What You Will Learn

• Configure Voice AI agents (LC Phone, voice/model selection, call behavior)
• Build and attach knowledge bases to prevent hallucinations and answer FAQs
• Create Conversation AI bots for multi-channel lead qualification and multi-calendar booking
• Implement actions, triggers, workflows, human handover and outbound compliance
• Test and iterate prompts, transcription, and performance using feedback loops

Study Guide

Introduction: The New Way to Build a Sales Team That Never Sleeps

The way businesses handle sales and marketing is changing. You've probably noticed it. Customers expect instant responses. They want answers at midnight. They want to book appointments without playing phone tag. And they definitely don't want to fill out a form and wait three days for someone to call them back. Enter the concept of AI employees. These aren't just chatbots that spit out canned responses. They're intelligent agents that live inside your GoHighLevel account, and they can handle real conversations. They can talk to leads on the phone. They can chat across Facebook, Instagram, WhatsApp, SMS, and even email. They can qualify prospects, answer questions about your business, book appointments into your calendar, and trigger automated follow-up sequences. This course is your complete guide to implementing these AI employees. We're going to break down the two main types. Voice AI agents that handle phone calls, and Conversation AI agents that handle messaging. You'll learn the exact configuration steps, the pricing structures you need to know about, how to train these agents with your business data, and how to deploy them in real-world scenarios. We'll walk through practical examples, including a full breakdown of how a martial arts academy implemented this system to qualify leads and book intro classes automatically. By the end of this course, you'll have the knowledge to set this up for any business. Whether you're a business owner looking to scale your operations, an agency owner who wants to offer AI implementation as a premium service, or just someone curious about how this technology works in practice. Let's get into it.

Section 1: Understanding the AI Employee Concept

First things first. What exactly is an AI employee in the context of GoHighLevel? Think of it as a digital worker that you assign tasks to. It lives inside your sub-account, which is your workspace within GoHighLevel. This worker doesn't clock out at 5 PM. It doesn't take sick days. It can talk to an unlimited number of leads simultaneously, and it never gets tired of answering the same question about your pricing structure. The core purpose of these AI employees is to execute sales and marketing operations. They handle the repetitive tasks that would normally require a team of human beings. When we talk about the tasks, we're talking about things like communicating with leads and customers, handling general frequently asked questions based on the information you provide, qualifying leads based on specific criteria that you define, booking appointments directly into your connected calendars, and triggering workflows. So if a lead says they aren't ready to buy, the AI can automatically enroll them in a six-week nurture sequence.
There are two primary types of AI employees you need to understand. The first is the Voice AI agent. This is the one that handles telephone communications. It picks up inbound calls when someone dials your business number. It also makes outbound calls to leads who maybe filled out a form but never completed their booking. The second type is the Conversation AI agent. This one handles text-based communications. And this is where things get interesting because it doesn't operate on just one channel. A single Conversation AI bot can simultaneously manage conversations via SMS, WhatsApp, Facebook Messenger, Instagram direct messages, your website's live chat widget, and even email. You can think of these two agents as different members of your team. One is your phone person. The other is your messaging person. And you can deploy them independently or together.

One of the first things you need to understand before diving into configuration is the pricing structure. This catches a lot of people off guard. AI agent usage is not included in your standard GoHighLevel subscription. Whether you're on the $97, $297, or $497 per month plan, your AI usage is billed separately. There are two options. The first is pay-as-you-go. This means you're billed based on how much you use the AI features. This is a good option when you're just starting out and trying to figure out how much usage you actually need. The second option is an unlimited AI employee plan. This costs $97 per month per sub-account. This covers Voice AI, Conversation AI, Content AI, and other AI agents bundled in. So if you're running a business where you expect high, consistent usage, this fixed cost might be the more sensible route. A good approach is to start with pay-as-you-go to gauge your usage and monitor how the agents perform. Then, if you find yourself using it heavily on a regular basis, switch over to the unlimited plan. There are also occasional promotional periods where GoHighLevel might waive AI usage charges for a limited time. So it's always worth checking for current offers.

Section 2: Voice AI Agent Implementation Deep Dive

Now let's get into the specifics of configuring a Voice AI agent. This is the agent that handles phone conversations, and the setup process involves several critical steps. Let's go through them in order.

