Why Isn't My ZazzyAgent AI Agent Using Its Knowledge?
You've trained your AI Agent, added your business information, but it still gives poor or generic answers.
Before adding more documents, check the configuration.
The problem may not be a lack of knowledge.
1. Is the knowledge source connected to the agent?
Open the AI Agent configuration.
Confirm that the intended knowledge source is attached to the agent you're testing.
A knowledge source existing in your account doesn't necessarily mean every AI Agent can use it.
2. Are you testing the correct agent?
If you have multiple agents, make sure you're talking to the one with the knowledge you expect.
For example:
Sales Agent
→ Product knowledge
Support Agent
→ Support knowledge
If you're testing Support but expecting Sales information, the behavior may appear incorrect.
3. Is the information actually in the knowledge?
Search your knowledge source.
Confirm that the answer exists.
Suppose the customer asks:
What is your return period?
Your knowledge should contain an actual answer to that question.
Don't assume the AI will infer your entire return policy from a document that only mentions returns once.
4. Is the information current?
Outdated knowledge can be worse than missing knowledge.
Check:
Pricing
Product information
Delivery times
Returns
Promotions
Policies
Business hours
Remove or update obsolete information.
5. Check the System Prompt
The System Prompt controls the agent's behavior.
You should tell it how to use its knowledge.
Example:
Use the connected knowledge source to answer
questions about our products, services and policies.
When the answer is not available in the knowledge,
do not invent information.
Tell the customer that you don't have enough information
and offer the appropriate next step.
6. Don't put live information into static knowledge
Suppose your knowledge says:
Orders usually ship within 2–3 business days.
That's useful.
But the customer asks:
Where is my order #12345?
The current order status should come from a live ecommerce action or API, not from static knowledge.
See How to Build a Shopify Order Status AI Agent on WhatsApp.
7. Ask natural customer questions
Don't test only with the exact wording used in your knowledge.
If your FAQ says:
Our standard delivery time is 3–5 business days.
Test:
When will my order arrive?
or:
How long does delivery take?
or:
How many days does shipping usually take?
This is closer to how customers actually speak.
8. Check conflicting information
Suppose one document says:
Returns are allowed within 7 days.
Another old document says:
Returns are allowed within 14 days.
The agent now has conflicting information.
Remove or update the outdated source.
Your knowledge should have one authoritative answer.
9. Keep knowledge focused
Don't upload every document your company has.
A Sales Agent doesn't necessarily need:
Internal HR policies
Employee documents
Unrelated technical manuals
Old campaign documents
Give the agent information relevant to its role.
10. Knowledge vs System Prompt
Keep this distinction clear.
Knowledge
Standard delivery is 3–5 business days.
System Prompt
Use the connected knowledge when answering delivery-policy questions. Never invent delivery information.
Knowledge provides the fact.
The System Prompt provides the instruction.
11. Knowledge vs API
Use static knowledge for:
What is your return policy?
Use an API or ecommerce action for:
Has my return for order #12345 been approved?
The second answer changes over time.
12. Test with a question you know the answer to
Choose one piece of information that definitely exists in your knowledge.
Ask the agent a natural version of that question.
If the response is still generic, inspect:
Knowledge connection
Agent selection
Prompt
Knowledge source
Source content
13. Test an unknown question
Now ask something that's intentionally not in your knowledge.
A good System Prompt should stop the AI from inventing the answer.
Example:
Do you offer free private jet delivery?
The agent should not pretend that it does.
Common symptoms
Agent gives generic answers
Check whether the appropriate knowledge source is connected.
Agent gives outdated information
Update or remove outdated content.
Agent invents information
Strengthen the prompt:
Never invent information that is not available in the connected knowledge or configured actions.
Agent uses policy knowledge for live questions
Add rules telling it when to use APIs or ecommerce actions.
Agent ignores part of a document
Make the information easier to understand and remove conflicting content.
A good knowledge setup
AI Agent
↓
Relevant Knowledge
↓
Clear System Prompt
↓
Natural-language question
↓
Knowledge answer
For live information:
AI Agent
↓
Recognizes live-data question
↓
API / Ecommerce Action
↓
Live result
↓
Customer answer
The most common mistake
Adding more and more documents isn't always the solution.
Before adding another file, ask:
Does the agent actually have access to the right knowledge source?
Then:
Does that source contain a clear, current answer?
Then:
Does the System Prompt tell the agent to use it?
Those three checks solve many knowledge-related problems.
zazzyagent.com
