Content Reviewer

Overview

Review written content to detect AI-generated patterns and suggest improvements for more natural, human voice. Works best on blog posts, marketing copy, and documentation.

When to Use

• After drafting content, before publishing
• When content "sounds AI-generated"
• When polishing prose for publication
• User asks to review content for AI-isms

Workflow

digraph review_flow {
    "Read content" [shape=doublecircle];
    "Pass 1: AI patterns" [shape=box];
    "Pass 2: Style issues" [shape=box];
    "Pass 3: Structure" [shape=box];
    "Present findings with options" [shape=box];
    "User approves changes" [shape=doublecircle];

    "Read content" -> "Pass 1: AI patterns";
    "Pass 1: AI patterns" -> "Pass 2: Style issues";
    "Pass 2: Style issues" -> "Pass 3: Structure";
    "Pass 3: Structure" -> "Present findings with options";
    "Present findings with options" -> "User approves changes";
}

Pass 1: AI Pattern Detection

Scan for these common AI-isms and flag with line numbers:

Pattern	Example	Issue
"Picture this:" openers	"Picture this: a user logs in..."	Cliché AI opener
Rhetorical question chains	"What if you could X? Tap into Y?"	Formulaic
Contrast structures	"The real unlock isn't X. It's Y"	Overused by AI
"Think of it as..."	"Think of it as a unified account"	AI filler phrase
"Consider how/a..."	"Consider how markets work:"	Sometimes ok, often AI-ish
Parallel structures	"What X does for Y, Z does for W"	Too symmetrical, formulaic
"This is a step towards..."	Generic progress language	Vague
"lays the foundation"	Cliché	Context-dependent
"This creates a flywheel"	Buzzy	Context-dependent
"hub" as noun	"interoperability hub"	Overused buzzword
Back-to-back rhetorical questions	"What if X? What about Y?"	AI pattern
Excessive bullet points	"How it will work: - X - Y - Z - W"	AI formatting, reads sloppy

Preferred Alternatives

AI Pattern	Human Alternative
"Picture this: X"	Just state X directly
"What if you could X?"	"You could X"
"The real unlock isn't X. It's Y"	Cut, or just state Y
"Think of it as X"	"It's X" or use a dash
"Consider how X works:"	"X already works this way:"
"What X does for Y, Z does for W"	"Z plays the same role for W"
Excessive bullet points	Convert to flowing prose; use bullets only for truly parallel items

Pass 2: Style Issues

Check for:

• Redundancy - Same point made twice in different sections
• Inconsistent formatting - Some bullets bold, some not
• Long sentences - Could be split for clarity
• Vague language - Could be more concrete/specific
• Salesy tone - "That's volume you're not capturing" → "That's a user you're losing"
• Jargon - Technical terms without explanation
• "settlement" vs "clearing" - "Clearing" is often better (DTCC does clearing)
• Excessive bullet points - Lists that should be prose; bullets only work for truly parallel, scannable items

Pass 3: Structure

• Does each section earn its length?
• Are examples concrete and current? (Avoid dated references)
• Does the flow feel natural?
• Are there abrupt transitions?
• Is there a clear narrative arc?

Output Format

For each issue found, present as a table:

| Line | Current | Suggested | Issue |
|------|---------|-----------|-------|
| 36 | "Picture this: a user..." | "A user..." | AI opener |
| 44 | "What if you could X? Y?" | "You could X, or Y." | Rhetorical chain |

Then for significant changes, offer options:

**Line X:** [Current text]
**Issue:** [What's wrong]
**Options:**
1. [Alternative A]
2. [Alternative B]
3. [Cut entirely]

Section-by-Section Review

When doing a full review, go section by section:

### **[Section Name] (lines X-Y)**
**Verdict:** ✓ Good | Needs minor tweaks | Needs work
**Issues:**
- [specific issues with line numbers]
**Suggestions:**
- [specific suggestions]

Final Summary

After all passes, provide:

## Summary
- **AI patterns found:** X
- **Style issues:** X
- **Recommendation:** Ready to ship | Minor tweaks needed | Significant revision needed

**Remaining TODOs:**
- [any placeholders or missing content]

Interaction Style

• Present options, don't force changes
• Be direct about issues
• Acknowledge when something is fine ("✓ Good")
• Know when to stop - diminishing returns are real
• Let user choose between alternatives
• Don't over-edit - preserve author's voice

Critical Constraints

DO:

• Flag specific lines with issues
• Provide multiple options for fixes
• Be honest about quality
• Stop when content is good enough

DON'T:

• Rewrite entire sections without asking
• Force changes the user doesn't want
• Keep nitpicking after major issues are fixed
• Remove personality/voice from content