Files
harness-tools/skills/document/unslop/references/taboo-phrases.md
T
2026-08-24 16:33:11 -04:00

49 KiB

Taboo Phrases Reference

AI-generated text follows predictable patterns. This reference catalogs common tells to eliminate, organized by category.

Quoted examples are exempt by default. When you're documenting bad AI writing, sentences inside quotes, markdown blockquotes, or code fences are illustrative examples rather than prose to rewrite.

Table of Contents

  1. Throat-Clearing Openers
  2. Emphasis Crutches
  3. Business Jargon
  4. Filler Adverbs & Phrases
  5. Meta-Commentary
  6. Performative Emphasis
  7. Overused Intensifiers
  8. Absolute Words
  9. Structural Patterns to Avoid
  10. Significance & Legacy Inflation
  11. Promotional Language
  12. Superficial -ing Analyses
  13. Vague Attributions
  14. Copula Avoidance
  15. Formulaic Challenges Sections
  16. Negative Parallelisms
  17. Communication Artifacts
  18. Reader-Steering Frames & Vague Endorsements
  19. Reasoning-Chain Artifacts
  20. Acknowledgment Loops
  21. Novelty Inflation
  22. False Concession Structures
  23. Rhetorical Question Openers
  24. Parenthetical Hedging
  25. Numbered List Inflation
  26. Knowledge-Cutoff Disclaimers
  27. False Agency
  28. Conclusion & Sequencing Scaffolding
  29. Generic Positive Conclusions
  30. False Ranges
  31. Elegant Variation / Synonym Cycling
  32. AI Vocabulary (Additional)
  33. Punctuation Abuse
  34. Research-Sourced Additions
  35. Macro Structure
  36. Quick Reference: Detection Regex

Throat-Clearing Openers

These phrases delay getting to the point. Cut them entirely.

Phrase Why It's Bad
"Here's the thing:" Delays content
"The uncomfortable truth is" Dramatic, unnecessary
"It turns out" Hedging
"The real [X] is" Implies prior deception
"Let me be clear" Defensive positioning
"The truth is," Implies prior lying
"I'll say it again:" Redundant
"I'm going to be honest" Implies prior dishonesty
"Can we talk about" Permission-seeking
"Here's what I find interesting" Self-referential
"Here's the problem though" Delays content
"Let's be real" Performative
"Here's the deal:" Informal padding
"Here's what nobody tells you:" Clickbait-style
"It's no secret that" Performative
"Here's why" Delays content
"Let's dive in" Workshop filler
"Let's unpack" Workshop filler
"Let's explore" Workshop filler
"Let's examine" Workshop filler
"Let's break this down" Workshop filler
"Let's take a look" Workshop filler
"This is where it gets interesting" Performative delay

Emphasis Crutches

These phrases try to manufacture importance. The content should speak for itself.

Phrase Why It's Bad
"Full stop." Performative finality
"Period." Same as above
"Let that sink in." Condescending
"This matters because" Tells instead of shows
"Make no mistake" Dramatic positioning
"Here's why that matters" Over-explaining
"Read that again." Condescending
"This is important." Shows insecurity in content
"This cannot be overstated." Overstating the overstating
"Why this is load-bearing" Pre-announces importance instead of showing it
"Why this matters" Same — just make the point
"The struggle is real." Meme-derived punch line; asserts stakes instead of showing them
"The stakes are real." Same — name the concrete consequence
"The X is real." " is/are real" inflates significance; spare the literal "is it genuine?" sense

Business Jargon

Replace with direct alternatives.

