Rules guide how your VWAM agent behaves. They are natural language instructions added to the agent’s system prompt to influence how it selects products, manages conversations, and expresses tone.
Rules are not hard-coded filters. Instead, they act as instructional guardrails that shape how the AI interprets user intent and generates responses. Clear, specific rules improve consistency, strengthen brand alignment, and increase recommendation quality.
During a conversation:
The product feed grounds available recommendations.
Rules guide how the AI behaves.
The clearer and more explicit the instruction, the more consistent the outcome.
Before you begin
Before creating rules:
Define the primary goal of the agent.
Confirm the product feed includes accurate attributes such as price, category, tags, or experience level.
Types of rules
Rules are organized into three sections:
Recommendation rules
Conversation rules
Tone of voice rules
Each section controls a different behavioral dimension of the agent.
Create and manage rules
Each section controls a different aspect of agent behavior.
Open the Editor.
Navigate to the Rules section in the Editor.
Open any of the following sections:
Recommendations
Conversation
Tone of voice
Create a rule
In the selected section, click Add rule.
Enter:
Rule name
Instruction (Prompt field)
Click Add.
You can create multiple rules in each section.
Edit or delete a rule
Select the three-dot menu next to the rule.
Choose Edit to modify the rule at any time.
Choose Delete to remove a rule that is no longer needed.
Choose Duplicate to create a copy of an existing rule.
Rules guide behavior. They do not replace structured product data or strict business logic.
Configure rules
Recommendation rules
Recommendation rules control how products are selected and presented.
Use this section to:
Prioritize specific products or categories
Emphasize important attributes
Enforce constraints such as price or experience level
Control how comparisons are framed
Weak instruction
Recommend good products.
Strong instruction
Only recommend products under $100.
Explain clearly when no products match the budget.
Ask for budget when it is not provided.
Explicit instructions improve adherence.
Example use cases:
“Only recommend products under $100”
“Prioritize items in the best seller category”
“If a user mentions “beginner,” suggest entry-level options first”
“Avoid recommending products from the “Clearance” category”
Conversation rules
Conversation rules control how the agent manages interactions.
Use this section to:
Require clarifying questions
Handle vague requests
Manage objections
Control response length
Address edge cases
Weak instruction
Be helpful.
Strong instruction
Ask clarifying questions when the request is vague.
Confirm budget, use case, and preferences before recommending products.
Do not recommend more than three products at once.
Example use cases:
“Always ask at least one clarifying question before recommending products.”
“If a user is vague, ask about budget and intended use.”
“Do not answer unrelated or inappropriate questions.”
“Keep responses under 120 words.”
Tone of Voice Rules
Tone of voice rules define how the AI sounds.
Use this section to control:
Formality
Personality
Brand alignment
Response structure
Weak instruction
Be friendly.
Strong instruction
Speak as a professional stylist.
Use confident and concise language.
Avoid slang.
Provide brief explanations for each recommendation.
Example use cases:
Professional and courteous
Playful and energetic
Expert stylist
Technical and precise
How Rules interact
All rules ultimately merge into a unified instruction layer.
This means:
Rules should not contradict each other.
Overly complex or conflicting rules may produce inconsistent behavior.
Clarity is more powerful than quantity.
If two rules conflict:
The AI will attempt to reconcile them, but results may vary.
Best practice: Keep rules specific and non-overlapping.
Write effective Rules
Follow these principles:
1. Be explicit
Avoid vague language.
Instead of:
Focus on affordability.
Write:
Only recommend products under $75 unless the user explicitly requests premium options.
2. Define behavior, not outcomes
Instead of:
Increase conversions.
Write:
After presenting recommendations, ask the user if they would like to compare options or see more details.
3. Limit scope per rule
Each rule should address one concept.
Avoid:
Only recommend under $100 and ask clarifying questions and be concise and don’t recommend clearance items.
Break this into multiple rules.
4. Test and iterate
After adding rules:
Use the live preview in the Editor.
Try edge cases.
Adjust language if adherence is weak.
If behavior is inconsistent, strengthen the instruction language.
Example rule sets
Budget-focused product finder | Premium brand stylist |
Recommendation: | Recommendation: |
Conversation: | Conversation: |
Tone of voice: | Tone of voice: |
Limitations
It is important to understand:
Rules guide the AI; they do not create strict filters.
The AI may occasionally deviate, especially if rules are vague.
Stronger, more detailed instructions produce better outcomes.
If strict filtering logic is required, ensure your product feed and business strategy align accordingly.
Next steps
Test responses in the Editor preview.
FAQ
What happens if rules conflict?
The AI attempts to reconcile them. Conflicting rules may lead to inconsistent responses. Keep instructions clear and non-overlapping.
Are rules guaranteed to be enforced?
No. Rules guide the AI but do not act as hard-coded filters. Clear and strong language improves adherence.