Prerequisites for Voice AI

Before you can even think about configuring the agent, you need a phone number. And not just any phone number. It needs to be purchased through GoHighLevel's integrated phone system, often referred to as LC Phone. You can access this through the settings in your sub-account. Once you've purchased the number, it needs to be assigned to your sub-account. This number serves a dual purpose. It's the number that inbound calls will ring into, and it's also the caller ID that will be displayed when your AI agent makes outbound calls.

Creating Your Agent and Selecting a Voice

Once your phone number is set up, the next step is creating the agent. In GoHighLevel, you navigate to AI Agents, then Voice AI, and you'll find the Agent List. From there, you can create a new agent. You have two choices here. You can start with a marketplace template that's already pre-configured for common use cases, or you can create a custom agent from scratch. For learning purposes and for full control, starting from scratch is the better route. Next, you'll pick the voice. The GoHighLevel voice library is extensive. You can filter voices by language, accent, and gender. The platform supports dozens of languages, so you can set your agent to speak Dutch, Italian, Hindi, Japanese, and more, beyond just English. This is relevant for businesses that serve multicultural audiences. There's also a custom voice cloning feature. This is a game-changer for brand consistency. If you're the business owner, you can upload a sample of your voice, and the AI agent will basically sound like you. Now imagine having an AI that sounds like you answering every call. That's a powerful way to maintain a personal touch even when you're scaling.

Choosing the AI Model

The underlying model determines the intelligence and the cost of your agent. GoHighLevel gives you several options. You have GPT-4.1, which is typically recommended as the default for its balance of performance and capability. There's also GPT-4.1 mini, which is less powerful but more cost-effective. This might be fine for simple FAQ conversations but might struggle with complex, nuanced booking scenarios. You also have options from Anthropic like Claude Sonnet and Haiku, and Google's Gemini Flash. The choice of model impacts both the quality of conversations and the cost per minute. The language setting is separate from the model. You can set the agent's language independently, which is useful if you want a specific model but need it to operate in a language other than English.

Configuring Actions and Event-Based Triggers

This is where the agent starts to become useful. Actions are capabilities your agent can perform, and they're activated based on events detected during a conversation. Let's break down the most important ones. The first is Call Transfer. This is essential for escalation. You might want to transfer the call to a human agent when the caller requests it. You can also set conditions, like transferring when the caller is identified as an existing paying client. You can configure a pre-transfer message, so the caller hears something like "Please wait while I transfer you to our team." This prevents an awkward silence. The second action is Appointment Booking. This is a core feature. When the caller expresses interest in booking, the agent can select a specific calendar. It will then collect the required information from the caller, which typically includes their full name, email, and phone number. The agent can offer available time slots in natural language, like "We have slots available on Tuesday at 3 PM and Thursday at 5 PM." And if you enable it, the agent can also manage rescheduling and cancellation for existing appointments. Other actions include sending an SMS, which can be useful for sending contact info or a link. There's also the Trigger Workflow action, which adds the contact to an automation sequence based on what happened in the conversation. And for advanced users, there's MCP integration, which is the Model Context Protocol. This expands the agent's capabilities to interact with external systems, though that's a more complex topic. All these actions are configured in the agent builder, and you define the specific event that should trigger them. For example, you might set a trigger that says, "If the user says they want to speak to a human, trigger the call transfer action."

Building and Integrating the Knowledge Base

A knowledge base is what makes your AI agent sound like an expert in your business. It's a repository of information about your services, pricing, policies, location, and other details. Without it, the agent is essentially working blind and will likely hallucinate inaccurate answers. Creating a knowledge base is straightforward. In the AI Agents area, you'll find a Knowledge Base section. You create a new one and give it a name. Then you upload data. The most efficient method is to provide the business website URL. The system will crawl the website and process the information on it. For example, you could paste in `https://thefitnessgym.com`, and the crawler will visit all linked pages. It will learn about the gym's class schedule, the monthly fee, the location, the trainers, and everything else that's on the website. After the crawl is complete, you attach this knowledge base to your agent. But you don't just attach it. You can tell the agent exactly when to use it. In the voice agent settings, you can write an instruction like: "Use this knowledge base whenever the caller asks about pricing, location, services, or operating hours." This gives the agent a clear rule for when to query that information.