Avoid Use Instead
Navigate (challenges) Handle, address, manage
Unpack (analysis) Explain, examine, explore
Lean into Accept, embrace, commit to
Landscape (context) Situation, field, market
Game-changer Significant, important, major
Game changer Significant, important, major
Navigating the complexities Handling the specifics
Delve into Examine, explain
Delve deeper Examine further
Double down Commit, increase, reinforce
Deep dive Analysis, examination, study
Take a step back Reconsider, pause, review
Moving forward Next, from now, going ahead
Circle back Return to, revisit, follow up
On the same page Aligned, agreed, in sync
Level up Improve, advance, grow
Bandwidth Capacity, time, availability
Synergy Cooperation, collaboration
Low-hanging fruit Easy wins, quick fixes
Pivot Change, shift, redirect
Disrupt Change, challenge, transform
Leverage Use, apply, utilize
Scalable Expandable, growable
Actionable Practical, usable, concrete
Ecosystem Environment, system, network
Stakeholder People involved, interested parties
Touch base Talk, connect, check in
Value-add Benefit, contribution
Thought leader Expert, specialist
Best-in-class Leading, top-tier
Cutting-edge Modern, advanced, new
Crucial Important, key (or cut)
Delve Explore, examine, look at
Enhance Improve, strengthen
Garner Get, earn, attract
Pivotal Important, key
Showcase Show, demonstrate
Testament Proof, evidence, sign
Underscore Show, highlight, prove
Fostering Building, encouraging, creating
Align with Match, fit, support
Robust Strong, solid, thorough
Comprehensive Full, complete, thorough
Utilize Use
Facilitate Help, enable, run
Spearhead Lead, start, run
Bolster Support, strengthen
Streamline Simplify, speed up
Harness Use, tap, apply
Multifaceted Complex, varied

Filler Adverbs & Phrases

These add words without adding meaning.

Phrase Action
"At its core" Cut entirely
"In today's [X]" Cut entirely
"It's worth noting" Just note it
"Interestingly," Cut—if interesting, it shows
"Importantly," Cut—importance should be evident
"Crucially," Cut
"At the end of the day" Cut
"When it comes to" Replace with "For" or cut
"In a world where" Cut entirely
"The reality is" Cut
"In terms of" Cut or use "for"
"With that said" Cut
"That being said" Cut
"All things considered" Cut
"By and large" Cut
"For the most part" Cut
"To be fair" Cut
"To be honest" Cut—implies prior dishonesty
"Needless to say" Then don't say it
"It goes without saying" Then don't say it
"The bottom line" Cut and state the point
"The key takeaway" Cut and state the point
"It's clear that" Cut, just state it
"What's clear is" Cut, just state it
"It's worth mentioning" Just mention it
"In an era of" Cut entirely
"A closer look" Cut entirely
"In order to" Replace with "To"
"Due to the fact that" Replace with "Because"
"Is able to" Replace with "Can"
"Has the ability to" Replace with "Can"
"At this point in time" Replace with "Now"
"In the event that" Replace with "If"
"Prior to" Replace with "Before"
"Subsequent to" Replace with "After"
"A large number of" Replace with "Many"
"The vast majority of" Replace with "Most"
"In terms of" Cut or use "for"
"It is important to note that" Cut
"It is important to note" Cut
"It's important to note" Cut
"It is worth noting" Cut
"The fact that" Cut and restructure

Meta-Commentary

These break the fourth wall unnecessarily.

Phrase Why It's Bad
"Hint:" Patronizing
"Plot twist:" Performative
"Spoiler:" Unnecessary framing
"You already know this, but" Then why say it?
"But that's another post" Content debt
"X is a feature, not a bug" Overused
"Dressed up as" Cliché
"Let me explain" Just explain
"To put it simply" Just put it simply
"In other words" Say it right the first time
"If you think about it" Patronizing
"As I mentioned" Redundant reference
"Spoiler alert" Performative
"Pro tip" Performative authority
"Hot take" Performative casualness
"Unpopular opinion" Usually neither unpopular nor an opinion

Performative Emphasis

These try too hard to sound genuine.

Phrase Why It's Bad
"creeps in" Dramatized
"I promise" Over-earnest
"They exist, I promise" Defensive
"This is genuinely hard" "Genuinely" undermines itself
"This is what X actually looks like" Implies deception elsewhere
"Trust me" Undermines trust
"Believe me" Same
"the million-dollar question" Manufactured drama; just ask the question
"Honestly" Implies prior dishonesty
"The elephant in the room" Over-dramatic framing
"It begs the question" Usually misused, always performative
"It's a no-brainer" Performative certainty
"Buckle up" Performative hype
"Food for thought" Performative
"Truly" Overemphasis
"Literally" Usually used incorrectly

Overused Intensifiers

These words have lost meaning through overuse.

