How Content is Graded

Score Range: 0-100 points across 7 categories

Letter Grades:

A+ (97-100), A (93-96), A- (90-92)
B+ (87-89), B (83-86), B- (80-82)
C+ (77-79), C (73-76), C- (70-72)
D+ (67-69), D (63-66), D- (60-62)
F (0-59)

Category Breakdown:

Substance (22 points) - Content depth, accuracy, value delivery
Voice (17 points) - Brand consistency, tone, conversational markers
Structure (17 points) - H2/H3 hierarchy, section balance, flow
Clarity (17 points) - Readability, jargon-free language, sentence length
Engagement (17 points) - Hooks, CTAs, reader connection
SEO (10 points) - Keyword placement, meta elements, search optimization
Visual (10 points) - Formatting, scannability (not penalized - images generated separately)

What Gets Detected

53 Issue Types Across 8 Categories:

Hook & Opening (5 types)

Weak opening - Generic first sentence
Generic intro - "In today's world..." phrases
Buried lede - Key point not in first 50 words
No promise - Opening doesn't state reader value
Slow start - Too much preamble

Structure (6 types)

Poor hierarchy - H3 without H2, skipped levels
Unbalanced sections - Sections with wildly different lengths
Missing H2 headers - No H2 tags at all (knocks Structure to 0)
Too few sections - Fewer than 3 H2 sections
Wall of text - Paragraphs exceeding 200 words
Overly thin sections - Sections under 100 words

Substance (7 types)

Thin content - Total word count too low
Lack of specifics - Vague claims without evidence
No examples - Missing concrete examples
Unsupported claims - Claims without backing
Missing data - No statistics or facts
Surface level - Doesn't go deep enough
Repetitive - Redundant points

Voice (12 types)

AI-isms - "Delve," "navigate," "realm," "landscape," "robust," "streamline," etc.
Clichés - "At the end of the day," "game-changer," "cutting-edge"
Conversational markers - Excessive "you," "your," "we"
Corporate jargon - "Leverage," "synergy," "holistic"
Fake specificity - "5 easy steps" without real specificity
Formulaic openings - "When it comes to," "In today's world"
Repetitive voice framing - Overuse of same framing device
Negative phrasing - "Don't miss," "avoid," "prevent"
Vague language - "Some," "many," "various"
Passive voice - Exceeding 10% of sentences
Em-dashes - Prohibited punctuation (highest priority AI signal)
Fabrication patterns - Phrases that signal made-up content

Clarity (7 types)

Long sentences - Sentences exceeding 35 words on average
Dense paragraphs - Paragraphs over 150 words
Unclear antecedents - Ambiguous pronouns
Run-on sentences - Multiple clauses without breaks
Complex vocabulary - Unnecessarily difficult words
Jargon heavy - Too much industry terminology
Poor transitions - Abrupt section changes

Engagement (6 types)

No hook - Missing engaging opening
Weak CTA - Call-to-action is generic or missing
No questions - Content doesn't engage reader thinking
Rhetorical questions - Overused rhetorical devices
No reader benefit - Doesn't explain "what's in it for me"
Boring subheads - H2/H3 tags don't create curiosity

SEO (5 types)

Missing keyword - Target keyword not in first 100 words
Keyword stuffing - Keyword density exceeds 3%
No keyword in title - Target keyword missing from H1/title
No bold keywords - Target keyword never bolded
Thin meta - Missing or poor meta description

Formatting (5 types)

BR tags - Using instead of
tags
Markdown remnants - **bold** instead of
Placeholders - [INSERT X], [ADD Y] still present
H1 tags in body - Title should be separate
Inconsistent formatting - Mixed styles

Total: 53 issue types monitored

Auto-Fixing Process

Content goes through a three-phase auto-fixing pipeline before grading:

Phase 1: Programmatic Fixes (No AI cost)

BR tags →
paragraphs
Markdown → HTML conversion
Placeholder removal ([INSERT], [ADD])
H1 tag stripping

Phase 2: Em-Dash Fixing (Highest priority)

Separate AI call
Context-aware replacements
Critical signal of AI-generated text

Phase 3: Style & Word Count (Combined call)

Negative phrasing removal
Rhetorical question removal
Cliché elimination
Formulaic opening fixes
Word count reduction (if over target by 10%+)

What is NOT Auto-Fixed:

Structure issues (heading hierarchy)
Substance issues (missing examples, thin content)
Most engagement issues (weak CTAs, no hook)
SEO issues (keyword placement)

These require regeneration with updated prompt instructions.

Penalty System

Issue Severity:

Critical: 3 points per occurrence (no cap)
High: 2 points per occurrence (capped at 10)
Medium: 1 point per occurrence (capped at 6)
Low: 0.5 points per occurrence (capped at 4)

Knockout Issues (Zero out entire category):

no_h2_headers - No H2 tags → Structure = 0 points

Example Scoring:

Content has 5 AI-isms (critical): -15 points from Voice
Content has 8 clichés (high): -10 points (capped) from Voice
Content has weak opening (high): -2 points from Engagement
Missing keyword in title (medium): -1 point from SEO
Total penalty: -28 points
Final score: 72/100 (C)

Where You See Grades

Content Hub - Grade badge next to each content piece

Content Detail Page - Full breakdown:

Overall grade (letter + score)
Category scores (7 individual scores)
Issues detected (full list with severity)
Issues fixed (what was auto-corrected)
Issues remaining (what needs manual attention)

Brief Detail Page - If content generated from brief

Email Reports - Content generation completion emails include grade

Improving Your Grade

For Substance Issues:

Regenerate with "Add specific examples" instruction
Manually add data, statistics, case studies
Deepen analysis in weak sections