Call Settings and Agent Behavior

This is where you fine-tune the conversational quality. There are several settings here, and each one plays a role in making the agent feel more natural. Time Zone is critical. It's used for calculating appointment availability. If you set this wrong, the agent will offer times that are outside of business hours. Maximum Call Time is what it sounds like. AI conversations are usually pretty short, so setting this to 10 minutes is a common recommendation. It prevents the call from going on indefinitely if the conversation starts looping. Idle Time and Reminder Frequency relate to how the agent handles silence. If the caller goes quiet, the bot will wait a certain number of seconds before prompting them. A common setup is to wait 4 seconds, then give a reminder. And you might set it to give two reminders before moving to a termination phase. After a longer period of silence, say 15 seconds, the call will end automatically. Wait Before Speaking is the pause between the caller finishing their sentence and the bot starting its reply. A delay of 0.5 to 1 second feels more natural than instant responses. Response Speed controls how fast the bot speaks. You can keep it at 1 for a normal pace. Dynamic Response Based on User Input is a nice feature. When enabled, the bot will try to match the pace and tone of the caller. If the caller is slow and casual, the bot will slow down. If the caller is fast and business-like, the bot picks up the pace. Interruption Sensitivity determines how the bot responds when it's interrupted. If this is set too high, the bot will get flustered and stop talking. If it's set too low, it might talk over the caller. Finding the right balance is key for a smooth conversation. LLM Temperature is a technical setting. Lower values, like 0, keep the bot consistent and on-track. They prevent it from going off-script. Higher values, like 65 or 80, add creativity. For voice AI, consistency is usually more important than creativity, so a lower temperature is recommended. Backchanneling. This is a powerful feature for realism. When enabled, the bot will interject natural spoken cues like "yeah," "uh-huh," and "okay." This simulates the behavior of a human listener. There's a frequency setting, and you also choose which function words get used. This makes the conversation sound much more authentic and less robotic. Stay Silent on Hold Phrases. This is self-explanatory. If the caller says things like "give me a minute," or "hold on," the bot will stop talking and wait rather than responding.

Transcription and Voice Settings

Transcription is how the AI converts the caller's speech into text. The standard mode works for general conversations. There's also a medical mode that's specifically trained on medical vocabulary, which is useful for healthcare-related businesses. Boosted Keywords are essential for accuracy. If your business uses specific terminology or if you have a complicated business name, you can add those keywords here. This tells the transcription engine to prioritize listening for those words. For example, a business with a brand name like "Straßenbau" would want to boost that term. Pronunciation is another area where you can intervene. If the agent is struggling to say a word correctly, you can provide the IPA (International Phonetic Alphabet) code for the correct pronunciation. A common workaround is to ask another AI tool, like ChatGPT, to provide the IPA format for the word you're struggling with. You then paste that into the agent settings. Voice Settings can add ambient background sounds to the call. You can set the bot to sound like it's in a coffee shop or an office environment. This adds another layer of realism that might make callers forget they're talking to an AI.

Post-Call Processing

What happens after the conversation ends? You have control over that too. You can set up notifications that get sent to admins, specific team members, or custom email addresses when a call is completed. This is useful for keeping tabs on performance. You can also trigger a workflow post-call. So if the purpose of the call was to book an appointment, you could automatically trigger a workflow that sends a confirmation SMS to the customer with the appointment details. There's also the option to include performance summaries in your email digests, which gives you an at-a-glance view of how your AI agent is performing on a regular basis.

Writing the System Prompt

The system prompt is the soul of your AI agent. This is where you define its role, its personality, its objectives, and its boundaries. It's not the time to write a one-liner. This deserves your full attention and some deep thinking about your business. The prompt should define the role clearly. For example: "You are Beth, an AI personal assistant for Dr. Miller's Dental Clinic. Your primary objective is to assist existing patients and answer questions from potential new patients. Your goal is to gather contact information from new callers." It also needs to include handling logic. You need to spell out what the agent should do in different scenarios. What if the caller is asking about a service the clinic doesn't offer? How should the agent respond? What if the caller is complaining? You need to set the boundaries. The prompt also needs to define rules for using actions. You might say, "If the caller expresses an interest in booking an appointment, use the appointment booking action." And you need to structure the call flow script. We're talking about the introduction, how to handle the main part of the conversation, and how to conclude the call professionally. The script might look something like this. Introduction: "Hi, thank you for calling Dr. Miller's. This is Beth. How can I help you today?" Handling: "Ask if they have a specific question or if they want to book an appointment. If they want to book, proceed to gather details and check availability." Conclusion: "Thank you for calling Dr. Miller's. We look forward to seeing you." The more detailed and specific your prompt, the better the agent will perform. You can use external AI tools to help draft and refine this prompt, but you should always customize it for your specific business needs.