Word Problem
Deeply Almost always deletable
Truly Rarely adds meaning
Fundamentally Often hyperbolic
Inherently Usually wrong
Simply Often untrue
Literally Overused
Inevitably Deterministic assumption
Essentially Vague
Incredibly Inflation
Absolutely Often unnecessary
Extremely Inflation
Really Weak intensifier
Very Almost always deletable

Absolute Words

These make claims that are usually false.

Word Problem
Always Rarely true
Never Rarely true
Everyone Rarely true
Nobody Rarely true
Everything Hyperbolic
Nothing Hyperbolic
All Often overstated
None Often overstated
Completely Usually exaggerated
Totally Usually exaggerated
Entirely Usually exaggerated
Perfectly Usually exaggerated

Structural Patterns to Avoid

Binary Contrasts (False Drama)

Pattern Fix
"Not because X. Because Y." State Y directly
"[X] isn't the problem. [Y] is." "The problem is Y"
"It feels like X. It's actually Y." Just state Y
"Not X. But Y." "Y matters here"
"X isn't Y. It's Z." State Z directly; drop the contrast scaffolding

Dramatic Fragmentation

Pattern Fix
"[Noun]. That's it. That's the [thing]." Complete sentences
"X. And Y. And Z." Normal flow
"This unlocks something. [Word]." Trust content
"The ___ loop." (as standalone fragment) Rewrite as a complete sentence or cut
"X things. One thing." State the single point directly

Rhetorical Setups

Pattern Fix
"What if [reframe]?" Make the point directly
"Here's what I mean:" Cut—redundant
"Think about it:" Cut—condescending
"And that's okay." Cut—unnecessary permission

Rhythm Problems

Pattern Fix
Three-item lists Use two or one
Questions answered immediately Let questions breathe or cut
Paragraphs starting with "So" Start with content
Every paragraph ends punchily Vary endings
Em-dashes before reveals Use periods or commas
Staccato fragmentation Don't stack short punchy sentences

Contrastive Definitions (False Drama)

The "affirm then negate" template asserts an identity by rejecting a foil. It reads as insight but only manufactures contrast. Soft: a genuine factual correction can look the same, so imperative corrections ("Use pnpm, not npm."), numeric corrections ("Latency fell 40%, not 4%."), and the authentication idiom ("The painting is real, not a forgery.") are exempted.

Pattern Fix
"X is a build step, not an output." State the positive claim on its own
"Failures lived in the shared substrate, not the models." State where they lived; drop the foil
"X isn't a Y, it's a Z." State what it is directly

Slogan & Template Cadence

Marketing copy from strong models leans on a few headline skeletons. Any one can be fine; the tell is the shape (and, for the fragment and the headline cadence, its repetition). Soft severity.

Pattern Fix
Two-beat imperative: "Emit 1,100 tokens. Ship 237KB." Combine into one sentence
Repeated "<plural noun> that <verb>." (2+ in a doc) Vary the structure or use full sentences
"One X, N Y." (numeric parallelism) Rewrite as a normal sentence
"The feedback loop ships inside the artifact." (abstraction ships inside/with) State the mechanism plainly
Headline slogan cadence: "Short statement. Short statement." on 3+ headlines in one doc Vary the headline shapes

The headline-cadence family generalizes the two-beat imperative slogan to ANY standalone line built wholly from two-to-four short sentences (each ≤5 words, subject-verb, imperative, or a bare noun phrase): "One command. A real URL." / "Reviewers click. The agent fixes." A single such headline is voice, so it fires only at document frequency ≥3. It is anchored to a full line — a line that opens with any long sentence, or a short-sentence run flowing inside a paragraph, is spared. Honest limit: in plain text "standalone headline" can only be approximated as "a line whose whole content is the short-sentence pair," so the tell is gated on frequency, not on proof that the line is a real heading.

Headings and Slide Titles

Write the title after the slide or section has earned its point. A strong title names the test, mechanism, result, decision, or completion criterion the viewer should inspect.

