Skip to main content
Prompt engineering is the practice of writing clear, structured instructions that tell your AI agent exactly how to behave. A well-crafted prompt is the single biggest factor in determining whether your Conversation AI or Voice AI agent delivers helpful, on-brand responses — or frustrates your customers. This guide introduces a proven 4-part framework you can apply to every AI agent you build in HoopAI. Whether you are setting up your first chatbot or fine-tuning a phone agent that handles hundreds of calls per day, these principles will help you get dramatically better results.

​
Why prompt engineering matters

Your AI agent is only as good as the instructions you give it. Without a clear prompt, even the most powerful language model will:
  • Hallucinate information it does not have
  • Give vague, generic answers that erode customer trust
  • Fail to collect the information your business needs
  • Miss opportunities to book appointments or qualify leads
A strong prompt turns your AI agent into a reliable team member that knows your business, follows your rules, and represents your brand consistently across every conversation.
You do not need a technical background to write great prompts. If you can write a clear email to a new employee explaining their job, you can write an effective AI prompt.

​
The 4-part framework

Every high-performing prompt follows the same structure: Role, Task, Guidelines, and Examples. Think of it as an onboarding packet for a new hire — you tell them who they are, what they need to do, what rules to follow, and show them examples of great work.

​
Part 1: Role — define who the AI is

The Role section establishes the AI agent’s identity, personality, and tone of voice. This context shapes every response the agent generates. A good role definition includes:
  • Business identity — the company name, industry, and what you do
  • Agent persona — the agent’s name and character (“You are Sarah, a friendly scheduling assistant for Bright Smiles Dental”)
  • Tone of voice — professional, casual, warm, authoritative, empathetic
  • Language — what language(s) the agent should use
You are Alex, a friendly and professional scheduling assistant for
Evergreen Family Dental. You work at a family dental practice in
Portland, Oregon that has been serving the community for 15 years.
Your tone is warm, patient, and reassuring — especially when speaking
with nervous patients. You always refer to the practice as
"Evergreen Family Dental" or "our office."
The more specific your role definition, the more consistent and natural the agent’s responses will be.

​
Part 2: Task — what the AI should accomplish

The Task section defines the agent’s primary objectives. Be explicit about what the agent should do in order of priority. Common tasks include:
  • Answer questions about your products, services, hours, and pricing
  • Book appointments by collecting name, preferred date/time, and service type
  • Qualify leads by gathering budget, timeline, and decision-making authority
  • Collect contact information such as email, phone number, and company name
  • Route conversations to the right team member based on the customer’s need
Your primary goals, in order of priority, are:

1. Answer questions about our services, pricing, and office hours
   using only the information in your knowledge base.
2. Book dental appointments by collecting the patient's full name,
   preferred date and time, and the type of service they need
   (cleaning, consultation, or emergency).
3. If the patient has a billing question or complaint, collect their
   name and phone number and let them know a team member will call
   them back within one business day.

​
Part 3: Guidelines — set guardrails and rules

Guidelines are the rules your agent must follow. They prevent the agent from going off-script, sharing inaccurate information, or behaving in ways that could harm your business. Effective guidelines cover:
  • What NOT to do — “Never provide medical advice,” “Do not discuss competitor pricing,” “Never make up information”
  • Knowledge boundaries — “Only answer questions using the information in your knowledge base. If you do not know the answer, say so.”
  • Off-topic handling — “If the customer asks about something unrelated to our services, politely redirect the conversation.”
  • Escalation rules — “If the customer asks to speak with a human, immediately transfer the conversation to a live agent.”
  • Compliance rules — “Never ask for credit card numbers, social security numbers, or other sensitive information in the chat.”
Rules you must always follow:

- Only answer questions using the information provided in your
  knowledge base. Never guess or make up information.
- If you do not know the answer, say: "That's a great question —
  let me connect you with someone on our team who can help."
- Never provide medical diagnoses, treatment recommendations, or
  medication advice.
- If the customer expresses frustration or asks to speak with a
  person, immediately offer to transfer them to a team member.
- Do not discuss competitor practices or pricing.
- Keep responses concise — no more than 2-3 sentences per message
  unless the customer asks for more detail.
- Always confirm appointment details before finalizing.

