) but adapted for text-based conversation.

Conversation AI has two modes: suggestive (drafts responses for a human to review and send) and auto-pilot (responds automatically without human approval). Start with suggestive if you’re nervous. Move to auto-pilot once you’ve seen enough conversations to trust the responses.

Train it on the business’s knowledge base: FAQs, service descriptions, pricing, hours, location. The more context it has, the better it handles real questions without hallucinating or making things up.

Where this connects:

• Conversation AI: configuring chat automation across channels, suggestive vs. auto-pilot modes

Step 8: Deploy the Chat Widget

Add the chat widget to the client’s website or funnel. Go to Sites > Chat Widgets, create or edit a widget, and select the Conversation AI agent. Configure the branding: colors, welcome message, agent name.

For voice-enabled chat, you can also deploy the Voice AI Chat Widget if it’s enabled in Labs. This lets website visitors talk to the AI through their browser microphone instead of typing. It’s the same agent, different interface.

Embed the widget code on the site. Test it by visiting the page and starting a conversation. Make sure the responses make sense, the tone matches the brand, and the handoff to a human works when triggered.

Where this connects:

• Live Chat Widget: the embeddable chat interface
• Voice AI Chat Widget: browser-based voice conversations through the widget

Step 9: Test Like a Real Customer

This is the step most agencies skip, and it’s the one that saves you from embarrassment.

Call the phone number. Talk to the agent. Ask the questions a real customer would ask. Then ask the questions a real customer would ask that you didn’t plan for: “What are your prices?” “Can I talk to someone?” “I have a complaint.” “Do you do [service you don’t offer]?” See how the agent handles the unexpected.

Chat on the widget. Ask weird questions. Give incomplete information. Try to break it. Test the handoff to a human. Does it work? Does the human get notified? Does the conversation context carry over?

Check the back end. Was the contact created? Were the fields updated? Did the workflow fire? Is the conversation logged in the conversations inbox? If anything is wrong, fix it before a real customer finds it.

No element links. This is QA discipline.

Step 10: Deploy to Your First Client

Everything you just built for yourself, replicate in a client sub-account. Adapt the prompt to their business, their voice, their services, their hours. Pick the right voice. Write a greeting that sounds like their brand.

If you’ve been building snapshots, the workflow and configuration templates can transfer. The prompt and branding always need customization. That’s the part that makes it feel like theirs, not yours.

Walk the client through what’s happening. Show them the conversations inbox where they’ll see call transcripts and chat threads. Show them the notification settings so they know when they’re being alerted. Show them how to reach a human when the AI transfers a call.

This is now a service you sell. The AI receptionist answers their phone, responds to their website visitors, and makes sure nothing falls through the cracks. You charge for it monthly as part of your package or as a standalone add-on.

Where this connects:

• Snapshots: packaging the agent configuration for repeatable deployment

The Sequence at a Glance

What This Playbook Does NOT Cover

• Outbound Voice AI calling (requires separate registration and approval, has daily limits and pacing controls. That’s a different use case.)
• Advanced Agent Studio builds with MCP integration and custom LLM nodes (see Agent Studio)
• Self-selling demo agents that role-play as the prospect’s receptionist (future playbook)
• Multi-language voice agent configuration (GHL supports 26 languages but that’s an advanced setup)
• Custom voice cloning through ElevenLabs import
• Third-party voice platforms (VAPI, Retell) for use cases that exceed GHL’s native capabilities

This playbook gives you a working AI receptionist deployed to a real client. Voice and chat, answering calls, handling web inquiries, collecting information, and handing off to humans when needed. It runs without anyone thinking about it, and the client never misses another lead.