Prefer:

  • "The Week 1 capstone is a clean run with recorded agreement."
  • "Order, verbosity, and self-preference bias the judge."
  • "Each capstone must clear three gates."
  • "Can one packet reconstruct both products?"

For a bad or conflicting evaluation, name the discrepancy or its human/system cause. Do not preserve tool agency by changing "the judge lies" into "the judge gives a false verdict"; for example, use "One response receives conflicting labels" when that is the supported claim.

Rewrite:

  • Calendar or program containers delivering outcomes: "Week 1 ends with a calibrated eval"; "Demo day closes the cohort."
  • Tools, scores, data, or packages given narrative or volitional agency: "Your judge is lying"; "A score hides five failures"; "The package outlives the cohort."
  • Abstract uplift transformations: "Taste becomes infrastructure"; "Unlocking the power of clinical intelligence."
  • Generic handles that make the body do all the work: "Architecture", "Proof plan", "Three pillars", "What the budget buys."
  • Repeated counting or journey frames that describe the deck instead of the claim.

This is a semantic rule, not a blanket subject-verb ban. "Examples calibrate the judge" and "The gate fails the build" describe mechanisms. "Week 1 ends on Friday" describes a literal boundary. The problem is category-error agency or a title that withholds the actual point.


Significance & Legacy Inflation

These phrases manufacture gravitas. If something is truly significant, the facts show it.

Phrase Why It's Bad
"stands as" Inflated copula — just use "is"
"stands as a testament to" Double inflation
"testament to" Let evidence speak for itself
"pivotal moment" Almost never actually pivotal
"enduring legacy" Manufactured gravitas; state what actually persisted
"underscore the importance" Manufactured importance; state the evidence
"underscores the importance" Manufactured importance; state the evidence
"underscoring the importance" Manufactured importance; state the evidence
"highlights the importance" Manufactured importance; state the evidence
"indelible mark" Cliché gravitas
"rich history" Vague filler
"rich tapestry" AI favorite — never use
"tapestry of" AI metaphor cliché
"rich mosaic" AI metaphor cliché
"mosaic of" AI metaphor cliché
"a legacy of" Unearned weight
"iconic" Overused to meaninglessness
"seminal" Usually hyperbolic
"trailblazing" Promotional, not analytical
"groundbreaking" Usually not
"transformative" Rarely earned
"cornerstone of" Inflated significance
"speaks volumes" Cliche inflation
"speaks for itself" False agency; state what proves the point
"speaks for themselves" False agency; state what proves the point
"sends a clear message" Cliche inflation
"sheds light on" AI-formal for explains or shows
"raises the bar" Cliche inflation
"the fabric of" Vague inflation
"a testament to" Let evidence speak for itself
"a deep understanding" Inflated substitute for understanding
"holds promise" Empty significance claim without specifics
"holds great promise" Empty significance claim without specifics

Promotional Language

These phrases read like tourism brochures or press releases. AI defaults to promotional framing.

Phrase Why It's Bad
"nestled" Tourism cliché
"boasts a" Promotional voice
"breathtaking" Unearned superlative
"must-visit" Promotional, not descriptive
"in the heart of" Geographic cliché
"world-class" Empty superlative
"state-of-the-art" Empty superlative
"second to none" Empty superlative
"a hidden gem" Tourism cliché
"vibrant community" Vague promotional (soft — context-dependent)
"bustling" Tourism filler
"picturesque" Lazy description
"renowned for" Promotional attribution
"a beacon of" Inflated metaphor
"at the forefront of" Promotional positioning

Superficial -ing Analyses

Participial clauses tacked onto the end of sentences as fake analysis.

Pattern Why It's Bad
", highlighting..." Pretends to analyze, just restates
", showcasing..." Same
", underscoring..." Same
", fostering..." Same
", demonstrating..." Same
", reflecting..." Same
", signaling..." Same
", paving the way for..." Cliché + fake analysis

Fix: Delete the participial clause. If the analysis matters, give it its own sentence with actual reasoning.


Vague Attributions

These attribute claims to unnamed sources, creating an illusion of authority.

