What is a prompt?
A prompt is the "brain" behind your AI Agent — a set of written instructions that tell it who it is, how it should act, and what it should do.
Think of your AI Agent as a new team member. The prompt is their onboarding guide: it explains their job, how to talk to customers, and what to avoid. If your prompt is too vague, your agent will improvise. If your prompt is clear and detailed, your agent will stay on-brand and accurate — just like a well-trained teammate.
Why prompts matter
Your AI Agent already knows how to chat, but it doesn't know your business until you tell it. The prompt defines your brand voice, priorities, and personality. It's what turns your AI Agent from "just another chatbot" into a virtual team member your customers trust.
A great prompt helps the AI:
Understand what to do — including its main tasks and responsibilities.
Adopt your brand's tone of voice — knowing how to chat, sound, and behave.
Recognise what to avoid — including taboos, boundaries, and things it should never say.
Your AI Agent follows instructions literally — so write your prompt in a way that a 10-year-old could follow. Short sentences, no guesswork, no "creative interpretation."
The building blocks of a great prompt
In Superchat, your prompt is split across two areas::
Task & Rules - With three fields: What should your agent do?, What should your agent never do?, and What should your agent know by heart?)
Tone & Voice - Separated by channel (Chat, Phone & Email) to ensure the AI Agent behaves appropriately for each type of communication.
What should your agent do?
This is where you explain who your AI Agent is and what exactly it should do. Define the agent's role and give clear step-by-step instructions. If your agent should use tools — like saving data or writing notes — explain when and how to do so.
Agent for phone and chat: Your agent automatically detects which channel it is currently active on — on the phone or in a text-based channel (chat, email). You can use this in the prompt to give channel-specific instructions.
When a dedicated phone agent makes sense: in most cases one agent for all channels is enough. But if phone and chat differ significantly — for example with large amounts of knowledge or very different workflows — it can make sense to create a separate agent for phone.
Ask yourself:
What is my AI Agent responsible for? What are its main tasks or functions?
What is the ultimate business goal: qualify applicants, support customers, close deals, prepare and schedule callbacks?
What is the AI Agent's name?
Is there a step-by-step process it should follow (first do this, then that)?
Should it use tools like Forward Call, Capture Contact Data or Write Internal Note — and when exactly?
Should it behave differently depending on the channel (phone vs. chat)?
Examples:
"You are Sophie, a support assistant at Super Sofa Co. Your job is to help customers with returns, delivery updates, and product info."
"If a customer shares an order ID, confirm that it matches our format (it always starts with an 'X' followed by eight numbers) before proceeding."
"Ask for first name, last name, and email. If any of these are already stored, skip that question."
"Once you've collected all three, save them."
"After finishing a conversation, summarise the key points and next steps in an internal note."
On the phone: "Ask if the caller can be reached on this number."
For more information on great prompting, watch our video below explaining how to train your AI Agent.
What should your agent never do?
This is where you define what the AI Agent must never do. It's your safety net — making sure the agent stays on brand and doesn't make false or sensitive statements. Without clear boundaries, the AI might make guesses or provide content it should avoid.
Ask yourself:
What mistakes must my AI Agent avoid at all costs? (e.g. confirming prices, giving medical advice)
What must it never say? (e.g. mention competitors, discuss internal policies)
What topics should it stay away from entirely?
Are there legal, brand, or compliance requirements it must always follow?
Examples:
"Never give medical, legal, or financial advice."
"Never confirm offers or prices — only share information from the knowledge base."
"Avoid mentioning competitors or comparing products."
"Do not discuss internal company policies or team details."
"Never invent data or order details."
"If unsure, ask for clarification instead of guessing."
💡 Pro tip: Handover to a human. Escalation and handover rules should not go in the prompt — set them up per channel instead.
Chat & Email: in the AI Agent node within the Automation. There you define custom exit conditions, specify when they trigger (e.g. when the contact asks for a human), and decide what happens next — such as assigning the conversation to a team member.
Phone: via the "Forward Call" tool. Enable the tool and describe in the prompt under "What should your agent do?" when it should forward (e.g. when the caller explicitly requests a human).
What should your agent know by heart?
Only instant facts that anyone on your team could answer immediately without looking them up — e.g. opening hours or location. This is not the place for your entire knowledge base: everything deeper and the bulk of your knowledge belongs in Knowledge & Data.
Language Style & Voice
You no longer set tone in the prompt text itself — instead you set it in the Language Style & Voice section, separately for Chat, Phone and Email. The following questions help you describe it per channel.
Ask yourself:
How should the AI sound — friendly, formal, casual, or professional?
Should it write/speak in short, direct sentences or longer, more detailed ones?
Should it use emojis — or avoid them entirely?
On the phone: it's heard, not read — short, natural sentences, no bullet points, one question at a time.
Examples:
"Be friendly, clear, and professional — sound like a helpful team member, not a bot."
"Write short, direct sentences. Don't repeat yourself."
"Only use emojis if the customer uses them first."
"Be empathetic and polite, but stay concise — avoid over explaining."
💡 Pro tip: If you're unsure how to describe your tone, write one short example reply in your style — the way your team would actually respond. The AI will automatically learn from that sample and use it as a reference for tone and writing style.
Writing tips for better prompts
Be literal. The AI does exactly what you say — word for word.
Be specific. Describe exactly when and how something should be done ("ask this first, then that").
Avoid filler words. Drop phrases like "usually", "try to" or "in general".
Use examples. Real examples help the AI understand your intent.
Keep facts separate. Don't mix company info into the prompt — use the knowledge base for that.
Think about the channel. Where needed, write separately how the agent should behave on the phone, in chat and by email.
Test regularly. Small tweaks can make a big difference.