Outbound Calling Configuration and Compliance

Making outbound calls is where compliance becomes critical. You can't just turn on outbound calling and start dialing random numbers. There are steps you need to take. First, you have to enable outbound calling in the settings. This triggers an identity verification process. GoHighLevel requires this to comply with anti-spam and anti-robocalling regulations. You'll need to upload some form of identification and potentially verify that you're a real business entity. Second, you have to accept the outbound terms of service. These are important. There's a cap of up to 1,000 calls per day per location. That's a hard limit. Calls are sent at a rate of 1 call per 10 minutes. This is to keep the phone lines stable and avoid being flagged as a spam caller. Calls can only be scheduled between 8:00 a.m. and 8:00 p.m. in the contact's time zone. And a contact cannot be called more than once per day. These terms shape your entire outbound strategy. You can't just blast out thousands of calls. You need to be strategic. There are also specific outbound settings you need to configure. The most important is the Disclosure Message. This is a mandatory automated message that plays at the start of the call. It sounds something like, "Hi, this is an AI assistant calling on behalf of Dr. Miller's Dental Clinic." It's a legal requirement to disclose that the caller is an AI, not a human. This message also needs to inform the recipient of their opt-out options. You should keep this message concise so it doesn't waste the recipient's time. You also need to configure voicemail and IVR detection. If the call goes to an automated directory (like "press 1 for English"), the agent should hang up. If the call goes to voicemail, you have a choice. A common setting is to leave a static voicemail message. This is better than letting the AI try to improvise a voicemail message, which can sound weird. Finally, you need to set the Outbound Call Intent. This is essentially the why behind the call. You explain to the agent why it's calling. For example: "You are calling to follow up with a lead who filled out a form on the website. They wanted to book an appointment but never completed the booking process." This intent guides the entire conversation flow. The agent will use this as the basis for its opening and subsequent questions.

Testing and Fine-Tuning Voice AI

You're going to test your agent, and you should expect it to be imperfect. GoHighLevel provides two ways to test. The first is a web call, which runs the simulation right in your browser. The second is a phone call, where the agent actually calls your real phone number. Initial tests often hit around 80% accuracy. That's normal. Some of the most common issues you'll run into are mispronunciations, like the agent saying "5:00 a.m." when it should be "5:00 p.m." Phone numbers and email addresses often get transcribed incorrectly. The agent might hear a "B" where there was a "P." Interruption handling might be too aggressive or too passive. The key is to iterate. When you spot a specific error, you make adjustments. If the time format is wrong, you add a specific instruction to the appointment booking prompt. Something like, "Always say the time in 12-hour format. For example, say 5 PM, not 5:00 a.m." If phone numbers are problematic, you adjust the transcription settings or add boosted keywords. If the agent is too slow or too fast, you adjust the wait-before-speaking or response speed settings. This is a process of refinement, and it pays off. After a few rounds of testing, the agent becomes reliable enough to deploy.

Section 3: Conversation AI Agent Implementation

Now let's switch our focus to Conversation AI. This is the chat-based agent, and it offers a different set of capabilities. The standout feature is multi-channel operation. One bot can be deployed across SMS, WhatsApp, Facebook Messenger, Instagram, website live chat, and email. The bot reads previous conversation history, so it has context. If a lead messaged the bot on Facebook yesterday and then emails today, the bot remembers the earlier conversation. There are three setup methods for Conversation AI. The first is Guided Prompt Setup. This is a streamlined wizard that's designed for beginners. It limits the amount of control you have, but it's fast. The second is Prompt-Based Setup. This is the recommended method for advanced users because it gives you full control over the bot's personality, behavior, and answer logic. This is what I'll focus on. The third is Flow-Based Builder. This is a visual drag-and-drop interface that's useful for creating rule-based conversations with a fixed decision tree.

Primary Use Cases for Chat AI

Conversation AI is an incredibly versatile tool, but there are certain use cases where it shines. The first is Lead Qualification. This is the highest-impact use case. If you're running ads on Meta or Facebook, you're generating leads. Let's say you get 100 leads a day. Without automation, your sales team would need to call all 100. But with the bot, you can deploy it to interact with those leads automatically. The bot asks them qualifying questions. It filters them. After the bot is done, you might only have 20 qualified leads that your human sales team needs to focus on. That's a massive increase in efficiency. The second use case is Social Conversations. Many businesses are on Facebook and Instagram, but they often struggle to manage incoming messages. The bot can automatically respond to messages on your business pages. It uses conversation history for context, so it doesn't feel like you're starting over each time. This is valuable for local businesses where you might get the same questions repeatedly. The third is 24/7 Support and Recovery. The bot acts as a constantly available agent. It answers questions after hours. It books appointments when your team is asleep. And it catches leads that might have gone cold and recovers them with automated follow-up.