Phrase Why It's Bad
"Experts argue" Which experts? Name them or cut.
"Industry reports suggest" Which reports? Cite or cut.
"Some critics" Who? Be specific or cut.
"Many believe" Weasel phrasing
"Research suggests" Which research?
"Research indicates / Research shows" (bare, clause-initial) Unattributed authority — cite the study or name the source. Attributed ("Research by the Kaiser group indicates…") and possessive ("Our research indicates…") forms stay clean.
"Studies show" Which studies?
"It is widely regarded" By whom?
"According to experts" Name the experts
"Observers note" Which observers?
"Analysts predict" Which analysts?

Copula Avoidance

AI avoids "is/are" by substituting fancier verbs. The result reads as pretentious.

Avoid Use Instead
"serves as a" "is a"
"stands as a" "is a"
"represents a" "is a"
"constitutes a" "is a"
"functions as a" "is a"
"operates as a" "is a"
"acts as a" "is a"
"features a" "has a" or restructure
"remains a" "is still a" or just "is a"

Formulaic Challenges Sections

AI-generated articles follow a predictable structure: history → significance → challenges → future outlook.

Pattern Why It's Bad
"Despite its [achievements], [X] faces challenges..." Formulaic concession structure
"Challenges and Controversies" (as a section header) AI structural tell
"Future Outlook" (as a section header) AI structural tell
"Looking Ahead" (as a section header) AI structural tell
"However, it is not without its challenges" Mechanical concession
"While [positive], [negative] remains a concern" Template balanced structure

Negative Parallelisms

"Not only...but also" and similar constructions that signal AI generation.

Pattern Fix
"Not only...but also..." State both points directly
"It's not just about X, it's about Y" State what it's about
"Not merely X, but Y" State Y
"Beyond X, it also Y" State the second point directly
"More than just X" State what it actually is
"No X, no Y, just Z" Stacked negation for false drama — state Z
"Not X. Rather, Y." Copula-avoidance + binary contrast — just state Y
"It's not just X, it's Y" Same as "not only…but also" — state both directly

Communication Artifacts

Phrases left over from chatbot interaction that don't belong in published text.

Phrase Why It's Bad
"I hope this helps" Chatbot sign-off
"as an AI assistant" Pasted chatbot boilerplate
"Certainly!" Chatbot compliance marker
"Great question!" Chatbot flattery
"That's a great point" Chatbot agreement
"Absolutely!" Chatbot enthusiasm
"Of course!" Chatbot compliance
"Happy to help" Chatbot sign-off
"Let me know if you need anything else" Chatbot sign-off
"I'd be happy to" Chatbot compliance
"Sure thing!" Chatbot casualness
"I am happy to address" Chatbot deference (offers to receive criticism)
"I am open to any suggestions" Chatbot deference
"I am open to feedback" Chatbot deference
"I would appreciate any feedback" Chatbot deference
"I am willing to address" Chatbot deference
"I assure you" Chatbot reassurance
"demonstrate my commitment" Performative compliance
"with the utmost care" Performative respect
"If you have any concerns or suggestions" Chatbot solicitation

Reader-Steering Frames & Vague Endorsements

These cues tell the reader how to feel before earning it.

Phrase Why It's Bad
"Here's what's interesting" Pre-interprets importance
"Here's what caught my eye" Reader-steering frame
"Here's what stood out" Same
"worth reading" Empty thumbs-up without reason
"worth paying attention to" Say why it matters
"worth a look" Vague endorsement
"worth exploring" Generic approval instead of substance
"worth checking out" Same
"worth your time" Says nothing about the value

Fix: State the actual point or the specific reason something matters.


Reasoning-Chain Artifacts

These leak internal scaffolding instead of delivering the point.

Phrase Why It's Bad
"Let me think step by step" Internal reasoning leak
"Breaking this down" Filler setup
"To approach this systematically" Procedural throat-clearing
"Here's my thought process" The reader wants the conclusion
"Working through this logically" Internal monologue, not prose

Fix: State the conclusion first. Give evidence after.


Acknowledgment Loops

These restate the prompt before answering it.

Phrase Why It's Bad
"You're asking about" Restates the obvious
"To answer your question" Pure filler
"The question of whether" Indirect setup

Fix: Just answer.


