What Gets Detected

The Pattern Detection Library monitors 48 issue types across 8 categories:

1. Hook & Opening (5 Types)

Weak Opening: Generic first sentences that don't grab attention
Generic Intro: "In today's world," "When it comes to," "It's no secret that"
Buried Lead: Key point isn't in the first 50 words
No Promise: Opening doesn't tell readers what they'll learn
Slow Start: Too much preamble before delivering value

2. Structure (6 Types)

Poor Hierarchy: H3 without H2, skipped heading levels
Unbalanced Sections: Some sections 2x longer than others
Missing Transitions: Abrupt jumps between sections
Weak Conclusion: Article ends abruptly or trails off
No Intro: Jumps straight to H2 without opening paragraph
Single-Item Lists: Lists with only one bullet point

3. Substance (8 Types)

Too Short: Article is significantly below target word count
Thin Section: Section has less than 100 words of substance
Repetition: Same point made multiple times
Off-Topic: Content strays from the main subject
No Examples: Claims without supporting examples
No Data: Missing statistics or research citations
No Expert Voice: Generic advice without expertise signals
Surface Level: Covers topic without depth

4. Voice & Style (9 Types)

AI-isms: "Delve," "embark," "realm," "tapestry," "navigating"
Clichés: 80+ overused phrases (see full list below)
Negative Phrasing: "It's not X, it's Y" constructions
Fabricated Scenes: Invented specific moments with unnamed people
Repetitive Voice Framing: Same pattern used multiple times ("In my experience...")
Formulaic Openings: 18 generic article starters
Corporate Jargon: "Synergy," "leverage," "paradigm shift"
Conversational Issues: Too formal or too casual
Brand Voice Mismatch: Doesn't match your established tone

5. Clarity (7 Types)

Passive Voice: Excessive use of passive constructions
Long Sentences: Sentences over 40 words
Dense Paragraphs: Paragraphs over 150 words
Complex Language: Unnecessarily complicated vocabulary
Vague Language: "Things," "stuff," "a lot of," "various"
Filler Phrases: "It is important to note that," "as mentioned earlier"
Unclear Pronouns: Ambiguous "it," "this," "they" references

6. Engagement (5 Types)

No Questions: Article never engages reader with questions
No Personal Examples: All abstract, no concrete stories
No Action Steps: Advice without clear next steps
No Call-to-Action: Article ends without directing reader
Monotone Rhythm: Every sentence same length/structure

7. SEO (4 Types)

Keyword Missing: Target keyword doesn't appear naturally
No Internal Links: Article has no internal link opportunities
Missing Meta Elements: No meta description or title optimization
Thin Alt Text: Images missing descriptive alt attributes

8. Formatting (4 Types)

Placeholder Text: [BRAND NAME], [INSERT X], [TODO]
Fake Specificity: Manufactured details that seem authentic ("last Tuesday at 3:42pm")
BR Tags: tags instead of proper paragraphs
Markdown Artifacts: Unconverted markdown in HTML output

Pattern Categories in Detail

Clichés (80+ Patterns)

Opening Clichés:

"Picture this"
"In today's world"
"Let's dive in"
"Imagine this"
"Have you ever wondered"

Filler Clichés:

"At the end of the day"
"When it comes to"
"Needless to say"
"The bottom line is"
"There's no denying"

Transition Clichés:

"That being said"
"With that in mind"
"On the other hand"

Business Clichés:

"Game changer"
"Think outside the box"
"Best practices"
"Low-hanging fruit"
"Move the needle"

AI-isms (40+ Patterns)

Typical AI Verbs:

Delve, delving
Embark, embarking
Navigate, navigating
Unlock, unlocking
Unravel, unraveling

AI Nouns:

Realm, landscape
Tapestry, mosaic
Nuance, nuanced
Myriad
Cornerstone

AI Phrases:

"Can I be honest"
"Here's the thing"
"The truth is"
"It's important to note"

Negative Phrasing (9 Patterns)

These "It's not X, it's Y" constructions feel artificial:

"It's not just about X, it's about Y"
"It isn't about X, it's about Y"
"This isn't your typical X"
"Don't just think of X, think of Y"
"Forget everything you know about X"

Placeholder Text (13 Patterns)

Unfilled template placeholders:

[Brand Name]
[Insert X here]
[Your company name]
[Add example]
[TODO]
Lorem ipsum

Fake Specificity (10 Patterns)

Manufactured details that seem authentic but aren't:

"Last Tuesday at 3:42pm"
"On March 15th when I..."
"Standing in my closet at..."
"67.8% of people..."
"Precisely at 9am..."

Conversational Markers (20 Patterns)

Good conversational writing includes:

"Here's what I've noticed"
"What surprised me"
"Personally"
"In my experience"
"Here's why"
"Pro tip"

These are POSITIVE signals. The system tracks them to identify engaging, human writing.

