The Secret Weapon — The Examples Node (Orange)
This is the single most important node. If you only spend time on one section, make it this one.
6 min readWhy this node matters more than anything else
Everything else in your prompt map tells your agent what to do. The Identity node says who it is. The Style node says how it should sound. The Knowledge node gives it facts. The Rules node sets boundaries.
This node is different. This node SHOWS your agent what to do.
You write example conversations — a customer says X, your agent responds with Y. The AI reads these examples and copies the style, the length, the tone, and the approach. It's exactly like training a new employee by sitting next to them and showing them how you'd handle each situation. "When someone asks this, here's what I say." That's what this node does.
We've seen it over and over: two agents with identical settings everywhere else, but the one with better examples performs dramatically better. This is your single biggest quality lever. Spend your time here.
How to open the Examples node
Click the orange box labeled "Examples" on your prompt map. It will expand to show the example conversations inside.
The pre-filled examples — and why they work
Your template already comes with a few example conversations. Before you change anything, read through them. Notice a few things:
Each example starts with a realistic customer message — not something generic like "Hello," but something a real person would actually type. Then the agent response is friendly, gives a clear answer, and ends with an offer to help more. That pattern — acknowledge, answer, offer — is what makes an example good. The AI picks up on all of it: the friendliness, the structure, the length, even the way the response ends.
How to write your own examples
Let's walk through real examples for three different businesses so you can see what "perfect" looks like in practice.
Bakery Example
Customer says:
"Do you have gluten-free options?"
Agent responds:
"Yes! We have a rotating selection of gluten-free treats. Right now we usually have our gluten-free chocolate brownies and almond cookies available daily. The full list changes weekly, so I'd suggest calling us at (555) 123-4567 or stopping by to see today's selection. Anything else I can help with?"
Why this works: It's warm, gives specific items, admits the selection changes (honest), and provides a next step. Notice it does NOT say "all our gluten-free items are safe for celiacs" — because the Rules node prevents that.
Online Store Example
Customer says:
"My order hasn't arrived and it's been 10 days."
Agent responds:
"I'm sorry about that — I know it's frustrating when a package takes longer than expected. Standard shipping usually takes 5-7 business days, so yours is running behind. Could you share your order number? I'll look into exactly where it is and get you an update right away."
Why this works: It empathizes first (doesn't jump straight to troubleshooting), gives context about normal timing, and asks for the specific info needed to help. Professional but human.
SaaS Company Example
Customer says:
"How do I export my data to a spreadsheet?"
Agent responds:
"Sure thing! Head to your Dashboard, click the Reports tab in the top menu, then hit the 'Export' button in the upper right corner. You'll get a dropdown — pick 'CSV' for spreadsheets. The file will download automatically. If you need it in Excel format instead, choose 'XLSX' from that same dropdown. Let me know if you run into anything!"
Why this works: Step-by-step, no jargon, anticipates a follow-up question (CSV vs XLSX), and ends casually. The agent sounds like a helpful coworker, not a robot reading a manual.
Bad examples — surprisingly powerful
This is optional but worth knowing about. You can toggle on "bad examples" — these show the agent what a BAD response looks like. Here's how that works:
Example of a bad response
Customer says:
"What's your return policy?"
Bad agent response:
"Our return policy is governed by our terms and conditions which can be found on our website. Generally speaking, returns are subject to various conditions depending on the product category, the time elapsed since purchase, the condition of the item, whether it was purchased during a promotional period, and other factors that may apply. For specific details regarding your particular situation, I would recommend reviewing the full policy documentation available in the footer of our website or contacting our customer service department during business hours for a more detailed explanation of how the policy applies to your specific circumstances."
What's wrong: Way too long, says nothing useful, sends the customer in circles. A real customer would read this and think "just answer my question."
The agent reads this and thinks "I should NOT do that." It sounds simple, but it's surprisingly effective. One or two bad examples can eliminate entire categories of annoying responses.
How many examples do you need?
Minimum: 2. That gives the agent enough to see a pattern.
Recommended: 3 to 5. This covers your most common scenarios and gives the agent a solid sense of how you want it to sound.
Important: 3 great examples beat 10 mediocre ones. Every time. Don't pad it with filler just to hit a number. Each example should be one you're genuinely proud of.
Tips for writing great examples
Cover different types of questions. Don't write five examples that are all simple factual questions. Mix it up:
A straightforward factual question — "What are your hours?" This shows the agent how to give a quick, clean answer.
A frustrated customer — "This is the third time I've had this problem." This shows the agent how to empathize and de-escalate.
An edge case or tricky question — "Can I return something I bought six months ago?" This shows the agent how to handle situations where the answer isn't a simple yes or no.
Make each response the exact length you want. If your examples are three sentences long, your agent will write three-sentence responses. If they're three paragraphs long, guess what — you'll get three paragraphs. The agent mirrors what it sees.
Read your examples out loud. Seriously. Does it sound like your brand? Does it sound like something you'd actually say to a customer standing in front of you? If it sounds stiff or robotic when you read it out loud, rewrite it until it sounds natural.
Opening the Examples node → reading a pre-filled example → typing a new customer message in the User box → writing a response in the Agent box → toggling on a bad example
Video • 60 seconds
Side-by-side comparison: testing an agent with no examples vs. the same agent with 3 well-written examples — showing the dramatic difference in response quality
Video • 30 seconds
You just showed your agent exactly how to do its job. The examples you wrote are the single most impactful thing you can do for response quality. Everything else matters, but this is what separates an agent that feels generic from one that sounds like YOU. Up next: optional power-ups to take things even further.