How to Create a Custom AI Content Reviewer

A custom AI content reviewer evaluates your content against your criteria — not generic grammar rules or readability scores. You define what matters (brand voice, accuracy, SEO structure, compliance), assign weights to reflect priority, and the reviewer scores every piece against that definition.

This tutorial walks through creating your first reviewer, from choosing criteria to rolling it out to your team.

Quick answer: Create a reviewer in 4 steps: (1) name it and describe what it checks, (2) write a system prompt that defines the reviewer's expertise, (3) add 4-6 weighted evaluation criteria, (4) optionally set a quality gate. The whole process takes 10-15 minutes manually or 30 seconds with Create with AI.

Before You Start: What You Need

You need answers to three questions:

1. What content type will this reviewer evaluate? — Start with your highest-volume type (usually blog posts or marketing emails).
2. What do you review for most often? — The feedback you give repeatedly becomes your criteria. If you always comment on brand voice, readability, and CTA quality, those are your first three criteria.
3. What matters most? — Not all criteria are equally important. For a compliance document, accuracy is more important than readability. For a social post, brand voice is more important than SEO.

Step 1: Name and Describe Your Reviewer

Give your reviewer a descriptive name that tells your team what it's for:

Good names:

• "Blog Post Quality Reviewer"
• "Client X Brand Voice Checker"
• "Healthcare Compliance Reviewer"
• "Email Campaign Effectiveness"

Bad names:

• "Content Reviewer" (too generic — you'll have multiple reviewers)
• "AI Checker" (doesn't describe what it checks)
• "Test" (you'll forget what it does)

The description should explain what this reviewer evaluates and when to use it. Write it for the writer who submits content, not for yourself.

Example: "Evaluates blog posts for brand voice consistency, readability, SEO structure, and CTA effectiveness. Use this for all blog content before submitting to the editor. Target score: 75+."

Step 2: Write the System Prompt

The system prompt is the most important configuration. It tells the AI how to review content — its expertise, focus areas, feedback style, and any domain-specific instructions.

A good system prompt includes:

Role Definition

Tell the AI what kind of reviewer it is:

"You are a senior content editor specialising in B2B SaaS marketing content. You have 10+ years of experience reviewing blog posts, whitepapers, and marketing emails for technology companies."

Focus Areas

Tell the AI what to pay attention to:

"Focus on: (1) whether the content provides genuinely useful information rather than surface-level advice, (2) whether the brand voice is confident and direct without being arrogant, (3) whether the structure supports scanability with clear headings and short paragraphs, (4) whether the CTA is specific and relevant to the content topic."

Feedback Style

Tell the AI how to deliver feedback:

"Provide specific, actionable feedback. Instead of 'improve readability,' say 'Paragraph 3 contains a 45-word sentence — break it into two sentences.' Instead of 'brand voice is off,' say 'The phrase "it's important to note" is passive hedging — replace with a direct statement.' Give both positive feedback (what works well) and improvement areas."

Domain Context

Add any knowledge the reviewer needs:

"Our target audience is content team leads at 10-100 person companies. Our content should read at FK Grade 8-9. We never use: 'leverage,' 'utilise,' 'cutting-edge,' 'industry-leading,' or 'in today's digital landscape.' Our brand voice is confident, helpful, and human — we write like a smart friend, not a corporate brochure."

The Shortcut: Create with AI

Don't want to write a system prompt from scratch? Use Create with AI. Describe what you need in a sentence — "A reviewer for healthcare marketing emails that checks patient privacy language, empathetic tone, and clear calls to action" — and AI generates the complete configuration.

→ Guide: Create with AI: Generate a Reviewer in 30 Seconds

Step 3: Add Evaluation Criteria

Criteria are the specific dimensions your reviewer measures. Each criterion has:

• Name — what it evaluates (e.g., "Brand Voice")
• Description — what specifically to check within this criterion
• Weight — how important it is relative to other criteria (1-5 scale or percentage)

Recommended Criteria by Content Type

Blog Posts:

Marketing Emails:

Product Descriptions:

Writing Good Criterion Descriptions

The description tells the AI what to look for within each criterion. Be specific:

Vague (bad): "Check if the content has good readability."

Specific (good): "Evaluate sentence length (target: average 15-20 words, no sentence over 35 words), paragraph length (target: 2-4 sentences), use of transition words, active vs. passive voice ratio (target: >80% active), and Flesch-Kincaid grade level (target: 8-9 for blog posts)."

The more specific your descriptions, the more useful the feedback.

→ Template: Content Scoring Rubric Template

→ Deep dive: How to Set Up Weighted Evaluation Criteria

Step 4: Set a Quality Gate (Optional but Recommended)

A quality gate is a minimum score that content must achieve to pass. Set it on any reviewer to create an objective pass/fail standard.

Starting recommendations:

Start lower than you think you should. A gate that's too high initially frustrates writers and creates resistance. A gate that's too low doesn't change behaviour. Find the sweet spot, then raise it by 5 points every quarter.

→ Guide: Content Quality Gates: Setting Standards

Step 5: Upload Your Knowledge Base

This step is optional but makes a dramatic difference in reviewer quality.

Upload your brand guidelines, style guide, product documentation, or any document that contains information the reviewer should know. The AI uses this context to give feedback specific to your organisation.

Without knowledge base: The reviewer gives generic feedback — "Consider whether the tone matches your brand guidelines."

With knowledge base (brand guidelines uploaded): The reviewer gives specific feedback — "The phrase 'leverage our platform' violates your brand guideline prohibiting 'leverage.' Your preferred alternative is 'use TeamBench to...' (brand guide p.3)."

The minimum useful knowledge base: your brand voice guidelines document.

Step 6: Test With Real Content

Before rolling out to your team, test the reviewer with 5-10 real pieces of content — a mix of recent published content (which should score well) and drafts or older content (which should score lower).

What to check:

• Are the scores reasonable? Published content should score higher than drafts.
• Is the feedback actionable? Each feedback point should tell the writer specifically what to change.
• Are the criteria weighted correctly? If a low-importance criterion is dragging down scores disproportionately, adjust the weight.
• Is the quality gate at the right level? If all content passes easily, the gate is too low. If nothing passes, it's too high.

Adjust criteria descriptions, weights, and the quality gate based on your test results. This calibration step takes 15-20 minutes and dramatically improves the reviewer's usefulness.

Step 7: Roll Out to Your Team

Once the reviewer is calibrated:

1. Share it with your team — add team members to the project where the reviewer lives
2. Explain the workflow — writers submit content → get scored feedback → improve → re-submit → pass the gate → submit for human review
3. Set expectations — the first week will be an adjustment. Scores will be lower than expected. That's normal and the point.
4. Monitor score trends — use the Score Trends dashboard to track team progress. Scores should improve noticeably within the first 2-3 weeks.

What Happens Next

After your first reviewer is running, you'll naturally want to:

• Create content-type-specific reviewers — different criteria for emails vs. blog posts vs. social media
• Add a brand voice reviewer — dedicated to checking brand consistency
• Set up panel reviews — submit to multiple reviewers simultaneously for high-stakes content
• Build knowledge bases — upload more context documents for more specific feedback

→ Start building: Create your first reviewer

→ Related:

• The Complete Guide to AI Content Review
• Best Practices for AI Content Reviewers
• Content Scoring Rubrics: The Definitive Guide
• 10 Reviewer Templates for Marketing Teams