Set Up a HighLevel AI Employee Steps 1 - 5
The short answer: setting up a HighLevel AI Employee starts by enabling the feature at the agency level (not the sub-account), then creating and naming your agent, building the knowledge base it answers from, and connecting that knowledge base to agent goals like booking and contact capture. The knowledge base is the step that decides whether your agent sounds sharp or generic, so it deserves the most attention. This page covers steps 1 through 5, the core configuration; deployment follows in steps 6 through 12.
The full walkthrough below covers each step with screenshots, the most common reason the feature seems to be missing, and how to keep the agent from giving vague answers. If you want the wider context first, see the cost breakdown or compare platforms on the AI voice agent pricing comparison. Prefer to watch? The video below walks through all twelve steps.
- The AI Employee feature must be enabled at the agency level before any sub-account can use it. Two places to check if it is missing.
- Your knowledge base is the single most important configuration decision. Garbage in, garbage out applies here more than anywhere else.
- The web crawler is the fastest way to populate a knowledge base if your website already has solid content on your services and FAQs.
- Start with knowledge base only in Step 5, then layer in appointment booking, contact capture, and call transfer as you get comfortable.
- A pre-configured snapshot can compress a multi-hour setup into minutes, with a working agent, knowledge base, and workflows already in place.
Step 1: Where the Setup Actually Starts: Agency-Level Access
Most people who hit a wall on their first HighLevel AI Voice Agent setup run into the same problem: they cannot find the feature. The reason is almost always the same. The AI Employee capability lives at the agency level, and it has to be turned on there before it shows up anywhere inside a sub-account.
If you are not sure which view you are in, look for a "click here to switch" prompt near the top of the screen. If you see it, you are inside a sub-account and need to use that switcher to get to agency view. From there, navigate to Settings, then AI Employee, find your sub-account in the list, and flip the toggle on:
If you still do not see the AI Employee option in that settings menu, there is a second place to check: Settings, then Company, then Basic Details, then Features. There is a master-level toggle buried there that controls whether the option appears at all.
It sounds like a minor thing, but this two-minute check saves a lot of frustration. Getting the feature enabled is the real Step 1, and everything else builds from it.
Step 2 & 3: Creating and Naming Your AI Voice Agent with Purpose
Once you are back inside your sub-account, the path to creating a new agent is: AI Agents, then Voice AI, then Agent List, then Create Agent at the top right. You will be given the choice to start from a prebuilt template or build from scratch.
Templates exist for common use cases and can save time once you know what you are doing. For a first build, though, starting from scratch is worth the extra effort. When you wire everything together yourself, the logic of how the knowledge base connects to the goals, and how the goals connect to call-transfer rules, becomes intuitive in a way that following a template does not quite replicate.
Give your agent a name that signals its function clearly. "Sales Assistant" and "Front Desk Agent" are better choices than generic names because they set expectations for anyone on your team who later needs to edit the configuration. Small naming decisions like this matter more as you scale. Then proceed to entering your agent details:
Step 4: The Knowledge Base Is the Whole Game
Step 4 in the setup process is where most of the value is created, and where most under-built agents fall flat. The knowledge base is what your agent draws on when a caller asks a question. It will not speculate, and it will not reach outside the content you provide. That constraint is actually a feature: it means it stays accurate and on-brand, rather than improvising in ways that could mislead a prospect or create liability. The full set of controls that keep it accurate is covered in the agent guardrails breakdown.
You have three options for loading content. The web crawler pulls text directly from URLs you specify. If your website has detailed service pages, FAQ sections, or pricing information, the crawler is by far the fastest option. The Morgan Systems snapshot, which is available as a free starting point, was built using the crawler across 27 URLs, and the resulting knowledge base was comprehensive enough to handle the majority of real caller questions immediately. Manual entry lets you add Q and A pairs and text blocks directly, which is useful for information that does not live on a public webpage. File upload handles PDFs and documents, which works well for rate sheets, menus, policy documents, or anything else that exists in a formatted file.
The content you want to prioritize: your most-asked questions, service descriptions with enough specificity to be genuinely useful, pricing or pricing ranges if you are willing to share them, business hours and location details, and common objections with your preferred responses. A knowledge base built around how your team actually answers calls will outperform one built around what you think callers might ask.