How to Train Your Bot

A bot is only as good as its training. There are three primary methods for training the Conversation AI agent. Method one is Website Knowledge. You provide the business website URL. The system crawls the site and stores all the information. If the site has a page about pricing, the bot will know about it. Method two is Custom FAQs. You manually add common questions and answers. If the website doesn't list a specific detail, like the price of a single session, you add that here. This is essential for filling in the gaps. Method three is the Real-Time Feedback Loop. This is unique and powerful. When you're testing or when the bot is live, you can give a thumbs-down to an unsatisfactory response. The system automatically generates a draft FAQ based on that feedback. You can review it, edit it, and approve it. The system then instructs the bot to use this new information in future responses. This turns the bot into a continuous learning system.

Creating the Bot: Step-by-Step

Let's walk through the actual creation process for a Conversation AI bot. Step one is to navigate to AI Agents, then Conversation AI, and create a new bot. You'll want to choose "Start from Scratch" to get full control. Step two is bot settings. You give it a name. You choose its status. Off means it's inactive. Suggestive mode means the bot prepares responses but a human has to review and approve them before they're sent. Autopilot mode means the bot operates independently and sends messages without human intervention. You also choose your channels. You can select Facebook, Instagram, SMS, Email, Live Chat, WhatsApp, or any combination. The bot settings also include a few Autopilot considerations. Bot Response Delay is important. If the bot responds instantly, it feels robotic. Setting a delay of 10 to 15 seconds makes the interaction feel more human. There's also a maximum messages setting, often set to 75 messages per conversation, which prevents the bot from going on forever. In Autopilot, you decide if it's allowed to respond to images and voice notes. And you should set the bot to sleep when a human agent starts manually messaging or when a workflow triggers a message. This ensures the bot doesn't interfere with other automation.

Defining Goals and Personality

This is the equivalent of the system prompt for Voice AI. It's where you define who this bot is and what it should do. The Personality section defines the bot's character. An example for a martial arts academy might be: "You are a friendly, warm, welcoming conversational AI agent for Legacy Martial Arts. Your primary goal is to build trust, qualify leads, and book them in for introductory classes." It also sets rules. "Always maintain a friendly and professional tone. Mirror the customer's language. Avoid using emojis." The Goals section defines the qualification flow. This is a step-by-step journey the bot should take a lead on. For our martial arts example, the goals might be broken down into steps. Step one: Ask if they live in the local area. If yes, continue. If no, politely end the conversation. Step two: Ask if the class is for them or their child. Step three: Ask for the age of the participant. Step four: Ask if they have any prior experience. Step five: Once qualified, propose booking an intro class and ask for their email and phone number. There are also Critical Guidelines. This is where you add nuance. For example: "Ask one question at a time. Weave the questions naturally into the conversation. Do not sound like a rigid survey." These guidelines are what separate a great bot from a robotic one.

Setting Up Calendars

Before the bot can book appointments, you need calendars configured in GoHighLevel. A calendar represents a specific type of appointment. You might create one called "Intro Class Adults" with a duration of 60 minutes. You might create another called "Intro Class Kids" with a duration of 50 minutes. You set availability windows for each. For example, Adults might be available on Mondays, Wednesdays, and Fridays at 11:00 AM. Kids might be available on Tuesdays and Thursdays at 5:00 PM.

Multi-Calendar Appointment Booking

This is where the magic happens. The bot won't just use any calendar. It will use intelligence to pick the right one based on the conversation. In the appointment booking action, you choose "Multi-calendars." Then you add both the adult and kids calendars. Next, you configure the AI's criteria. You tell the bot something like, "Identify who the introductory class is for. If the contact says the class is for them, use the Intro Class Adults calendar. If the contact says the class is for their child, use the Intro Class Kids calendar." This logic means you're not asking the bot to assume. You're giving it rules for decision-making. You also need to decide on a fallback behavior. What if the bot can't determine which calendar to use? You can either instruct it to send a booking link, or tell it not to book at all and escalate to a human. A critical piece here is data collection. On social channels like Facebook or Instagram, you often don't have the contact's email or phone number. So you need to instruct the bot. "Before booking any appointment, you must collect the contact's email address and phone number." This ensures the appointment is properly associated with a real contact.

