Dehumidifier skill

The Dehumidifier — AI Writing Tell Removal Skill

A systematic protocol for detecting and eliminating AI-generated writing patterns. Named because it removes the artificial moisture that bloats and stiffens AI prose.

---

Core Principle

Don't swap one template for another. Replacing "Moreover" with "Picture this:" is still a tell. The goal is authentic voice with natural irregularity — not a cleaner-looking AI pattern.

Make multiple passes. Patterns hide behind other patterns.

---

PASS 1 — Title & Opening Audit

Title Red Flags

• Colon addiction — "Topic: A Practical Guide" structure
• X and Y formulations — "Benefits and Challenges of..." / "Opportunities and Risks in..."
• The [Adjective] [Noun] — "The Digital Transformation" / "The Essential Guide"
• Gerund openings — "Understanding...", "Navigating...", "Exploring..."
• How-to/Why/What titles — formulaic question-based titles

Fix: Title should reflect a specific point of view, not a filing system label.

Opening Red Flags

• Scene-setting openers — "Picture this:", "Imagine:", "Consider this scenario:"
• In medias res attempts — "You're staring at a blank screen..." (trying too hard)
• Temporal markers — "In today's world,", "In recent years,", "As we move forward"
• Question openers — "Have you ever wondered...?" / "What if I told you...?"

Fix: Start mid-thought. Drop the warm-up. The first sentence should do actual work.

---

PASS 2 — Sentence Structure Audit

Structure Red Flags

• Hypotactic over-complexity — excessive subordinate clauses daisy-chained with "while," "although," "whereas"
• Front-loaded transitions — 80%+ of sentences begin with "Moreover," "Furthermore," "Additionally," "However"
• Parallel structure obsession — "to explore, to examine, and to evaluate" (triplet rigidity)
• Em dash dependency — using em dashes for pseudo-profound contrasts ("not X — but Y")
  • Watch for: "Clear the grunt work – Generate boilerplate..."
  • Watch for: "not just about the destination — it's about the journey"
• Contraction overcorrection — either too many ("we're", "you're" everywhere) or none at all

Clause Red Flags

• Anticipatory "it" — "It is important to note that..." / "It should be emphasized that..."
• Existential "there" — "There are several factors that..." / "There exists a need for..."
• Passive voice hedging — "It has been shown that..." / "It can be argued that..."
• Cleft sentences — "It is through X that Y occurs" (false emphasis structure)

Fix: Rewrite as direct subject-verb statements. Kill the preamble.

---

PASS 3 — Vocabulary Audit

Word Choice Red Flags

• Academic register inflation — "utilize" (use), "commence" (start), "demonstrate" (show)
• Buzzword density — "leverage," "synergy," "paradigm shift," "revolutionary," "transformative"
• Modifier stacking — three or more adjectives per noun: "comprehensive, multifaceted, innovative approach"
• Latin/Greek preference — choosing Latinate over Germanic words ("demonstrate" vs. "show")

Stock Phrase Red Flags

• "plays a crucial role in..."
• "dive deep"
• "unpack the nuances"
• "navigate the complexities"
• "in today's rapidly evolving world"
• "it's worth noting that"
• "significant," "notable," "considerable" (semantic bleaching — sounds meaningful, adds nothing)
• "numerous," "various," "multiple" without specifics

Fix: Use the simpler word. Use the specific number. Cut the filler phrase entirely.

---

PASS 4 — Paragraph & Structure Audit

Organization Red Flags

• Topic sentence rigidity — every paragraph opens with an obvious thesis statement
• Mechanical transitions — same transition words rotating predictably
• Five-paragraph essay syndrome — intro, three body paragraphs, conclusion
• Circular conclusions — ending by restating the introduction nearly verbatim
• Paragraph length uniformity — consistent 3-5 sentences throughout, no variation
• Mechanical paragraph structure — topic sentence → 2-3 supporting sentences → transition, every time

Cohesion Device Red Flags

• "As previously mentioned," / "As we have seen," / "Building upon this"
• Additive conjunction chains — "Also... Moreover... Furthermore... Additionally..." in sequence
• Overuse of "This approach," "These factors," "That concept"

Fix: Vary paragraph length deliberately. Use single-sentence paragraphs for punch. Let some paragraphs run long when the idea warrants it.

---

PASS 5 — Logic & Argumentation Audit

Content Red Flags

• Surface-level analysis — stating obvious points as if profound
• False dichotomies — "Not just X, but also Y" when both are obviously true
• Hedging overload — "Perhaps," "might," "could potentially," "may possibly" (pick one or cut)
• Tautological statements — "Success requires successful strategies"
• Both-sides-ism — including counterpoints even when they don't strengthen the argument
• Exhaustive listing — covering every possible angle instead of the important ones
• Defensive writing — excessive qualifiers to avoid being wrong

Information Structure Red Flags

• Given-new violations — presenting known information as novel insights
• Coverage anxiety — every potentially relevant point included rather than what matters

Fix: Make a claim. Commit to it. Cut the hedge or keep it — not both.