For Voice Issues:

Use Learning Quality System (platform learns from your edits)
Mark false positives as "Not an issue" to suppress detection
Regenerate with updated brand voice samples

For Structure Issues:

Manually fix heading hierarchy (H2 → H3 nesting)
Add more H2 sections if too few
Break up long sections into subsections

For Clarity Issues:

Break long sentences into shorter ones
Simplify complex vocabulary
Add transitions between sections

For Engagement Issues:

Add strong opening hook (first 2 sentences)
Include clear CTAs in conclusion
Make subheadings more curiosity-driven

For SEO Issues:

Manually bold target keyword 2-3 times
Add keyword to first 100 words
Ensure keyword in title/H1

Learning Quality System

The platform learns from your edits to improve over time:

How It Works:

Content is graded with issues detected
You edit the content or mark issues as "Not an issue"
Platform tracks which issues you consistently ignore or fix
Future content auto-prevents recurring issues per tenant
False-positive detections get suppressed

Result: Quality improves 60% in first month of use

Example:

Week 1: "Rhetorical questions" flagged 12 times
You mark 8 as "Not an issue" (these are strategic questions)
Week 4: Platform stops flagging strategic questions in your content
Only excessive rhetorical questions get flagged

Quality Monitoring System

ILLIXIS runs background monitoring to detect quality issues across your content generation pipeline.

Brief Quality Monitoring

Schedule: Daily

Checks for:

Invariant violations: Briefs missing required fields (keyword, title, content strategy)
SERP pending: Briefs stuck waiting for SERP analysis
Incomplete data: Briefs with partial analysis data
Error rates: Percentage of briefs failing during creation

What happens when issues are detected:

Dashboard notification appears in Strategy Hub
If error rate exceeds 10%, system alert is logged
Admins can view details in Settings > System Health

Content Generation Monitoring

Schedule: Daily

Monitors:

Truncation rate: Content cut off due to token limits
Chart failures: Charts that failed to render
Image failures: Hero images that didn't generate
Success rates by brief type: Which brief types produce best results

Quality metrics tracked:
| Metric | Threshold | Action if exceeded |
|--------|-----------|-------------------|
| Truncation rate | >5% | Prompt optimization review |
| Chart failure rate | >10% | Playwright health check |
| Image failure rate | >15% | AI image generation investigation |
| Overall error rate | >10% | System alert generated |

Viewing quality metrics: Settings > System Health shows:

7-day rolling quality metrics
Breakdown by brief type (keyword, trend, link magnet, etc.)
Comparison to previous week

Automation Schedule

Content grading runs automatically at key points to ensure your content quality is always tracked:

On Content Generation:

Grading runs immediately when new content is generated
Auto-fixing pipeline (Phase 1-3) executes before grading
Grade is assigned and stored with the content piece
Improvement suggestions are generated alongside the grade

Weekly Bulk Re-Grading:

All existing content is re-graded every Saturday at 2:00 AM UTC
This catches improvements from Learning Quality System updates
Re-grading uses latest detection rules and thresholds
Only content that has changed since last grading is re-processed

Grade History:

Every grading run is stored for trend analysis
View grade changes over time in Content Detail page
Track how your content quality evolves month-over-month
Identify patterns in which categories improve or decline

Improvement Suggestions:

Fresh suggestions generate with each grading run
Suggestions prioritized by impact (highest point gains first)
Suggestions become more refined as platform learns your preferences
Stale suggestions (issues already fixed) are automatically cleared

Grade vs. Publish Decision

Grade ≠ Ready to Publish

A C-grade article with strong substance may outperform an A+ article with thin value.

Consider:

B+ or higher - Usually ready with minor tweaks
B to C+ - Review issues, fix critical ones
C to D - Significant issues, consider regeneration
F - Major problems, regenerate with better prompt

Strategic Exception:

Pillar content targeting high-competition keywords: aim for A- or higher
Quick blog posts for long-tail keywords: B is often sufficient
Social repurposing content: C+ is fine (brevity over depth)

Viewing Grade Details

Navigate: Content Hub → [Content Piece] → View Details

You'll see:

Overall grade badge
Score breakdown by category
Full issue list (detected, fixed, remaining)
Recommendations for improvement
Grading checklist (what was checked)

To regrade after manual edits:

Click "Regrade Content" button
Platform re-runs detection on current content
New grade calculated based on remaining issues

FAQs

Q: Why did my content get a C when it looks good? A: Check the category breakdown. You may have high substance but poor structure (no H2 headers) or excessive AI-isms in voice.

Q: Can I disable certain issue detections? A: Use the Learning Quality System. Mark false positives as "Not an issue" and the platform will stop flagging them for your tenant.

Q: Why aren't all issues auto-fixed? A: Some issues (structure, substance, engagement) require regeneration with updated prompts. Auto-fixing is limited to style and formatting issues that the AI can safely correct without changing meaning.

Q: Does a higher grade mean better SEO performance? A: Not necessarily. Grade measures quality, not keyword targeting or search intent match. A B-grade article perfectly matching search intent will outrank an A+ article targeting the wrong intent.

Q: What's the average grade for generated content? A: First-generation content typically scores B to B+ (83-89). After auto-fixing and learning from your edits, average grade improves to A- (90-92) within 30 days.

Q: Why is Visual worth 10 points if it's not penalized? A: Visual scoring was removed December 2024 because images are generated separately. The 10 points were redistributed to other categories (Substance, Voice, Structure, Clarity, Engagement each gained 2 points).

Q: Can I export grade data? A: Yes. Export grade data via Content Hub filters or API access.

Content Grading