Novelty Inflation

These frame routine ideas as secret discoveries or inventions.

Phrase Why It's Bad
"introduced a term" Often factually risky
"coined the phrase" Same unless sourced
"a concept nobody's naming" Fake scarcity
"a problem nobody talks about" Engagement bait
"the insight everyone's missing" Inflated exclusivity
"what nobody tells you about" Clickbait framing

Fix: Describe what the person explained or what the concept does. Don't claim novelty unless you can prove it.


False Concession Structures

These simulate balance without saying anything precise.

Pattern Why It's Bad
"While X is promising, Y remains a challenge" Template balance
"Although X has made strides, Y is still an open question" Same
"While X has limitations, Y is still remarkable" Empty concession

Fix: Name the actual tradeoff or pick a side and argue it.


Rhetorical Question Openers

These stall instead of making the point.

Phrase Why It's Bad
"What does this mean for..." Just tell the reader
"Why should you care?" Engagement bait
"What's next?" Often empty transition

Fix: Replace with the direct answer.


Parenthetical Hedging

Parenthetical asides often smuggle in weak nuance instead of committing.

Pattern Why It's Bad
"(and perhaps more importantly, ...)" Performs nuance
"(and, increasingly, ...)" Weak hedge
"(or, more precisely, ...)" Usually should be a sentence
"(arguably ...)" Hedged aside

Fix: If the aside matters, give it its own sentence. If it doesn't, cut it.


Numbered List Inflation

AI overuses counted lists because they are structurally safe.

Phrase Why It's Bad
"Three key takeaways" Number chosen for format, not meaning
"Five things to know" Same
"Top seven" Inflated list framing
"Here are 7 reasons..." Usually padding

Fix: Use a list only when the count matters. Otherwise write the point directly.


Knowledge-Cutoff Disclaimers

Phrases that betray AI-generated text by referencing training data limitations.

Phrase Why It's Bad
"as of my last" Direct AI tell
"based on available information" Hedging that reveals source
"as of my knowledge cutoff" Direct AI tell
"I don't have access to real-time" Direct AI tell
"based on my training data" Direct AI tell
"as of [date]" (when used as disclaimer) AI hedging pattern

False Agency

AI hands rhetorical agency to inanimate things so it can assert significance without making the claim itself. (Signal adapted from stop-slop.)

Phrase Why It's Bad
"the numbers speak for themselves" The writer should say what they show
"the data tells a (clear) story" Data doesn't narrate; state the finding
"tells a story" Data or artifacts do not narrate; state the finding
"the results speak for themselves" Dodges the actual interpretation
"X paints a clear picture" Inflated stand-in for "X shows"

Fix: State the figure and what it means. "Revenue doubled" beats "the numbers speak for themselves."

Tool Anthropomorphism

A close cousin of false agency: marketing copy hands an inanimate tool or artifact volitional or cognitive agency — machinery personified. Two shapes, both soft.

Pattern Why It's Bad
Reflexive self-agency: "The suite defends itself." / "The rules update themselves." / "It graded its own reflection." A tool acting on itself is personification; state what it does and to what
Volitional verb on a standalone headline line: "The bench decides which model does which job." / "It hunts instances, not word lists." A tool doesn't decide, hunt, want, or know; say what it does mechanically

The reflexive shape fires anywhere the tool-noun subject takes a reflexive object directly after the verb. The volitional shape (decides, hunts, wants, knows, believes, cares, refuses, judges, thinks) fires only when the whole line is the short sentence — the headline/slogan position. Ordinary technical register is spared: "the parser reads the file", "the model learns the distribution", "the gate fails the build", and the dev idiom "the test cleans up after itself" all stay clean, and human roles ("the judge decides the case") are not tool-nouns.

Fix: Name the mechanism. "The bench routes each job to a model" beats "The bench decides which model does which job."


Conclusion & Sequencing Scaffolding

Visible essay scaffolding. Ordinals and wrap-up openers announce structure instead of just delivering it.