---

PASS 6 — Rhetorical Device Audit

Device Red Flags

• Rule of three addiction — everything in triplets: "efficient, effective, and elegant"
• False balance — "On one hand... on the other hand" for non-opposing ideas
• Mechanical rhetorical questions — used to transition between sections
• Rhythmic monotony — similar sentence lengths throughout with no variation
• Fortune cookie endings — "Those who keep their fundamentals sharp will ship better software, faster"

Analogy & Metaphor Red Flags

• Crystalline/technical metaphors — "crystal balls," "wiring models into shells"
• Forced workplace analogies — comparing everything to tools, crafts, or journeys
• Mixed metaphor chains — too many metaphors in succession
• Time-relatability attempts — "while your coffee cools" (trying too hard)

---

PASS 7 — Verbosity Audit

Expansion Red Flags

• Elaboration compulsion — never letting a simple statement stand: "The sky is blue" → "The sky exhibits a blue coloration due to atmospheric light scattering"
• Example inflation — three examples when one would do
• Redundant clarification — "In other words," "That is to say," "To put it another way" (pick one, cut the rest)

Padding Red Flags

• Nominalization excess — "give consideration to" → "consider"; "make a decision" → "decide"
• Periphrastic constructions — "is able to" → "can"; "has the ability to" → "can"
• Filler phrases — "It is worth noting that," "It should be mentioned that," "It goes without saying"
• Unnecessary precision — "approximately 7-10 days" → "about a week"

---

PASS 8 — Fake-Casual / "Humanizing" Attempt Audit

These patterns appear when AI tries to sound human and fails. They survive most editing passes.

Conversational Template Red Flags

• Fake casual openers — "You know what?", "Here's the thing:", "The real [x]?", "But here's the catch —"
• Ellipsis abuse — using "..." to simulate thoughtful pauses
• Calculated informality — strategic profanity or "gonna/wanna" that doesn't match the rest of the voice
• Meta-commentary interruptions — "(Did you notice I just...?)" / "(See what I did there?)"
• Performance markers — "Watch this:", "Check it out:", "Get this:"

Structural Tells That Survive "Humanizing"

• Em dashes for definition — "Clear grunt work – Generate boilerplate" (persists even after edits)
• List items starting with verbs — "Generate...", "Flag...", "Propose..." (action-verb bullet lists)
• Binary constructions — "juniors climb... seniors stay..."
• "Of course" as transition — sounds natural but AI overuses it
• The payoff framing — "For teams, the payoff is..."
• Industry insider voice — "Nobody misses the slog" (performing experience rather than having it)
• Confident prediction endings — "will ship better software, faster"
• Temporal bridge phrases — "Researchers are already..." / "AI will not... yet it is already..."
• Setup-list-conclusion format — opening paragraph → bulleted list → philosophical ending

Template Switching Red Flags

Watch for swapping one AI template for another:

• "Moreover" → "Picture this:" ❌ still a tell
• "It is important to note" → "Nobody misses" ❌ still a tell
• "In conclusion" → "Those who... will..." ❌ still a tell
• Academic voice → Fake startup blog voice ❌ still a tell

---

PASS 9 — The Human Messiness Check

The most damning tells are absences. After all edits, check: does the writing have any of these?

• Sentence fragments used for emphasis
• Inconsistent stylistic choices (intentional ones, not errors)
• Personal linguistic tics or recurring phrasing
• Emotional overflow affecting syntax
• At least one moment of digression or natural tangent
• At least one paragraph that breaks the structural pattern
• A point that's made and then not countered with a caveat

If the answer to all of the above is "no" — the writing is still AI.

---

Usage Modes

Mode A: Audit Only

Read the text and produce a flagged report by section. List every pattern found with the offending phrase quoted. Do not rewrite yet.

Trigger phrase: "audit this for AI tells"

Mode B: Full Edit

Run all 9 passes and produce a clean rewrite. Preserve the original argument and key facts. Do not introduce new tells while fixing old ones.

Trigger phrase: "dehumidify this" / "de-robot this" / "humanize this"

Mode C: Single Pass

Target one specific category (e.g., "just fix the vocabulary" or "just fix the structure"). Run only the relevant pass.

Trigger phrase: "dehumidify [specific issue] only"

Mode D: Score Report

Produce a score (0–100, where 0 = indistinguishable from AI, 100 = fully human) with a breakdown by category and a short diagnosis.

Trigger phrase: "score this for AI tells"

---

Quick Reference: Worst Offenders

These patterns are the highest-signal AI tells. Flag these first in any audit:

1. Em dash pseudo-profundity — "not X — but Y"
2. Fortune cookie endings
3. Temporal openers — "In today's rapidly evolving world"
4. Rule of three triplets
5. Front-loaded transition chains — Moreover / Furthermore / Additionally in sequence
6. "It is important to note that..."
7. Modifier stacking — three adjectives per noun
8. Fake casual openers replacing formal ones
9. Uniform paragraph lengths throughout
10. Exhaustive listing / coverage anxiety

---

Anti-Patterns: What NOT to Do When Editing

• Don't replace "Moreover" with "Picture this:" — template swap, not a fix
• Don't add profanity to signal authenticity — if the rest of the voice doesn't match, it reads worse
• Don't fragment every third sentence to seem punchy — pattern repetition is still a pattern
• Don't cut all hedges and overcorrect to false certainty
• Don't introduce new metaphors to replace old ones if they don't come from the subject matter naturally

---

Last updated from: Comprehensive Analysis of AI-Generated Writing Tells (source document)