Skip to content
Optimizely CertificationOptimizely Certification Optimizely Certification

AI-Powered Content Review Pipeline

⏱ 40 minutes intermediate
📜CoreOpalCMPcms

Why manual content review does not scale

Section titled “Why manual content review does not scale”

Content teams produce dozens of assets per week — blog posts, landing pages, product descriptions, email copy. Every piece must comply with brand guidelines: tone of voice, approved terminology, legal disclaimers, accessibility standards. Manual review catches issues, but reviewers are bottlenecks. They miss deadlines, apply rules inconsistently, and burn out on repetitive checks.

An AI-powered review pipeline automates the consistent, rule-based parts of content review. Opal agents check every piece of content against your brand instructions before a human reviewer sees it. The human reviewer then focuses on strategic judgment — is this the right message? — rather than catching formatting errors and banned words.

Architecture overview

Section titled “Architecture overview”

This pipeline spans three Optimizely products:

┌────────────────────────────────────────────────────┐
│          CMP: Content Creation                      │
│  Authors create content in CMP tasks and briefs.    │
│  Content reaches "Ready for Review" status.         │
└────────────────────────┬───────────────────────────┘
                         │ content submitted
                         ▼
┌────────────────────────────────────────────────────┐
│          Opal: AI Brand Compliance Review           │
│  1. Agent retrieves brand guidelines via RAG        │
│  2. Agent reviews content against guidelines        │
│  3. Agent produces a compliance report:             │
│     - Tone alignment score                          │
│     - Flagged terms or phrases                      │
│     - Suggested corrections                         │
│     - Pass / needs revision verdict                 │
└────────────────────────┬───────────────────────────┘
                         │ compliance report
                         ▼
┌────────────────────────────────────────────────────┐
│          CMS: Publish Approved Content              │
│  Content that passes review is formatted and        │
│  published. Flagged content returns to CMP          │
│  for author revision.                               │
└────────────────────────────────────────────────────┘

Prerequisites

Section titled “Prerequisites”
  • Opal with Agent Builder access and RAG enabled for CMP
  • CMP with an editorial workflow that includes a “Ready for Review” status
  • CMS (SaaS) configured to receive content from your publishing workflow
  • Brand guidelines documented and stored in CMP (as briefs, documents, or assets)

Step 1: Document your brand review criteria

Section titled “Step 1: Document your brand review criteria”

Before building the agent, codify what your human reviewers check. Common review criteria include:

CategoryExample rules
Tone of voiceConversational but professional. No slang. No jargon without explanation.
Banned terms”Cheap” (use “affordable”), “Guarantee” (use “designed to”), “Best” (use “leading”)
Legal requirementsFinancial content must include a disclaimer. Health claims require sourcing.
Formatting standardsHeadlines under 70 characters. Paragraphs under 4 sentences. One CTA per page.
AccessibilityImages require alt text. Links must be descriptive (no “click here”).

Store these rules in a CMP document or brief so RAG can retrieve them.

Step 2: Create the brand review agent

Section titled “Step 2: Create the brand review agent”
  1. Navigate to Opal > Agents
  2. Click Create Agent > Specialized Agent
  3. Name it “Brand Compliance Reviewer”
  4. Write instructions that encode your review criteria:
## Objective
You review content for brand compliance. You receive a piece of content
and produce a compliance report that identifies issues and suggests
corrections.


## Process
1. Retrieve our brand guidelines from the content library
2. Read the submitted content thoroughly
3. Check against each guideline category: tone, banned terms, legal,
   formatting, accessibility
4. For each issue found, quote the problematic text and suggest a
   correction
5. Score overall tone alignment from 1-10
6. Provide a verdict: PASS (no issues), REVISE (minor issues),
   or REJECT (major compliance failures)


## Output Format
### Compliance Report
**Content:** [title of reviewed content]
**Verdict:** PASS | REVISE | REJECT
**Tone Score:** X/10


#### Issues Found
1. [Category] — "[quoted text]" → Suggestion: "[corrected text]"
2. ...


#### Summary
[2-3 sentence summary of overall compliance status]


## Constraints
- Never rewrite the entire content. Only flag specific issues.
- Do not evaluate content quality or strategy. Focus only on
  compliance with documented guidelines.
- If no brand guidelines are found via retrieval, state this
  clearly and do not fabricate rules.
  1. Enable the content retrieval tool so the agent can access brand guidelines via RAG
  2. Add an input variable: content_to_review (required, text)
  3. Save the agent

Step 3: Configure the CMP workflow trigger

Section titled “Step 3: Configure the CMP workflow trigger”

Set up your CMP editorial workflow so content routed to “Ready for Review” is sent to the Opal agent.

  1. In CMP, navigate to Workflow Settings
  2. Locate the status transition that moves content to “Ready for Review”
  3. Add an action that sends the content to the Brand Compliance Reviewer agent:
    • The content body is passed as the content_to_review input variable
    • The compliance report is attached as a comment on the CMP task
  4. Configure the workflow to:
    • Move content to “In Review” while the agent processes it
    • Move content to “Needs Revision” if the agent returns REVISE or REJECT
    • Move content to “Approved” if the agent returns PASS

Step 4: Test the review pipeline

Section titled “Step 4: Test the review pipeline”

Run through the pipeline with sample content to verify each handoff works.

  1. Create a CMP task with intentional compliance issues:
    • Use a banned term (“This is the cheapest option on the market”)
    • Omit a required disclaimer
    • Write a headline over 70 characters
  2. Move the task to “Ready for Review”
  3. Verify the Opal agent produces a compliance report
  4. Check that the report:
    • Flags the banned term with a correction
    • Notes the missing disclaimer
    • Flags the headline length
    • Returns a REVISE verdict
  5. Fix the issues and resubmit
  6. Verify the agent returns PASS on compliant content

Step 5: Route approved content to CMS

Section titled “Step 5: Route approved content to CMS”

Content that passes review moves to CMS for publication.

  1. In CMP, configure the “Approved” status to trigger content export to CMS
  2. Map CMP fields to CMS content types:
    • CMP title to CMS page title
    • CMP body to CMS content area
    • CMP metadata (author, category, tags) to CMS properties
  3. In CMS, the content author:
    • Reviews the imported content
    • Adds media, internal links, and page-specific formatting
    • Previews and publishes

Step 6: Verify the end-to-end pipeline

Section titled “Step 6: Verify the end-to-end pipeline”

Run a complete cycle from creation to publication.

  1. Create a new piece of content in CMP
  2. Submit it for review (move to “Ready for Review”)
  3. Check the Opal compliance report appears as a task comment
  4. Fix any flagged issues and resubmit
  5. Confirm the PASS verdict moves content to “Approved”
  6. Verify content arrives in CMS ready for final formatting
  7. Publish in CMS and confirm the page is live

When to use this pattern

Section titled “When to use this pattern”

This pipeline works best when:

  • Your team produces high-volume content that must comply with documented guidelines
  • Brand consistency across multiple authors is a priority
  • Review bottlenecks delay publication timelines
  • You want human reviewers to focus on strategic decisions rather than rule-checking

For low-volume or highly creative content (thought leadership, executive communications), human-only review may be more appropriate because the judgment involved goes beyond rule-based compliance.

Customization options

Section titled “Customization options”
  • Multiple review agents — Create separate agents for legal review, accessibility review, and tone review, then chain them in a workflow agent
  • Severity thresholds — Configure the agent to auto-approve content with only minor issues (REVISE with tone score above 8)
  • Feedback loops — Track which issues recur most and use that data to train authors, reducing review cycles over time