Why Best Practices Matter
As your community grows, moderation complexity scales with it. A few poorly organized policies may work in the early days, but as more users, channels, and content types are added, misconfigured rules can create chaos, over-blocking safe content, missing harmful messages, or overwhelming moderators. Best practices help ensure your moderation remains:
- Scalable: grows with your platform without breaking workflows
- Consistent: applies the same rules across environments and teams
- Maintainable: easy to test, update, and explain to new admins or auditors
Keep Policies Purpose-Built
Resist the temptation to build one giant policy that covers everything. Instead, create separate policies for distinct use cases.
Examples:
- General Chat Policy – broad coverage for day-to-day conversations.
- Livestream Policy – stricter thresholds for spam, self-harm, and harassment in live events.
This modular approach ensures each policy is easier to test and refine without unintended side effects elsewhere.
Use Clear Naming Conventions
Confusion often arises when teams manage multiple environments (dev, staging, prod) or multiple communities. Avoid “Policy1” or “DefaultPolicy.”
Best Practice:
- Adopt a consistent, descriptive naming convention such as:
- Chat-Prod-Policy
- Gaming-Community-Policy
- Livestream-Staging-Policy
This avoids costly mistakes, like applying an experimental staging policy to production by accident.
Start with Defaults, Then Tune
Stream provides strong default harms and blocklists out of the box. Start with these defaults to establish a baseline.
- Step 1: Launch with defaults enabled.
- Step 2: Monitor results in the queue, look at false positives and negatives.
- Step 3: Tune severity thresholds, harms, and actions incrementally.
Pro Tip: Don’t over-customize too early. Fine-tuning before you’ve observed real community behavior often leads to excessive false positives or missed harms.
Test in Staging Before Production
Always validate new or updated policies before going live.
- Send sample messages/images/videos that represent both harmful and safe content.
- Use the dashboard’s Test Policy feature to paste text or URLs and see how the system reacts in real time.
- Roll out changes to production only after verifying performance in staging.
This avoids disrupting real conversations with misconfigured rules.
Layer Your Rules for Coverage
Each rule type has strengths and weaknesses:
- AI Rules → catch nuance and context
- Semantic Filters → enforce plain-language standards
- Blocklists → enforce strict “do not allow” terms
- Regex Filters → catch spam patterns and obfuscations
- Domain Allowlists/Blocklists → control links
When combined, these create a defense-in-depth model: AI handles edge cases, while blocklists and regex provide hard boundaries.
Document and Version Policies
Policies aren’t static; they evolve. Without documentation, teams forget why changes were made, or worse, reintroduce old mistakes.
- Version your policies (v1, v2, etc.)
- Keep an internal changelog noting what changed, why, and who approved it
- Archive old versions for compliance and audits
This creates transparency and protects against accidental regressions.
Regularly Review and Update
Online behavior evolves constantly, slang, memes, and coded language shift over time. A policy that worked six months ago may be ineffective today.
Best Practice:
- Schedule quarterly policy reviews.
- Update blocklists with new terms.
- Evaluate false positive/negative rates.
- Adjust thresholds for harms that trend up or down.
Pro Tip: Involve moderators in reviews, they see edge cases first-hand and can surface issues that admins may miss.
Common Mistakes to Avoid
- Over-reliance on Blocklists: They’re easy to set up, but too rigid. Use them only for absolutes.
- Skipping Staging Tests: Deploying untested rules to production leads to accidental censorship or policy gaps.
- Duplicating Rules Across Policies: Leads to redundancy and inconsistency. Use modular, purpose-built policies instead.
- Unclear Naming: “Policy 1” vs. “Policy 2” causes confusion across environments. Always name by purpose and environment.
- Not Involving Moderators: Admins may design rules in isolation; moderators see how they actually perform in real workflows.
Policies are only as effective as their organization. By keeping them purpose-built, well-named, tested, documented, and regularly updated, you ensure moderation that scales smoothly with your community.
Next, we’ll examine moderation categories and labels, explaining how harms are classified, what each label means, and how to interpret them effectively in the dashboard.