​
Part 4: Examples — show ideal interactions

Examples are the most underrated part of prompt engineering. Showing the AI what a great conversation looks like is often more effective than writing dozens of rules. Include 2-3 examples that cover your most common scenarios:
Example interactions
Example 1 — Booking an appointment:
Customer: "I'd like to schedule a cleaning."
Agent: "I'd be happy to help you schedule a cleaning! Could I get
your full name and your preferred date and time? We have
availability most weekdays between 8 AM and 5 PM."

Example 2 — Handling an unknown question:
Customer: "Do you accept MetLife insurance?"
Agent: "That's a great question. I want to make sure I give you
accurate information, so let me connect you with our front desk
team. They can verify your specific plan. Could I get your name
and phone number so they can reach out?"

Example 3 — Off-topic request:
Customer: "Can you recommend a good restaurant nearby?"
Agent: "Ha, I wish I could help with dinner plans! Unfortunately,
I'm only able to help with dental appointments and questions about
our services. Is there anything dental-related I can assist with?"
Examples teach by showing, not telling. When your agent sees a pattern in your examples, it naturally follows that pattern in new conversations. Start with your 3 most common customer interactions.

​
How context and prompt length affect quality

There is a balance between giving enough detail and keeping your prompt manageable. Here are the key principles:

​
More context is usually better — up to a point

  • Specific business details (hours, services, pricing tiers) help the agent give accurate answers
  • Detailed examples reduce ambiguity and improve consistency
  • Clear escalation paths prevent the agent from getting stuck

​
But overly long prompts can cause problems

  • Contradictory instructions buried deep in a long prompt get ignored or cause unpredictable behavior
  • Redundant instructions waste context space without improving quality
  • Extremely long prompts can dilute the priority of your most important rules

​
The sweet spot

For most businesses, a well-structured prompt is between 300 and 800 words. Your prompt should be long enough to cover your core scenarios and short enough that every sentence earns its place.
If your prompt exceeds 1,000 words, consider moving detailed product information and FAQs into your Knowledge Base instead. The knowledge base is designed to hold large volumes of reference material that the agent can search through as needed.

​
Quick-start template

Copy this template and fill in the bracketed sections to create your first prompt:
Prompt template
ROLE:
You are [Agent Name], a [tone — e.g., friendly and professional]
[role — e.g., customer service assistant] for [Business Name].
[1-2 sentences about your business and what makes it unique.]

TASK:
Your primary goals are:
1. [Primary goal — e.g., Answer questions about our services using
   only the knowledge base]
2. [Secondary goal — e.g., Book appointments by collecting name,
   date/time preference, and service type]
3. [Tertiary goal — e.g., Collect contact info for leads and pass
   them to the sales team]

GUIDELINES:
- Only answer using information from your knowledge base. If you
  don't know, say so and offer to connect them with a team member.
- Never provide [prohibited topics — e.g., medical advice, legal
  guidance, specific pricing not in your knowledge base].
- If the customer asks to speak with a human, immediately
  [transfer/collect their info for a callback].
- Keep responses [concise/detailed] and [tone descriptor].
- [Any channel-specific rules — e.g., Keep SMS messages under
  160 characters when possible.]

EXAMPLES:
[Include 2-3 example conversations covering your most common
scenarios]

​
Prompt engineering across HoopAI

The 4-part framework applies whether you are writing prompts for Conversation AI chatbots or Voice AI phone agents — but each channel has unique considerations. Dive into the detailed guides below to learn channel-specific best practices.

Conversation AI prompts

Step-by-step guide for crafting chat prompts across SMS, web chat, and social channels

Voice AI prompts

Write natural, effective prompts for AI phone agents that handle real calls

Advanced techniques

Multi-step prompting, conditional logic, tone calibration, and more

Common mistakes

The top 10 prompt mistakes and how to fix them with before-and-after examples

Optimization and testing

Systematic methods to test, measure, and improve your prompts over time

Knowledge base

Train your AI agents with custom data sources for accurate, on-brand answers

​
Next steps

Writing prompts for Conversation AI

Apply the 4-part framework to your Conversation AI chatbot with channel-specific tips

Writing prompts for Voice AI

Adapt the framework for natural-sounding phone conversations

Set up Conversation AI

Create your first Conversation AI bot and configure its channels and settings

Configure bot settings

Assign channels, set bot status, and manage primary vs. non-primary bots
Last modified on March 5, 2026