Creating effective policies is key to successful content moderation. This section covers best practices specific to the Deep Mod platform for writing clear, maintainable policies that balance automation with accuracy.

Quick Reference

Platform Limits by Tier

Limit	Free Tier	Paid Tiers
Rules per policy	25	100
Active policies	5	Unlimited
Rule content length	5-160 characters	5-160 characters

Additional Requirements:

• Webhook must be configured at organization level before activating any policy

• Policy descriptions: 10-500 characters (optional)

Policy Structure

Start with Clear Organization

Policy Naming: Use descriptive names that reflect the policy's purpose, like "Community Guidelines" or "Product Review Standards." Policy names must be 2-100 characters. Each policy gets a unique friendly identifier (URI) automatically generated from the name for API usage.

Rule Groups: Organize rules into logical categories:

• Safety: User protection, harassment, harmful content

• Legal: Compliance, copyright, privacy, regulatory requirements

• Brand: Brand standards, messaging guidelines, competitor mentions

• Spam: Promotional content, repetitive posts, unwanted links

This organization makes policies easier to understand, manage, and tune over time.

Rule Limits and Planning

Rule Limits:

• Free tier: Up to 25 rules per policy

• Paid tiers: Up to 100 rules per policy

Active Policy Limits:

• Free tier: Maximum 5 active policies

• Paid tiers: No hard limit

Start Small: Begin with 5-10 high-impact rules covering your most critical content issues. You can always add more rules as you learn how the system performs with your content.

Writing Effective Rules

Rule Content Guidelines

Rule Names: Keep names concise but descriptive (5-80 characters)

• Good: "Personal Information Disclosure"

• Avoid: "PII" or "Must not contain any personally identifiable information including but not limited to..."

Rule Content: Be specific and clear (5-160 characters)

• Good: "text containing phone numbers, email addresses, or home addresses"

• Good: "content promoting violence or illegal drug sales"

• Avoid: "personal information" (too vague) or "bad content" (too broad)

Using Rule Presets

Deep Mod provides pre-written rule templates to help you get started quickly:

Available Presets: Financial information, personal data, HIPAA violations, and other common content categories

When to Use Presets:

• Starting a new policy in a familiar domain

• Adding standard compliance rules

• Learning effective rule-writing patterns

Customization: You can modify preset content to match your specific needs after adding them to your policy.

Cloning Existing Policies

You can clone an existing policy to create a new one with the same structure:

• Clone organization policies you've already configured

• Clone from system-provided policy templates

• Optionally override the name and description during cloning

This is useful when creating similar policies for different content types or regions.

Configuration Best Practices

Confidence Thresholds

Confidence thresholds determine how certain the system must be before flagging content. Thresholds are expressed as decimals (e.g., 0.75 = 75%).

Start Conservative: Begin with a 0.75-0.80 (75-80%) confidence threshold

• Reduces false positives while you learn system behavior

• Easier to lower threshold later than deal with user complaints

Adjust Based on Results:

• Too many false positives? Increase threshold

• Missing obvious violations? Decrease threshold

• High ambiguous rate? Consider if human review is needed

Review Mode Selection

No Review Mode (noReview): Choose when:

• You can tolerate some false negatives

• Content volume is high

• Processing speed is priority

• You don't have review team capacity

Human Review Mode (humanReview): Choose when:

• Accuracy is critical

• Content decisions have significant consequences

• You have trained reviewers available

• Regulatory compliance requires human oversight

When using Human Review mode, reviewers can make granular per-rule decisions—approving or rejecting individual rule matches rather than just the overall content result.

Common Patterns and Anti-Patterns

Effective Rule Patterns

Specific Content Types:

• "text containing credit card numbers or banking information"

• "content promoting illegal drug sales or usage"

• "messages containing threats of physical violence"

Brand Guidelines:

• "content mentioning competitor brand names"

• "text using inappropriate language for customer communications"

• "promotional content without proper disclaimers"

Anti-Patterns to Avoid

Overly Broad Rules:

• Avoid: "negative content" (too vague)

• Better: "content expressing hostility toward specific individuals"

Conflicting Rules:

• Don't create rules that contradict each other

• Example: One rule allowing "educational content about sensitive topics" while another blocks "any mention of sensitive topics"

Context-Insensitive Rules:

• Consider how legitimate content might trigger your rules

• Educational, news, and creative content often discusses sensitive topics appropriately

Testing and Iteration

Policy Testing

Use Test Mode: Always test new policies and rule changes before activating

• Test with representative content samples

• Check for unexpected false positives

• Verify confidence scores align with expectations

Gradual Rollout:

1. Test with small content samples

2. Review results and adjust rules/thresholds

3. Activate for live content monitoring

4. Monitor initial results closely

5. Fine-tune based on real-world performance

Ongoing Optimization

Regular Review: Monthly assessment of policy performance

• Check false positive/negative rates

• Review ambiguous case patterns

• Analyze which rules trigger most frequently

Rule Refinement:

• Merge similar rules that always trigger together

• Split overly broad rules that catch too much benign content

• Remove rules that never trigger or add no value

Documentation: Keep notes on why rules were added and any modifications made. This helps with future maintenance and team handoffs.

Webhook Integration

Prerequisites

Important: You must configure a webhook before activating any policy. Policy activation will fail without a valid webhook endpoint.

Configure your organization webhook in Organization Settings before attempting to activate policies.

Activation Requirements

Webhook Testing: Verify your webhook endpoint before activating policies:

• Confirms your system can receive results

• Validates response format handling

• Tests error handling for failed deliveries

Result Handling

Plan for All Result Types: Ensure your webhook handler processes all result types appropriately:

Result	Description
success	Content passed all rules
failure	Content violated one or more rules
ambiguous	Results unclear (requires human review)
abandoned	Moderation process was abandoned

Metadata Usage: Include relevant metadata with moderation requests to help with content routing and analytics

Scaling Considerations

Multiple Policies

Policy Separation: Create separate policies for different content types or risk levels:

• User-generated content vs. customer support messages

• Public content vs. internal communications

• Different geographic regions with varying standards

Shared Learning: Apply lessons learned from one policy to improve others

Tier Considerations: Remember that free tier accounts are limited to 5 active policies. Plan your policy architecture accordingly.

Team Collaboration

Clear Ownership: Assign responsibility for policy maintenance and updates

Change Management: Establish processes for:

• Requesting policy changes

• Testing and approving modifications

• Communicating updates to relevant teams

Advanced Features

AI-Powered Policy Generation

Deep Mod can automatically generate policies from existing documents:

• Upload PDF documents containing your content guidelines

• The system extracts rules and organizes them into rule groups

• Review and refine the generated policy before activating

This is useful when migrating existing moderation guidelines to the platform.

Policy Templates

System-provided templates offer pre-configured policies for common use cases:

• Use templates as starting points for new policies

• Clone and customize templates to match your specific requirements

• Templates cannot be edited directly—clone them first

What's Next

• Plan Usage & Limits - Monitor policy usage and manage plan limits

• Interpreting Results - Analyze policy performance and optimize based on results

• Human Review Workflows - Set up human oversight for ambiguous cases

Effective policy authoring is an iterative process. Start simple, test thoroughly, and refine based on real-world performance to build robust content moderation that serves your specific needs.

Need help with policy design? Contact our support team for guidance on rule writing, threshold tuning, and policy optimization strategies.