Step 5: Agent Goals: Starting Simple and Building Up
After the knowledge base is built, Step 5 connects it to what the agent actually does. This is where you attach the knowledge base you just created, configure appointment booking and select the target calendar, define which contact fields to collect during the conversation, and set the rules for when a live transfer should happen.
The temptation at this stage is to turn everything on at once. Resist it. Start with knowledge base only. Get familiar with how the agent handles questions in test mode before adding appointment booking. Then add contact capture. Then configure call transfer rules. Each addition changes how the conversation flows, and troubleshooting is much easier when you have introduced one variable at a time.
Call transfer rules deserve particular attention. Defining clearly when the agent should hand off to a human is one of the most important decisions in the entire configuration. A rule as simple as "transfer if the caller asks to speak to someone" is a good start. You can refine it as you learn which calls actually benefit from a live conversation. Once the agent is booking, the full booking automation chain (confirmations, reminders, CRM logging) fires from here.
Why a Knowledge Base Gives Vague Answers, and How to Fix It
The most common complaint after a first build is that the agent answers in vague, generic language. That is almost never a limitation of the model. It is a content problem, and a fixable one. The web crawler pulls whatever lives on the URLs you point it at, marketing copy included, so if you crawl your homepage hero section, the agent learns to talk in slogans. Point it at your substantive pages instead: service detail, pricing, FAQs, policies. Crawl the pages a customer would actually read to get a real answer.
Structure beats volume. A knowledge base of fifteen tightly scoped question-and-answer pairs will outperform one built from fifty rambling paragraphs, because the agent retrieves the closest-matching chunk and reads from it. Phrase your entries the way callers actually ask ("Do you charge for estimates?") rather than the way you would write a brochure. Keep a single source of truth, too. If your business hours appear in three places with three slightly different values, the agent will pick one unpredictably.
Treat gaps as a to-do list. Anything missing from the knowledge base becomes an "I will have someone follow up," so the fastest way to improve coverage is to run real conversations for the first week, note every question it could not answer, and feed those misses back in. A knowledge base built from the questions people genuinely ask will always beat one built from the questions you imagined they might.
When You Cannot Find the AI Employee Feature Anywhere
If the AI Employee option is nowhere to be found, there are three usual causes, and it is worth checking them in this order. First, you are probably looking from inside a sub-account rather than agency view. The feature is switched on at the agency level, so use the agency switcher at the top left to get there before you go hunting for it.
Second, the master company toggle may be off. Even in agency view, if AI Employee is not enabled under Settings, Company, Basic Details, Features, the per-sub-account switch will not appear at all. Turn the master toggle on first, and the sub-account option shows up where you expected it.
Third, the feature can be enabled and still invisible to you because of permissions. If everything is toggled on but you still do not see Voice AI under AI Agents, your user role likely does not include Voice AI access, and an admin has to grant it. That one trips up a lot of people, because nothing looks broken. The feature is simply hidden from your role.
Voice, Language, and the First Message People Rush Through
The agent details (voice, language, and initial message) are easy to click past, but the initial message in particular does more work than almost any other single setting. It is the first thing a caller hears or a visitor reads, and it decides whether they engage or drop off in the first few seconds. Preview a few voices and choose one that matches how your business actually sounds: measured and professional for a law office, warmer and quicker for home services.
Keep the opening line short and pointed. The job of the first message is to say who is speaking and what it can do, then hand the conversation back. It should not deliver a speech. Compare two versions. A rambling opener like "Hello and welcome, thank you so much for reaching out, we are so glad you are here, how may I assist you with your inquiry today?" gives the caller nothing to grab onto. A tighter line works better: "Hi, thanks for calling. I can answer questions about our services or get you booked in. What do you need?" That tells them exactly what is on offer and invites the next step.
Resist the urge to over-script it. The knowledge base carries the detail. The opening line only needs to orient the caller and point them forward. If you find yourself writing a full paragraph, you are doing the knowledge base's job in the wrong place.
The Bigger Picture for Small Business
There is a version of this conversation that frames the technology as a cost-cutting measure. Fewer receptionists, reduced after-hours labor costs, leaner headcount. That framing misses what is actually interesting about this tool.
