How to Set Up Your GHL AI Employee: Step-by-Step Guide

Any client sub-account can have the AI Employee flipped on in under two minutes. Whether it's still answering correctly, booking into the right calendar, and staying quiet about things it shouldn't promise a week later depends entirely on what happened in those two minutes — and in the hour of setup most agencies skip.

This is the hands-on setup walkthrough: the screens, the order, and the checks, for an agency configuring the AI Employee on a real client sub-account. It assumes you've already decided to sell it. If you're still working out what to charge and which offer to package it as, that's a separate question — see how agencies sell and deliver GoHighLevel AI in 2026. This guide is what you do after that conversation, before you hand the client a live bot.

Step 1: Turn on the AI Employee at the agency level

Everything starts with one toggle. In Agency View, go to Settings > Company and turn on the AI Employee add-on. That's the agency-level setting — it unlocks access for each client sub-account afterward, and it runs roughly $97/month per sub-account on the unlimited usage tier.

Do not enable it for every client at once. Pick one client and one narrow, low-risk task to prove it on first:

• After-hours lead replies — the bot only answers when the front desk is closed
• FAQ answers pulled strictly from an approved knowledge base
• Draft replies a staff member approves before anything sends

Set one success signal before you touch anything else — faster first-reply time, or fewer missed leads — something concrete enough to point to in week one. If the use case rides on SMS, sort the number's A2P 10DLC registration before launch too; an unregistered number gets filtered by carriers no matter how good the bot is. That's part of our email & SMS infrastructure work — full detail in the deliverability and A2P guide.

Step 2: Build the knowledge base before you grant access

Access without a knowledge base is how a bot starts improvising — and AI improvisation has a name: hallucination. Before you touch the client sub-account, put a short brief in writing:

• Approved services, exact prices, hours and service locations
• Brand voice, the questions customers actually ask, and what the bot must never claim
• Lead rules — when it qualifies, books, tags, or hands off
• Who gets escalated for refunds, legal issues, or complaints

Only once that brief exists, open AI Suite > Plans & Limits, find the sub-account, and enable AI Agents. Give client staff only the access they need to run it — HighLevel's own permissions let you restrict by role and limit records to Only Assigned Data, which is worth using here.

Step 3: Load the brief into the bot's actual settings

The brief from Step 2 isn't a reference document — it's the instruction set. Put the brand voice and the banned claims into the AI Agent's real configuration, not into a shared doc nobody reopens. Two rules earn their keep on every build:

• The bot never states a price, discount, or timeline that isn't on the approved list. If it's not written down, the answer is "let me get someone to confirm that for you."
• The bot never diagnoses, guarantees an outcome, or takes a legal or medical position on the client's behalf — no matter how the question is phrased.

Every setup that skips this step still ships a fast, confident bot. It's just fast and confident about the wrong things.

Step 4: Wire the booking and escalation logic

A knowledge base answers questions; booking logic is what turns an answer into an appointment on the right calendar. From the same brief, connect:

• Which calendar the bot books into, and what it needs to ask before it offers a slot
• What separates a qualified lead from one that just gets an FAQ answer
• Where escalations actually land — refunds, legal, anything emotional — a real inbox or phone, not a queue nobody watches

Route this through the client's existing pipelines and automations rather than bolting on a parallel system next to them. This is the line between a chatbot and something that actually replaces a step in the client's day — and it's where a lot of setups quietly stop short.

Step 5: Choose rebilling or reselling before you flip the switch

One decision has to be made before the sub-account goes live: who pays for AI usage, and how.

• Rebilling passes HighLevel's usage cost through to the sub-account, with your markup on top — it moves with actual use.
• Reselling sets one fixed monthly price you choose; HighLevel treats it as a normal recurring subscription.

Get the client's written approval on whichever model you pick before enabling access, and don't promise unlimited results — state what's included and who covers overages. That's the operational half of this decision; for what to actually charge a client and how to package the offer, see how agencies sell and deliver GoHighLevel AI.

Step 6: Run the guardrail test before a real customer does

This is the fifteen-minute check that catches what the brief couldn't predict. Before anything goes live on a real phone number or inbox:

• Send it a normal question, an unclear one, and a deliberately urgent or upset one — see what it actually does with each
• Confirm the booking action lands on the calendar, and the contact record updates correctly
• Confirm the human handoff actually alerts a person — not just a log entry nobody checks
• Log every failure, fix it, and run the same questions again

Test five real customer questions before launch. Check the answers, the routing, and the booking actions — not just that it replied.

Step 7: Launch on one client — capped and monitored

Go live on the one client from Step 1, not the rest of the book. Before you do:

• Set a monthly spending cap on the sub-account, and set it to Block AI at the limit rather than let it run past budget
• Put a weekly slot on the calendar to review cost and transcripts — HighLevel lets you cap usage by agency, sub-account, and user, with alerts or automatic blocking when a limit hits, so use it

Neither step is optional. An AI Employee nobody is watching isn't running unattended — it's running unaccountable, on a client's live phone number.

Step 8: Tuning — assign an owner and review weekly

Setup is not a one-time event you finish and forget. Name one person — even part-time — who owns:

• Reading transcripts and patching the knowledge base when the bot gets something wrong
• Approving any change to the brief or the prompt rules
• Reviewing missed or mishandled leads every week, not every quarter

Never launch without an owner for prompt updates, client approval, and missed-lead review.

If your team doesn't have the hours to run this checklist on every client, it's exactly what our AI setup team does for agencies white-label, drawing from your monthly hour block instead of a separate line item.

FAQ: Setting up the GoHighLevel AI Employee

How do I enable the GHL AI Employee for a sub-account?

Turn on the AI Employee add-on at the agency level under Settings > Company, then open AI Suite > Plans & Limits, find the sub-account, and enable AI Agents. Test it with a real lead before rolling it out to other clients.

What's the difference between rebilling and reselling?

Rebilling passes HighLevel's usage cost through to the client sub-account with your markup — it moves with actual use. Reselling sets one fixed monthly price you choose, billed as a normal recurring subscription.

Can I restrict AI Employee access to specific clients?

Yes. Enable AI Agents only on approved sub-accounts, and limit staff roles and data access — HighLevel supports per-role permissions and an "Only Assigned Data" restriction for exactly this.

What do I need ready before I turn the AI Employee on?

A written brief: approved services and exact prices, hours and locations, brand voice and banned claims, lead-qualification rules, and an escalation contact for refunds, legal issues or complaints. That brief becomes the knowledge base and the prompt rules — skip it and the bot fills the gap by guessing.