Phrase Why It's Bad
"In conclusion" Don't announce the conclusion; state it
"In summary" / "To summarize" Filler wrapper
"Firstly / Secondly / Thirdly" Stiff scaffolding; use "first" or nothing
"Ultimately," (as an opener) Filler throat-clear before the point
"underscore(s) the importance of" Inflated stand-in for "shows"

Fix: Cut the scaffold word and lead with the content.


Generic Positive Conclusions

AI defaults to optimistic, empty endings.

Phrase Why It's Bad
"The future looks bright" Empty optimism
"What remains clear is" Generic conclusion scaffold
"Exciting times lie ahead" Content-free conclusion
"Only time will tell" Cliché non-conclusion
"One thing is certain" False certainty wrapper
"What remains clear is" Filler conclusion opener
"continues to shape" Vague ongoing-ness
"future prospects" Generic forecast label
"continues to evolve" Says nothing
"poised for growth" Promotional conclusion
"remains to be seen" Non-conclusion

False Ranges

"From X to Y" constructions that create a false sense of breadth.

Pattern Why It's Bad
"from X to Y, from A to B" Stacked ranges = AI tell
"ranging from X to Y" Often unnecessary — just list items
"spanning everything from X to Y" Inflated scope
"whether X or Y" (repeated) Parallel false range

Fix: Pick the most relevant items. If the range matters, state the endpoints with actual data.


Elegant Variation / Synonym Cycling

AI rotates synonyms to avoid repeating a word, producing unnatural prose.

Signal Example
Entity cycling "The company... the firm... the organization... the enterprise"
Verb rotation "said... stated... noted... remarked... observed"
Adjective swapping "significant... notable... remarkable... considerable"

Fix: Repeat the natural word. English prose allows and expects repetition of key terms. Forced variation is more distracting than repetition.


AI Vocabulary (Additional)

Single words that appear at disproportionately high rates in AI text.

Word Problem Severity
delve AI-favorite verb hard
garner Rarely used by humans in casual writing hard
interplay AI-formal hard
intricate Overused AI descriptor hard
tapestry AI metaphor cliché hard
underscore AI-formal for "show" hard
multifaceted AI-formal for "complex" hard
nuanced Overused AI qualifier soft
myriad AI-formal for "many" soft
plethora AI-formal for "many" soft
paramount AI-formal for "important" hard
resonate with AI-favorite for "matters to" hard
encompass AI-formal for "include" soft
pertaining to AI-formal for "about" hard
aforementioned AI-formal for "this" or "that" hard
henceforth AI-formal — use "from now on" hard
whereby AI-formal — restructure sentence hard
therein AI-formal hard
moreover AI transition word soft
furthermore AI transition word soft
nevertheless AI transition word soft
notwithstanding AI-legal — restructure hard
burgeoning AI-formal for "growing" hard
ubiquitous AI-formal for "everywhere" or "common" soft
resonates AI-favorite for "matters" or "connects" hard
intersection As in "at the intersection of" hard
double-edged sword AI cliche hard
sheds light AI-formal for "explains" or "shows" hard
strikes a balance AI-formal for "balances" hard
paints a picture AI metaphor cliche hard
meticulous AI-formal for "careful" or "thorough" hard
meticulously AI-formal adverb (companion to "meticulous") hard
bolster the argument/case/claim (any tense) AI-formal rhetorical support; literal reinforcement (a wall, a cushion) is fine hard (collocation-gated)
emphasizing Common AI participial padding soft
enduring AI-formal for "lasting" — significance inflation adjacent soft
vibrant Tourism/promotional default — frequent AI tell soft
valuable Empty positive descriptor when used without object soft
substrate Soft when used metaphorically for "foundation," "base," or "layer"; exempt literal lab, construction, and electronics uses soft
load-bearing AI-favorite metaphor for "essential" or "critical"; exempt literal construction and mechanics hard
wedge AI-favorite metaphor for "angle," "lever," "differentiator," or "entry point"; exempt literal construction uses hard

Hard severity: Always an AI tell regardless of context. Soft severity: Context-dependent — may be appropriate in academic or formal writing.


Punctuation Abuse

AI text has distinctive punctuation habits. These are high-signal tells.