For a small business, the problem is rarely too many people answering the phone. It is the opposite: calls that go to voicemail during a meeting, prospects who do not leave messages and move on to the next search result, after-hours inquiries that sit in a queue until morning. It addresses a real gap, not a staffing redundancy. That is the same case behind answering every call around the clock.
When the AI Voice Agent answers a call, gathers the caller's name and interest, books them into a consultation slot, and routes the complex questions to your team, it is not replacing anyone. It is catching the lead that would have been lost. Your human staff still handles the conversations that actually require judgment, empathy, and expertise. It handles the intake and triage that nobody had bandwidth for. The person on your team who learns to configure and optimize these tools becomes more valuable as a result, not less, because they are operating with better information and more qualified conversations to work from.
This is what AI augmentation looks like in practice: not a replacement for the front desk, but a layer that makes the front desk more effective.
The Snapshot Option: When You Want It Running Today
For those who want to compress the setup timeline significantly, a pre-configured snapshot is available through the Morgan Systems demo page. Rather than building each component manually, the snapshot installs a working AI Voice Agent, a pre-populated knowledge base built from real-world content, and connected workflows directly into your HighLevel account.
The snapshot is paired with an extended 30-day free trial when you sign up through the demo link, along with live bootcamp training and 90 days of premium support including screen-share. For a business owner who wants to see the system working before committing to the full manual build, it is the fastest way to form a real opinion about whether the tool fits.
The full 12-step process, including chat widget installation, phone number assignment, live call testing, and workflow connections, is documented with screenshots in the setup guide linked above. Steps 1 through 5 covered here are the foundation; Steps 6 through 12 cover deploying the agent, and they are considerably easier once the core configuration is solid.
Frequently Asked Questions
Do I need agency-level access to set up the HighLevel AI Employee?
Yes. The AI Employee feature must be enabled at the agency level before it becomes available inside any sub-account. You toggle it on under Settings, then AI Employee in agency view. If you only have sub-account access, you will need to ask your agency admin to flip that switch first. There is also a master toggle under Settings, Company, Basic Details, Features that controls whether the option appears at all.
What is the knowledge base and why does it matter so much?
The knowledge base is the content your agent draws on when answering questions. It is what keeps the agent accurate and on-brand. Without a well-built knowledge base, the agent has nothing to work with and will either give vague answers or fall back on generic responses. Loading it with your FAQs, service details, pricing, and objection-handling content is what transforms a generic bot into a genuine business representative.
Can I use a prebuilt template instead of building my agent from scratch?
HighLevel does offer prebuilt templates when you create a new agent, and they can save time if you are in a hurry. That said, building from scratch your first time is worth the extra 20 minutes. It forces you to understand how the knowledge base, goals, and call-transfer rules connect, so you know what to adjust later when something is not performing the way you want.
Will this AI Employee replace my receptionist or front desk staff?
No, and that is not the right frame for thinking about it. The agent handles calls that would otherwise go to voicemail or get missed entirely. It answers questions, qualifies callers, and books appointments, then routes complex situations to a real person. Your team handles the high-value conversations; the agent catches everything that would have slipped through the cracks. It is a force multiplier, not a headcount reduction.
What contact information can the AI Employee collect from callers?
Inside the agent goals configuration you can specify exactly which fields to capture: name, email, phone number, service interest, and any custom fields you have defined in HighLevel. The agent gathers this conversationally during the call and writes it directly to the contact record, so your CRM stays current without any manual data entry.
How does the web crawler option work for building the knowledge base?
You provide the URLs you want HighLevel to crawl and the system pulls the text content from those pages automatically. It is by far the fastest way to get a knowledge base populated if your website already has detailed service pages, FAQs, or blog content. The Morgan Systems snapshot was built by crawling 27 URLs, which covered enough content to handle the most common caller questions right out of the box.
What happens if a caller asks something the agent does not know?
The agent stays within the boundaries of your knowledge base, so it will not fabricate answers. When a question falls outside its training content, or when call-transfer rules are triggered, it hands the conversation off to a live person. This is a feature, not a flaw. It means the agent stays reliable and on-message, while your team handles the edge cases that actually benefit from a human touch.