Advanced Actions for Conversation AI

Conversation AI is more than just a chatbot. It has a suite of advanced actions that let it perform real business functions. Trigger Workflows is a key one. The bot can add conversations to automation pipelines based on outcomes. If the lead isn't ready to book, the bot can trigger a workflow that adds the lead to a 30-day nurture sequence. This keeps the lead warm without requiring any human effort. Human Handover is the ability for the bot to transfer the conversation to a human agent. This happens when the bot encounters a question it can't handle, or when the contact explicitly asks for a person. Auto Follow-Up Messaging is for unresponsive contacts. If the conversation goes silent, the bot can send a series of follow-up messages at scheduled intervals. It might wait 5 minutes, then send "Hey, are you still there?" and then another after 10 minutes. These messages are AI-generated, so they feel personal. Image Analysis is another feature. Using the underlying AI models like OpenAI or Gemini, the bot can analyze images sent by users. In real estate, the bot could look at a property photo and provide insights. This opens up doors for creative implementations. Voice Note Processing is similar to image analysis. The bot can understand voice notes sent through Facebook or Instagram. This makes interaction much easier for users who might not want to type out a message.

Deploying to Channels

Getting the bot live involves connecting it to your channels. For Facebook and Instagram, you'll go to Settings, then Integrations, and connect your business pages. Once connected, you'll come back to the Conversation AI agent list and set the bot as the Primary Bot. Once it's the primary bot, any message to your Facebook page or Instagram account will trigger the bot to respond. For SMS, the bot will handle text messages sent to your sub-account's phone number. For the website live chat, the bot operates within the GoHighLevel chat widget. And for email, incoming emails to the sub-account's email can be handled by the bot.

Section 4: Real-World Implementation Example

Let's tie everything together with a comprehensive walkthrough of a real scenario. We'll return to the martial arts academy example because it demonstrates all the concepts working together. The business is a martial arts school that offers free introductory classes to attract new students. They want to use AI to handle the influx of leads from their Facebook ads. Their goal for the bot is simple. Qualify leads and book them into an introductory class. They defined the qualification criteria as four questions. Location, do they live in Kerry, NC? Intended participant, is it for them or their child? Age, what is the participant's age? And previous experience, have they done martial arts before? Each of these questions serves a purpose. Location ensures they can actually attend. Participant and age help route them to the right class. Experience helps the instructor know what to expect. The agency implementing this wrote a detailed prompt. They emphasized building the bot with a conversational tone. They didn't want it to sound like a rigid survey. They instructed the bot to ask one question at a time and to reference the knowledge base for any specific questions about class times or pricing. For the calendar system, they created two distinct calendars. The "Intro Class Adults" calendar was set for 60 minutes, available on Monday, Wednesday, Friday at 11:00 AM. The "Intro Class Kids" calendar was set for 50 minutes, available on Tuesday and Thursday at 5:00 PM. The multi-calendar logic was configured. If the contact said "for me," the bot used the adult calendar. If they said "for my child," the bot used the kids calendar. The bot was also instructed to collect contact data. On social channels, the name and email aren't always included in the lead data. So the bot was told to ask for email and phone before booking. They set the bot to Autopilot. This meant it was operating and replying to potential customers without any human intervention. During testing, the bot demonstrated its capabilities. It answered general questions about the school. It asked the qualification questions in a logical order. It processed a voice note from the lead. It extracted the email and phone number from the conversation. And it booked the appointment into the correct calendar for an adult intro class. It wasn't perfect at first. But through iterative refinement of the prompt and calendar settings, the bot's accuracy improved significantly.

Handling Fallback Scenarios and Follow-Ups

Not every lead is ready to book. Some are traveling for a month. Some want to think it over. This is where the fallback workflow comes into play. The agency implemented a scenario for this. If the lead says they aren't ready, the bot collects their contact info regardless. Then it adds the contact to a follow-up workflow. This workflow is a sequence of automated actions. It might wait 5 days, then send an SMS saying, "Hey, we're still here if you have any questions about the intro class." Then it waits another 3 days and sends another message. This continues until the lead either books or specifically opts out. This ensures that even if a lead doesn't book immediately, they stay in the loop, and the sales team gets a second chance to convert them later.

