How to Train ZazzyAgent AI With FAQs, Website Content & Files
An AI Agent is only as useful as the information it has access to.
A strong System Prompt tells the agent how to behave.
Your knowledge sources give it the business information it needs to answer customers.
ZazzyAgent's AI training workflow is designed around this separation.
System Prompt vs Knowledge
This is the first concept to understand.
System Prompt
The System Prompt tells the agent:
Who it is
What it should do
How it should behave
How it should communicate
When it should ask questions
When it should use an action
When it should transfer a conversation
What it must avoid
Knowledge
Knowledge provides information such as:
Products
Services
Policies
FAQs
Website information
Documents
Other business content
The current underlying AI system uses a knowledge/training campaign connected to the AI Agent, while the System Prompt defines how the agent should use it.
What can you use to train the AI?
Depending on the ZazzyAgent setup, supported knowledge sources can include:
FAQs
Useful for structured questions and answers.
Example:
Question: What is your delivery time?
Answer: Orders are usually delivered within 3–5 business days.
Website content
Useful when important business information already exists on your website.
Examples:
Product pages
Service pages
FAQ pages
Shipping policy
Return policy
About page
Documents
Useful for internal or detailed business information.
Examples:
PDF brochures
Product manuals
Service documents
Policy documents
Training documents
Create a knowledge/training campaign
Open the AI training/knowledge area in ZazzyAgent and create a new knowledge source or training campaign.
Give it a useful name.
Bad:
AI Knowledge 1
Better:
Ecommerce Sales Knowledge
Or:
Customer Support Knowledge
Or:
Product Catalogue Knowledge
The name should tell you what information belongs inside it.
Keep knowledge focused
Do not automatically put every document you own into every AI agent.
A Sales Agent should primarily receive sales-related information.
A Support Agent should primarily receive support-related information.
A product recommendation agent should have product information.
A separate order-support agent may need shipping, returns and order policies.
Focused knowledge makes it easier to control what the agent should rely on.
Write useful FAQs
Bad FAQ:
Shipping?
Yes.
That doesn't give the AI enough context.
Better:
Question: How long does delivery take?
Answer: Standard delivery usually takes 3–5 business days after the order has been dispatched. Delivery times may vary by location.
A good FAQ contains the full answer rather than relying on the AI to guess missing details.
Train the AI using website pages
Website training is useful when your website already contains accurate information.
Choose pages that contain useful customer-facing information.
Good sources include:
Product information
Pricing pages
FAQ pages
Shipping policy
Return policy
Service descriptions
Avoid adding irrelevant pages just to increase the amount of training data.
Train the AI using files
Documents are useful for information that doesn't live neatly on your website.
Before uploading a document, check that:
The content is current.
The document is readable.
Old policies have been removed.
Product information is accurate.
Internal information that customers should never see is excluded.
Keep knowledge updated
An AI can only answer correctly from information it has.
If your policy changes, update the relevant knowledge.
If a product changes, update the product information.
If your delivery policy changes, update the shipping information.
Don't expect the System Prompt to compensate for outdated business information.
Knowledge does not replace live data
Static knowledge is suitable for things like:
What is your return policy?
But not necessarily for:
Where is my order right now?
That type of question requires current data.
A better architecture is:
Knowledge
→ general order policy
Action/API
→ actual order status
The current AI Agent system supports this pattern: general information can come from the knowledge source while live information can be retrieved through an API or ecommerce action.
Give every agent the right knowledge
A useful approach is to create separate knowledge collections.
Sales Knowledge
Products
Pricing
Product comparisons
Sizes
Recommendations
Shipping
Returns
Support Knowledge
Shipping
Returns
Refunds
Order policies
FAQs
Support instructions
Appointment Knowledge
Services
Appointment rules
Locations
Availability rules
Cancellation policy
Then connect each AI Agent to the appropriate knowledge.
How to test your training
After training, don't ask only questions that are directly copied from your knowledge.
Test natural customer questions.
If your knowledge says:
Customers may return products within 7 days of delivery.
Ask:
Can I send this back next week?
The AI should understand that the question relates to the return policy.
Also test:
Do you accept returns?
What if the product doesn't fit?
Can I exchange it?
What is your return window?
Good training should help the AI understand variations of customer questions, not simply reproduce exact FAQ wording.
A simple training checklist
Before connecting your knowledge to a live agent:
[ ] Information is current
[ ] Policies are accurate
[ ] Product data is correct
[ ] FAQs are complete
[ ] Website pages are relevant
[ ] Documents are current
[ ] Private/internal information is excluded
[ ] Live data is handled through actions/APIs where needed
[ ] The knowledge matches the agent's responsibility
[ ] Real customer-style questions have been tested
Good AI training is less about uploading the most information possible and more about giving the agent the right information in the right structure.