Technologies
How do you create an AI agent in GoHighLevel?

Cristian Cabrera
Published · 8 min

To create an AI agent in GoHighLevel, go to AI Agents > Conversation AI, create a bot based on instructions or on a visual flow, train it with your website, your FAQs and your commercial terms, define its goal and when it hands off to a person, and test it in suggestive mode before switching to autopilot.
The agent replies on WhatsApp, Instagram, Facebook, website chat, SMS and email, and there is also a voice version for phone calls. It is not included in the GoHighLevel plan: you pay for it with an AI Employee plan or by usage. In this guide we show you how to build it step by step according to HighLevel's documentation (updated in November 2026), how we use it in real stores at We make it Lab and how much it costs.
AI agent or chatbot: what is the difference
A rule-based chatbot follows a fixed menu and gets stuck when something unexpected comes up.
An AI agent understands the question as people actually write it, looks for the answer in the information you loaded and can run actions: book an appointment, trigger a workflow, move an opportunity in the pipeline or hand off to a person. In GoHighLevel, that is Conversation AI. If you are still wondering whether your business needs one, read do I need a chatbot in my business?
How to create the agent step by step
1. Define a single job
Write in one line what it has to achieve, for example: "answer product, price and shipping questions on WhatsApp and take the customer to the purchase link". An agent that handles everything answers everything badly.
2. Choose the bot type
HighLevel's setup guide offers two methods:
- Prompt Based: you write the personality, goals and rules. It is the one we use for stores: it answers open-ended questions and lets you choose the AI model.
- Flow Based: you build the path in a visual editor with branches and steps. It fits when you need to qualify a lead with several questions in order.
3. Train it with real information
In Bot Training you load what the agent can use to answer. Our method has three layers:
- The website pages: products, shipping, returns, FAQs. The crawler reads them on its own.
- FAQs loaded by hand for what cannot be answered wrong: payment methods, installments, shipping times, warranty.
- Documents with commercial terms, current promotions and, if the catalog is large, a product table with price and stock exported from the store.
4. Write the instructions
What has the biggest impact on answer quality:
- Brand tone: in Argentina, the informal "vos" form, short sentences, one question per message.
- What to do when it doesn't know: "Let me confirm and I'll get back to you". Never invent prices or lead times.
- Search before denying: if a product doesn't show up in the first search, it should try another term before saying you don't sell it.
- Transparency: if the customer asks whether they are talking to a person, it should say it is an assistant and offer to hand them off.
5. Set up actions and handoff
According to the setup guide, the agent can book appointments, trigger workflows, pass the conversation to a person, stop replying and follow up automatically. It can also transfer to another bot that specializes in something, for example one for sales and another for after-sales.
Define in writing what always goes to a person: complaints, returns, invoicing, custom orders and large purchases.
6. Test in suggestive mode, then move to autopilot
The Conversation AI documentation describes three modes: off, suggestive (the agent drafts and a person reviews and sends) and autopilot (it replies on its own). Test it first in the test panel, then leave it in suggestive mode with real conversations and switch to autopilot only when you hardly need to correct it. Same rule as in our team of AI agents: whoever writes doesn't approve.
For the agent to answer on WhatsApp, you must first have the channel connected. We explain it in how to connect WhatsApp to GoHighLevel.
How we use it in real stores
Nexa Soluciones, an interlocking-floor brand we built from scratch, uses GoHighLevel as its CRM. We added an AI agent on WhatsApp that answers questions, gives prices, follows up outside business hours and sends the payment link when the customer asks for it. If the customer stops replying, the follow-up writes again at growing intervals and stops on its own when they answer or have already booked. Sales closed in the CRM go back to Google Ads and Meta as conversions.
In an electrical supplies store, the agent is set up for WhatsApp, Instagram and the website chat, with an FAQ base, commercial-terms documents and the store's product table. It started in suggestive mode and hands off by type of inquiry and by amount. The most useful lesson: at first it denied products that were in the table because it only searched the FAQs. The search-before-denying rule fixed it; the table needs to be refreshed from the store often.
Voice agent: answer calls with AI
Voice AI is the same concept applied to the phone. According to the inbound call guide, you connect a HighLevel phone number to the agent from Settings > Phone System > Manage Numbers > Edit Configuration > Call Forwarding.
At Nexa we set up a voice agent that answers inbound calls with the same information as the WhatsApp one, collects the customer's details and, when the call ends, a workflow sends them the payment link or the factory location by WhatsApp. Calls carry a separate telephony cost.
How much the AI agent costs in GoHighLevel
Data from the official AI Product Pricing page (updated 10/11/2026), in dollars and per sub-account:
| Option | Price | Conversation AI (text) | Voice AI (voice) |
|---|---|---|---|
| Pay as you go | Based on tokens or minutes used | Charged by usage | Charged per minute |
| AI Employee Starter | USD 20 per month (agencies on the Founder plan only) | Not included, charged by usage | Not included, charged by usage |
| AI Employee Growth | USD 50 per month | 1,000 replies per month | 100 minutes per month |
| AI Employee Unlimited | USD 97 per month | Unlimited, fair use | Unlimited, fair use |
Note: when what's included runs out, the agent pauses unless you turn on additional usage. Telephony and the WhatsApp messages Meta charges are on top.
Mistakes we see often
- Turning on autopilot on day one, skipping suggestive mode.
- An outdated knowledge base: prices or promotions that no longer run.
- Instructions that don't say what to do when it doesn't know: that is where the agent makes things up.
- A single bot for everything: sales, after-sales and complaints mixed together.
- Not reading conversations: the dashboard and the agent's logs show where it fails.
Conclusion
Creating an AI agent in GoHighLevel is not writing a good prompt and switching it on. It means defining a job, giving it real information, telling it what to do when it doesn't know, marking when it hands off to a person and testing it before letting it loose. That way it answers outside business hours and doesn't make things up.
At We make it Lab we work as an agency on GoHighLevel with our own Agency account and build the agent inside a digital partner package for stores that already sell, together with the CRM, WhatsApp, email and ads (at Nexa, Meta Ads reached x7.7 with 30 purchases). If you are going to hire someone, use these questions to choose a provider. See how we do it in CRM implementation and automation, or write to us. You can also send us a WhatsApp at +54 9 11 5042-0646. And if you are still choosing a platform, read Kommo or GoHighLevel.
Sources consulted
You may also like
Kommo or GoHighLevel: which one is better for selling on WhatsApp?Frequently asked questions
- Does GoHighLevel include an AI agent? Some AI assistants (for creating content or sites) come included for agency users, with fair-use limits, but the agents that answer customers (Conversation AI and Voice AI) are paid separately, with an AI Employee plan or by usage.
- Does GoHighLevel's AI agent reply on WhatsApp? Yes. WhatsApp is one of the agent's channels, along with Instagram, Facebook, website chat, SMS and email. You must first connect WhatsApp to the sub-account.
- Can it sell on its own? It can answer, recommend and send the payment link through a workflow. Large purchases and complaints are better handled by a person.
- How long does it take to be ready? It depends on the catalog and the exceptions in your operation. What takes time is testing in suggestive mode with real conversations.
About the author

Cristian Cabrera
Tech & Data · We make it Lab
Explore more on