Section 5: Best Practices and Strategic Insights

Based on everything we've covered, there are several strategic takeaways that will help you implement these AI employees more effectively. First, your prompt is everything. The system prompt for Voice AI and the personality/goals for Conversation AI are the heart of your agent. Businesses that fail to invest time in writing detailed, specific prompts end up with agents that sound generic and perform poorly. Businesses that write the prompts well get an agent that feels like a member of the team. Second, understand that events drive actions. The Voice AI agent only performs actions like transferring or booking when it detects an event. You need to configure these triggers explicitly. Don't assume the agent knows when to transfer a call. Tell it. Third, build your knowledge bases thoroughly. The value of your agent is directly tied to the data it can access. Crawl the website, add custom FAQs, and use the feedback loops to continuously improve. If pricing is missing, the agent will guess, and that's bad news. Fourth, compliance is non-negotiable. Outbound AI calls require identity verification, a played disclosure, opt-out mechanisms, and strict adherence to time and frequency caps. Ignoring these rules puts your business at risk. Fifth, plan for testing. Both Voice AI and Chat AI will show small errors in the beginning. The 80% accuracy rate is a realistic starting point. Build testing into your deployment timeline. Iterate, adjust the prompt, tweak the settings, and get to that 95% plus range before going fully live. Sixth, use AI for lead qualification first. This is the highest-impact use case. The bot can filter out the wrong leads, ensuring your human sales team only spends time on qualified prospects. This solves the "too many unqualified leads" problem that plagues many businesses. Seventh, explore the advanced features. Image recognition, voice note reading, multi-calendar booking, and fallback workflows are powerful. They transform a simple chatbot into a true sales assistant. Don't be afraid to implement these advanced pieces.

Common Pitfalls to Avoid

Let's be clear about the mistakes people make, so you can avoid them. The first is rushing the configuration. Jumping into the flow builder or writing a 2-line prompt will result in a poor agent. Take the time to plan out the conversation flow and write detailed prompts. The second is ignoring data collection. On social channels, you often miss out on contact info. If you don't instruct the bot to harvest email and phone info before booking, you'll have appointments without a way to contact the lead afterward. The third is setting it and forgetting it. AI employees aren't a one-time setup. You need to review call transcripts and chat logs. You need to monitor the feedback loops. You need to update the knowledge base as your business evolves. A neglected bot will start performing poorly. The fourth is underestimating the importance of the knowledge base trigger. When you configure the knowledge base usage, you need clear instructions. "Use this whenever the lead asks about pricing." Vague instructions confuse the agent.

Conclusion: Putting It All Together

You now have the full picture of what GoHighLevel's AI employees can do and how to set them up. You've learned about the Voice AI agent, the one that handles phone calls. We walked through the prerequisites, from the LC Phone System number to the billing structure. We covered voice selection, model choice, and the deep configuration of actions and behaviors. We dove into the system prompt and highlighted the compliance steps required for outbound calling. You've also learned about the Conversation AI agent, the one that manages messaging. We covered its multi-channel abilities. We looked at the setup methods, from guided to prompt-based. We went through the training process, using website data, FAQs, and the feedback loop. We broke down the goal setting, the multi-calendar booking logic, and the advanced actions like image analysis and voice note processing. And we walked through the entire example of the martial arts academy to see how all these pieces come together in a real-world deployment. The core takeaway here is that these tools are not about replacing human interaction. They're about handling the volume, the speed, and the consistency. They handle the initial qualification, the routine questions, and the simple bookings. This frees up your human employees to focus on the high-value activities that require empathy and judgment. Things like closing complex deals, managing relationships, and providing service that requires a human touch. The potential for a business that uses this correctly is significant. You have a sales team that works 24/7. You have a response time that's measured in seconds, not days. You have a cost structure that doesn't scale linearly with your leads. Whether you are a local business, an e-commerce brand, or an agency implementing this for clients, the skills you've learned here are transferable. The implementation process is the same across industries. Only the specifics change,the prompts, the knowledge base, and the calendar setup. Now, the best thing to do is to start. Fire up a sandbox sub-account. Experiment with a test agent. Build a knowledge base for a hypothetical business. Write a detailed prompt. Test it. Break it. Fix it. That process is where true learning happens. This isn't just theoretical knowledge; it's a practical skill that you need to practice. So go build your AI employee, and put this knowledge to work.

Frequently Asked Questions

Fundamental Concepts

What is an AI Employee in GoHighLevel?

An AI Employee is an AI-powered agent that lives inside your GoHighLevel sub-account and handles sales and marketing operations autonomously. It can communicate with leads via chat and over the phone, performing tasks such as:

Handling general FAQs, qualifying leads before human follow-up, booking appointments directly into your calendar, managing both inbound and outbound phone calls, and responding to messages across multiple channels (SMS, WhatsApp, Facebook, Instagram, live chat, email).