Corporate Jargon (40+ Terms)

Business speak to avoid:

Leverage, synergy, paradigm shift
Holistic, ecosystem, bandwidth
Deliverables, stakeholders, actionable
Low-hanging fruit, circle back
Touch base, deep dive, drill down

Formulaic Openings (18 Patterns)

Generic article starters detected in first 500 characters:

"In this article, we will..."
"Are you struggling with..."
"Welcome to this guide"
"According to the dictionary"
"Have you ever wondered"

How Detection Works

Automatic Scanning

Every generated article gets scanned immediately after creation. The system:

Strips HTML tags for pure text analysis
Runs pattern matching across all 200+ patterns
Logs detected issues with line numbers
Assigns severity levels (Critical, High, Medium, Low)
Groups issues by category for display

Issue Severity Levels

Critical - Must fix before publishing (placeholder text, broken formatting) High - Strong recommendation to fix (weak openings, AI-isms, clichés) Medium - Should fix for quality (poor transitions, vague language) Low - Nice-to-fix (minor style issues)

Viewing Detected Issues

After generating an article, scroll to the Content Quality section to see:

Total issue count by severity
Issues grouped by category
Specific examples with context
Auto-fix availability for each issue

Auto-Fixing Issues

The system can automatically fix many detected issues through three methods:

1. Programmatic Fixes (No AI Cost)

Converts tags to proper paragraphs
Removes placeholder text
Strips H1 tags (one per page)
Converts markdown artifacts to HTML

2. AI-Powered Fixes

Rewrites weak openings
Removes clichés and AI-isms
Fixes negative phrasing patterns
Strengthens conclusions
Adds transitions

3. Manual Review Required

Poor heading hierarchy (requires structural changes)
Missing internal links (requires content knowledge)
Thin sections (needs more substance)

Running Auto-Fix

Click Auto-Fix Issues on any article to trigger the three-phase fixing process:

Em-dash fixing (highest priority)
Style editing (clichés, AI-isms, negative phrasing)
Word count reduction (if needed)

Programmatic fixes run first, then AI-powered fixes. The system shows a progress indicator during processing.

Learning Quality System

The Pattern Detection Library learns from your edits to improve over time.

How Learning Works

When you edit an article after generation:

System compares original vs. edited version
Identifies which detected issues you actually fixed
Tracks which "issues" you ignored or kept
Adjusts future detection sensitivity per tenant

False Positive Suppression

If the system flags something that's actually correct for your brand voice, the system learns:

You kept "leverage" 3 times → Stops flagging it as jargon
You kept first-person voice framing → Recognizes it as your style
You kept specific industry terms → Understands your niche vocabulary

Improvement Timeline

Week 1: Baseline detection (some false positives expected) Week 2-3: System learns your edit patterns Week 4: 60% improvement in detection accuracy Month 2+: Highly tuned to your specific voice and style

Viewing Learning Stats

Go to Settings > Content Quality to see:

Total issues detected
Auto-fix success rate
False positive trends
Most common issue types

Multilingual Support

Pattern detection works across 6 languages:

English (EN)

Full 200+ pattern library
All 48 issue types

Spanish (ES)

80+ Spanish-specific patterns
Clichés: "Imagina esto," "En el mundo de hoy"
Conversational markers: "Te cuento algo," "La verdad es que"
Reader address detection (tú/usted forms)

Other Languages

Pattern detection for Portuguese, French, German, and Italian focuses on:

Structural issues (heading hierarchy, transitions)
Formatting issues (placeholders, BR tags)
Passive voice percentage
Sentence length

Language-specific cliché and AI-ism detection is under development.

Best Practices

1. Review Before Auto-Fixing

Scan the detected issues list first. Some "issues" might be intentional for your brand voice.

2. Don't Blindly Trust Detection

The system is good, not perfect. A flagged phrase might be exactly right for your audience.

3. Fix High-Severity Issues First

Focus on Critical and High issues. Low-severity items are optional polish.

4. Use Learning to Your Advantage

Keep editing articles naturally. The system learns from every edit you make.

5. Check After Auto-Fix

Always review auto-fixed content. ILLIXIS makes intelligent edits, but you know your brand best.

6. Dismiss Irrelevant Issues

If an issue type never applies to your content, dismiss it consistently so the system learns.

FAQ

Q: Why does the system flag phrases I like? The Pattern Detection Library starts with generic "good writing" rules. As you edit and keep certain phrases, it learns your preferences. Give it 2-3 weeks to tune to your voice.

Q: Can I disable specific pattern checks? Not directly, but the learning system effectively disables checks you consistently ignore. If you keep "leverage" every time, it stops flagging it.

Q: Does auto-fix change my article's meaning? No. Auto-fix focuses on style improvements (removing clichés, strengthening openings) without altering factual content or core arguments.

Q: How many issues are too many? For a 1,500-word article:

0-5 issues: Excellent
6-15 issues: Good (minor cleanup)
16-30 issues: Needs work (auto-fix recommended)
30+ issues: Major revision needed

Q: What if auto-fix makes it worse? Every auto-fix creates a new version. You can always revert by clicking Regenerate to go back to the original.

Q: Why don't I see any issues on some articles? Either the article is genuinely clean, or the learning system has adapted to your style and no longer flags patterns you consistently keep.

Q: Can I see which patterns were detected? Yes. Click View Details on any issue category to see specific examples with surrounding context and line numbers.

Q: Does pattern detection affect my quota? No. Pattern detection runs automatically at no cost. Only auto-fixing consumes quota.

Q: What's the difference between pattern detection and content grading? Pattern detection identifies specific issues. Content grading assigns an overall letter grade (A+ to F) based on 7 categories. They work together: pattern detection feeds into the grading system.

Pattern Detection vs. Human Editor

The Pattern Detection Library catches mechanical issues (clichés, AI tells, placeholders) but can't judge:

Strategic fit: Does this content serve your business goals?
Audience resonance: Will your specific audience connect with this?
Brand alignment: Does this sound like YOUR brand?
Factual accuracy: Are claims correct and current?

Think of pattern detection as line editing, not developmental editing. It makes good writing better, but can't fix fundamentally wrong strategy.

Pattern Detection Library