How to Train Using Briefs, Instructions, and Edits
Quack AI is built to learn and improve based on your feedback. We offer three powerful training methods, each designed for a different type of improvement.
Method 1: Editing Answers
This method is ideal for correcting specific responses directly.
How It Works
Ask a Question in the Training Center
Review the AI’s Response
If correct: Click 👍 to mark it as verified.
If incorrect: Click the ✏️ pencil icon to edit the response.
Write the Correct Answer
Replace the AI-generated response with a clear and accurate one. Then click Save and click 👍 to mark it as verified.
Test with Variations
Ask the same question using different wording to make sure the AI understands the concept.
💡 Pro Tip: Training this way helps Quack learn exactly how to respond to similar questions in the future.
Method 2: Using Briefs
Briefs give Quack topic-level guidance, perfect for complex subjects or style rules.
What Is a Brief?
A brief is a written set of knowledge and logic that tells Quack how to respond to a specific topic. It can include tone guidelines, do’s and don’ts, key points to highlight, and examples to follow.
How to Add a Brief
Select the Topic you want to train from the Training Center
On the right menu, click “Add Brief”
Enter clear, specific guidance for the AI. For example:
“Always clarify whether the user is on mobile or desktop before giving steps.”
“Don’t mention support hours in this topic.”
Add Multiple Guides If Needed
You can add more than one brief for the same topic to guide different use cases.
Save and Test Responses
Re-ask benchmark questions to confirm the new behavior.
💡 Pro Tip: Use briefs to maintain a consistent tone, prevent incorrect assumptions, or reinforce product rules.
Method 3: Instructions
Use instructions for precise behavioral rules - ideal for quick, repeatable directives.
What Is an Instruction?
An instruction is a direct command, it tells the AI to always or never do something in a specific context.
How to Use Instructions
Go to the Training Center and select the topic or question you want to fine-tune.
Enter Your Question and Click Improve:
Type your question as you would normally.
Click Improve.
In the Improve panel, describe how Quack should respond differently. Use clear instructions, such as:
never ask for an email address
💡 Pro Tip: Use simple, direct commands to get the best results.
Quack will generate an improved instruction based on your input
Review the suggestion and edit it if needed.
Click Submit when you're satisfied.
Manage Your Instructions
You can view and delete instructions anytime from the Instructions section within your topic.
💡 Pro Tip: Use instructions to enforce compliance, avoid sensitive actions, or support team preferences.
Our Recommendations
To get the most value from Quack AI training, we recommend starting with edits and briefs:
Use edits to correct specific answers
Use briefs to set clear expectations for how Quack should respond across a topic
These two methods provide strong foundational guidance and quickly improve answer quality. Instructions are best used later, once you’ve identified consistent behavioral patterns or compliance rules that need reinforcement.
Best Practices for AI Optimization
Be specific – Clearly define how the AI should respond to different queries.
Use structured instructions – Guide the AI on how to format responses effectively.
Test frequently – Run simulations to ensure high response accuracy.
Refine continuously – Monitor AI behavior and update instructions as needed.
Next Steps: Start Training!
Benchmark Your AI: Create and test 10–20 common customer questions
Review the Results: Give a thumbs up to good answers and a thumbs down to incorrect ones
Edit and Save Corrections: Use the pencil icon to fix any bad answers
Add Briefs Where Needed: Guide the AI on how to respond across an entire topic
Use Instructions if needed: Add targeted rules to fine-tune behavior
💡 Pro Tip: Re-test your original questions after training to confirm the AI is improving. Repeat the process as new topics emerge.