The AI Employee essentially acts as a frontline representative that engages prospects and customers 24/7, so your human team only interacts with the most qualified leads.

What are the two main types of AI Employees covered here?

There are two core AI agent types:

1. Voice AI Agent - Handles real-time phone conversations. It can receive inbound calls to your GoHighLevel phone number and make outbound calls to leads. The caller hears a natural-sounding AI voice that can answer questions, qualify callers, transfer calls, and book appointments.

2. Conversation AI Agent (Chat AI) - A multi-channel chatbot integrated into your sub-account. It operates through text-based channels including SMS, WhatsApp, Facebook Messenger, Instagram DMs, website live chat widgets, and email. It reads conversation history, answers questions based on trained data, qualifies leads, and books appointments.

Both agents share the same underlying goal: automate and scale your sales and marketing conversations.

How is AI Employee usage priced?

AI Employee usage is not included in your standard GoHighLevel subscription (whether you pay $97, $297, or $497 per month). AI agents have a separate pricing model:

- Pay-as-you-go - You are charged based on actual AI usage, such as the number of calls or conversations handled and the AI model used.
- Unlimited AI Employee Plan - A flat $97 per month per sub-account that covers Voice AI, Conversation AI, Workflow AI, and other AI agents without usage limits.

The recommendation for most users is to start on pay-as-you-go and only upgrade to the unlimited plan once usage increases significantly.

Is there any promotional pricing available?

GoHighLevel periodically runs promotional offers. Notably, the "Summer of AI" offer waived all AI usage charges across all sub-accounts until August 31st of the offer year. If you are evaluating the platform, check GoHighLevel's current promotions, as these offers can eliminate AI costs during your testing and early adoption phase.

How do Voice AI and Conversation AI work together in a unified system?

You can deploy both agents simultaneously to create a complete AI-driven front office. A lead might first message your business on Facebook and get qualified by the Conversation AI bot. If that lead prefers to talk by phone, the bot can capture their number and trigger a workflow that schedules an outbound Voice AI call. Alternatively, a caller who reaches your Voice AI agent can have a summary of the conversation sent to the Conversation AI bot for follow-up via SMS.

The key is that both agents can share the same knowledge base, calendars, and workflow triggers. This means you build the training data once and both agents reference it. Many businesses set up the Conversation AI bot as the primary intake system and use Voice AI for follow-up calls to leads that go cold or require more personal interaction.

What types of businesses get the most value from AI Employees?

Any business that generates leads through phone calls or messaging can benefit, but the highest ROI typically appears in:

- Local service businesses - Gyms, martial arts schools, salons, and clinics that rely on appointment bookings.
- Agencies - Marketing agencies managing multiple client accounts can deploy AI employees across each sub-account.
- Real estate - Qualifying property inquiries and booking showings automatically.
- Healthcare - Handling appointment requests and FAQs while routing sensitive cases to humans.

The common thread: these businesses receive high volumes of repetitive inquiries that follow predictable patterns. AI Employees handle the pattern work so humans focus on complex cases and closing deals.

Voice AI Agent Configuration

What are the prerequisites for using the Voice AI agent?

Before you can configure Voice AI, you must have:

- A phone number purchased through LC Phone System within your sub-account. This is the number that receives inbound calls handled by the agent and is also used as the caller ID for outbound calls.
- An active GoHighLevel sub-account with access to the AI Agents section.

You can have multiple phone numbers and configure different agents to work with different numbers, but at least one is required to begin.

What options are available for the agent's voice?

GoHighLevel provides a built-in voice library with a wide range of options. You can filter by:

- Language (English, Dutch, Italian, Hindi, Japanese, and many more)
- Accent (e.g., British, American, Australian)
- Gender (male, female)

You can listen to each voice before selecting it to ensure it matches your brand tone. Additionally, GoHighLevel supports custom voice cloning, allowing you to create an agent that speaks with your own voice. This is set up through the "Add Custom Voice" option.

Which AI models can power the Voice AI agent?

You can choose the underlying language model that drives the agent's reasoning and responses. Available options include:

- GPT-4.1 (recommended default)
- Claude Sonnet
- HighQ
- Gemini Flash
- Various other OpenAI models

Each model has a different pricing structure, so you should review the pricing before selecting. For most general use cases, GPT-4.1 is sufficient, but more complex scenarios may benefit from models with stronger reasoning capability.