AI and agents

How to write the prompt for your AI bot in GoHighLevel, with examples

By Edgar Jaimes 5 min read

Two white paper speech bubbles on a brown background

Role, script, rules, examples and knowledge base: how to write the Conversation AI prompt in GoHighLevel so it qualifies leads and books appointments, with examples.

You turned on the AI bot in GoHighLevel, wrote "you are a salesperson, help customers book" and in the very first conversation it booked someone who was only asking for your hours. The bot isn't broken: it's doing exactly the little you asked of it. The difference between a bot that gets in the way and one that sells is how you write its prompt.

This guide builds on LeadConnector's official AI Prompting 101 guide, the platform behind GoHighLevel, and takes it further: examples adapted to Latin American businesses, a transparency rule the original doesn't include, and role, script and rules texts ready to adapt.

What parts should your bot's prompt have?

Three: the role, the task with its script, and the rules. That's how LeadConnector frames it, and it's the simplest way to organize what the bot needs to know.

PartWhat it answersExample
RoleWho are you and how do you talk?Company assistant, friendly tone
TaskWhat must you achieve and in what order?Qualify with 3 questions and book
RulesWhat must you never do?Don't book without qualifying, 20 words max

If this is your first bot, before writing check what GoHighLevel is and who it's for.

How do I write the bot's role?

Say who it works for, what its goal is and how it should sound, with concrete details. Compare:

  • Worse: "You are a salesperson for {{ai.business_name}}. Help customers book."
  • Better: "You work for {{ai.business_name}}. Your goal is to assist, qualify the customer with the questions in the Script and, if they qualify, help them book. Use a professional but friendly tone. Use expressions like 'Done', 'Perfect' or 'Got it' instead of formal apologies. Call the customer by their name, {{contact.name}}, only when greeting and saying goodbye."

Heads up: the original guide suggests the bot shouldn't reveal it's an AI unless asked. We don't recommend it. Have it introduce itself as a virtual assistant and hand over to a person when the customer asks. You build trust and stay aligned with what we explain in Meta's rules for AI assistants on WhatsApp.

How do I write the task and the script?

Turn "qualify and book" into steps, with the exact questions and what to do with each answer. An example for a services business:

Hand ticking the boxes of a list on a tablet with a digital pen
A script with clear steps tells the bot what to ask and when to move forward. Photo: Unsplash
  • Worse: "Qualify customers with questions and then ask them to book. If they're not interested, take their details."
  • Better: "Script: 1. Ask what service they're looking for. If they're not interested, ask for their email to stay in touch. 2. If they respond with interest, ask when they need it. 3. Only if they answered well to everything, offer to book a call. 4. Before booking, ask for their email if you don't have it. 5. Once the appointment is confirmed, thank them and say goodbye."

If you also want to prioritize who to call first, combine the script with lead scoring in your CRM.

What rules should I give it?

Measurable rules, not wishes. "Keep it short" is ambiguous; "20 words max" isn't. LeadConnector's guide proposes, for example:

  • Short, direct answers of 20 words max.
  • Wait for the customer's answer before asking the next question in the Script.
  • Book only if the customer responded positively to the Script questions.

LeadConnector also recommends repeating what's critical: if the Script is the most important thing, mention it in the role and again in the rules. And giving examples of what to say and what to avoid, because sometimes it's easier to show the tone than to describe it:

  • Avoid: "I did not understand your response." Use: "Sorry, could you say that again?"
  • Avoid: "I apologize for the confusion." Use: "Sorry if that wasn't clear."

Where do I put the business information?

The essentials in the prompt, separated with delimiters; everything else in the knowledge base. LeadConnector recommends keeping the context inside the prompt under 100 to 200 words: if it grows too much, the instructions lose weight compared to the information.

To separate that block, use delimiters, such as a heading with # and tags like <oferta> and </oferta> around the text. Anything long (prices, policies, catalog) goes in the knowledge base. According to the Conversation AI training guide, you can combine pages from your website, FAQs, rich text, files, tables and Google Drive.

Common mistake: training the bot on your entire website. LeadConnector advises excluding outdated, duplicated, navigation-only, legal boilerplate and unrelated pages, because they make the answers worse.

How do I test the bot before launching it?

In two stages: first that it finds the right information, and then the full conversation. LeadConnector's guide says to use the knowledge base retrieval tester and then Test your Bot, in Bot Training, with four kinds of question: a common one, the same one rephrased, a frequent one and an off-topic one. Reset the test before each new scenario.

Open spiral notebook on a white desk
Write down every failure from your tests: each one is a missing rule or example. Photo: Unsplash

Once live, the Response Info option on each message shows you which sources the answer came from. Writing a prompt is iterative: test, adjust and test again. To go deeper into techniques, see how to write the instructions for your AI agent.

In short

The bot that booked someone who was only asking for your hours didn't need another tool, but a prompt with a role, a script, measurable rules and examples, plus a well-organized knowledge base. Test it before letting it loose and reply fast, because speed with your leads sells too. If you'd rather we set it up for you, see our GoHighLevel service.

Frequently asked questions

What parts should a bot prompt in GoHighLevel have?

LeadConnector's official guide proposes three: the role (who the bot is and how it talks), the task (what it must achieve and with which script) and the rules it must follow.

How much business information should I put inside the prompt?

LeadConnector recommends keeping that context under 100 to 200 words. Everything else is better loaded into the bot's knowledge base.

Should the bot say it is an artificial intelligence?

The original guide suggests the bot shouldn't reveal it's AI unless asked. We recommend being transparent: say it's a virtual assistant and hand over to a person when the customer asks.

How do I test the bot before turning it on?

First with the knowledge base retrieval tester and then in Test your Bot, inside Bot Training, with common, rephrased, frequent and off-topic questions.

What are delimiters for in a prompt?

To separate blocks such as the offer or the context from the rest of the instructions, using symbols like # or tags like <oferta></oferta>.

Do not stop at the theory: apply it to your business

We implement GoHighLevel end to end, with your real operation. The first conversation is free.

Keep reading