Writing Standards Knowledge Pack Files

Writing Quality

A writing quality gate that detects and eliminates AI-generated writing patterns. Merges 29 detection patterns from humanizer with 5-dimension scoring from stop-slop. Use this module as the final quality check before publishing any content.

This module is content-agnostic — it works on landing pages, blog posts, social posts, emails, documentation, or any prose. It does not need brand context to function, though brand context helps with voice calibration.

See ATTRIBUTION.md for licensing details.

Two-Pass Audit Process

Pass 1: Pattern Detection

Scan the content for AI writing patterns across 6 categories. Each detected pattern is a signal that the text sounds machine-generated.

Category	Patterns	What to look for
Content	1-6	Significance inflation, notability name-dropping, superficial analyses, promotional language, vague attributions, formulaic conclusions
Language	7-13	AI vocabulary overuse, copula avoidance ("serves as" instead of "is"), tailing negations, rule of three, synonym cycling, false ranges, passive voice
Style	14-19, 26-29	Em dash overuse, boldface overuse, inline-header lists, title case, emojis, hyphenated pairs, authority tropes, signposting, fragmented headers
Communication	20-22	Chatbot artifacts ("I hope this helps!"), knowledge-cutoff disclaimers, sycophantic tone
Filler & Hedging	23-25	Filler phrases, excessive hedging, generic positive conclusions
Structural	S1-S6	Binary contrasts, negative listings, dramatic fragmentation, rhetorical setups, false agency, narrator-from-distance

Full pattern catalog with before/after examples: references/banned-patterns.md

Pass 2: Scoring & Verdict

Rate the content 1-10 on five dimensions. The scoring system measures how human the writing sounds, not whether it's "good" writing in general.

Dimension	Measures	7+ (good)	3- (bad)
Directness	Statements vs announcements	Direct assertions, confident claims	Throat-clearing, "here's why this matters"
Rhythm	Sentence length variation	Natural mix of short and long	Every sentence roughly the same length
Trust	Reader respect	States facts, lets reader conclude	Over-explains, hedges, hand-holds
Authenticity	Human voice	Specific, opinionated, some rough edges	Generic, corporate, sanitized
Density	Signal-to-noise	Every word earns its place	Filler phrases, redundant qualifiers

Threshold: 35/50 to pass. Any single dimension below 5 = automatic revision, regardless of total score.

Full rubric with scoring examples: references/scoring.md

Core Rules

These 8 rules from stop-slop are the active editing directives. Apply them when rewriting flagged content:

1. Cut filler phrases — if removing a phrase doesn't change the meaning, remove it. See references/banned-phrases.md.
2. Break formulaic structures — binary contrasts, dramatic fragments, rhetorical striptease. See references/banned-structures.md.
3. Use active voice — every sentence needs a named human subject doing something. "The decision was made" → "Sarah decided."
4. Be specific — replace vague declaratives with concrete details. "The results were significant" → "Output increased 2.4x."
5. Put the reader in the room — "you" beats "people", "users", "one." Second person creates immediacy.
6. Vary rhythm — mix sentence lengths. Two items in a list beat three. No em dashes (use commas or periods instead).
7. Trust readers — state facts directly. Skip softening ("it's worth noting that"), hedging ("arguably"), and meta-commentary ("let me explain").
8. Cut quotables — if a sentence sounds like a pull-quote or inspirational poster, rewrite it. Real writing doesn't pose.

Quick Checks

A 12-item pre-delivery checklist. Run mentally before submitting any content:

1. Any adverbs (-ly words)? Cut them or replace with stronger verbs.
2. Any passive voice? Name the actor.
3. Any false agency? (Objects doing human things: "the data tells us", "the market rewards")
4. Any paragraphs starting with Wh- words? (What, When, Where, Which, Who, Why, How) Restructure.
5. Any throat-clearing openers? ("Here's the thing:", "Let me be clear:") Delete.
6. Any binary contrasts? ("Not X. But Y.") Rewrite as a direct statement.
7. All sentences roughly the same length? Vary them.
8. Every paragraph ending punchily? That's a pattern — vary endings too.
9. Any em dashes? Replace with commas, periods, or parentheses.
10. Any vague declaratives? ("The implications are significant.") Replace with specifics.
11. Any narrator-from-distance? ("Nobody designed this.", "This happens because...") Name actors.
12. Any meta-joiners? ("And that's okay.", "Here's what I mean:", "Think about it:") Cut.

Voice Calibration

When rewriting content to remove AI patterns, don't strip all personality. Good human writing has:

• Opinions — take a stance, don't hedge everything
• Varied rhythm — mix short punchy sentences with longer flowing ones
• Specificity — concrete details, not abstract claims
• First person (when appropriate) — "I" and "we" are fine in blog posts and social
• Imperfection — not every transition needs to be smooth. Some mess is human.
• Tension — good writing creates and resolves small tensions. Don't flatten everything.

If a brand context is available, calibrate voice to match its tone and personality. The goal is to sound like a specific human writer, not a generic one.

Common Tasks

1. Full quality check — run both passes (pattern detection + scoring), return findings and verdict
2. Quick scan — pattern detection only, no scoring. For fast feedback during drafting.
3. Targeted revision — fix specific pattern categories (e.g., "fix only filler and structural patterns")
4. Score only — skip pattern detection, just rate on 5 dimensions. For content that's already been through pattern cleanup.
5. Voice calibration — analyze a writing sample and extract style fingerprint. Use when adapting content to a specific brand voice.

Tips

• Don't just remove bad patterns — inject personality. Copy that's merely "not AI-sounding" is still bland.
• Two items in a list always beat three. The rule of three is one of the strongest AI tells.
• Read the content aloud (mentally). If it sounds like a keynote speech, it's too polished.
• After cleaning up AI patterns, re-read the whole piece. Sometimes cleanup strips voice. Add it back.
• Some patterns are occasionally correct in context. An em dash can be fine in informal writing. A list of three can work if the items are genuinely distinct. Use judgment.

Gotchas

• Voice calibration can over-correct, making copy bland and lifeless. Always re-read after cleanup.
• The blanket adverb ban (rule 1) is aggressive. "Quickly" is usually cuttable, but "previously" is sometimes necessary. Apply with judgment.
• False agency (S4) is one of the strongest AI tells but also one of the hardest to fix without restructuring sentences significantly.
• Narrator-from-distance (S5) detection can false-positive on legitimate explanatory writing. Context matters.
• The 35/50 threshold is calibrated for marketing copy. Technical documentation may legitimately score lower on Rhythm and Authenticity dimensions without being "AI-sounding."

References

• references/banned-patterns.md — full 29-pattern catalog with detection signals, before/after examples, and fix guidance
• references/banned-phrases.md — categorized phrase lists: throat-clearing, emphasis crutches, jargon, adverbs, meta-commentary
• references/banned-structures.md — structural anti-patterns: binary contrasts, false agency, dramatic fragmentation
• references/scoring.md — 5-dimension rubric with detailed 1-10 descriptors and scoring examples