Em-Dash Overuse (HIGH PRIORITY)

Em-dashes are the single most reliable AI punctuation tell. AI uses them for dramatic pauses, parenthetical asides, and reveals where humans would use commas, periods, or parentheses.

Pattern Fix Severity
Em-dash before a reveal Use a period or comma hard
Multiple em-dashes in one paragraph Max zero. Prefer commas/periods. hard
Em-dash as dramatic pause Let sentence structure create rhythm hard
Em-dash for parenthetical aside Use commas or parentheses soft
En-dash used as em-dash Fix the punctuation soft

Rule: Default to zero em-dashes. If one is absolutely necessary for rhythm and a comma or period genuinely won't work, allow one per several paragraphs at most. Two or more in any paragraph is always wrong.

Colon Overuse

AI loves the "setup: reveal" structure. Colons before dramatic reveals are a strong tell.

Pattern Fix Severity
"The answer is: [reveal]" State the answer directly hard
"The key takeaway: [point]" Just make the point hard
"The bottom line: [conclusion]" Just state the conclusion hard
"Here's the thing: [point]" Cut the setup entirely hard
Multiple colons per paragraph Restructure into separate sentences soft

Exclamation Mark Overuse

AI-generated text, especially in emails and social content, over-uses exclamation marks to simulate enthusiasm.

Pattern Fix Severity
Multiple exclamation marks in one paragraph Max one per several paragraphs hard
Exclamation marks in professional prose Use periods. Let content convey energy. soft
"!" after every sentence in a list Remove all but possibly the last hard

Boldface Abuse

Pattern Fix
Bold on every key term Reserve bold for truly critical items, max 1-2 per section
Inline headers in body text Use actual headers or let prose flow

Title Case in Body Text

Pattern Fix
"Digital Transformation Journey" Lowercase unless proper noun
"Thought Leadership Strategy" These aren't titles, don't capitalize

Curly Quote / Smart Punctuation Artifacts

Watch for inconsistent quote styles, a sign of copy-paste from AI output into different editors.


Research-Sourced Additions

Patterns added from external sources. Most academic single-words are soft (legitimate but over-represented in LLM prose — flag when clustered).

Significance & legacy puffery (Wikipedia: Signs of AI writing)

Phrase Note
"plays a vital/crucial/significant role" The canonical AI academic frame
"leaves a lasting impact", "enduring/lasting legacy" Vacuous legacy claims
"watershed moment" Soft — fine for a genuinely sourced turning point
"in the realm of" Almost always replaceable with "in"
"serves/stands as a testament/reminder/symbol…" Inflated copula; "serves as a kitchen" is fine

Promotional drift (Wikipedia)

Phrase Note
"rich cultural heritage" Travel-brochure cliché
"stunning natural beauty" Soft — acceptable in overt travel writing
"picturesque" Soft — promotional default

Vague attribution / weasel wording (Wikipedia)

Phrase Note
"has been described as", "is widely regarded as" Soft — fine when the source is named
"some critics argue", "observers have noted" Name the critics/observers
"to the best of our knowledge", "a growing body of" Cite the specific evidence

Editorializing / framing (Wikipedia)

Phrase Note
"(no discussion) would be complete without" Editorial voice
"In this article, we will explore…" Reader-addressing; alien to reference voice
"as an AI language model", "as of my last knowledge update" Pasted chatbot artifacts

Academic excess vocabulary (Berens & Kobak, Science Advances 2024 — excess-word frequency study of 14M PubMed abstracts)

Words / phrases Note
elucidate, delineate, underpin, unveil, seamless, invaluable, noteworthy Soft — overused style verbs/adjectives
revolutionize, revolutionizing, "holds (great) promise", "new avenues", "shed new light" Hype/discussion-section closers
"warrants further investigation", "it should be noted that" Empty closers / filler
"in recent years", "with the advent of", "taken together," Formulaic openers/summarizers
"garnered significant/considerable attention" Manufactured-consensus framing
"a deeper/more nuanced understanding of" Inflated stand-in for "understanding"
"treasure trove", "ever-evolving", "ever-changing" Cliche AI vocabulary

Sources: Wikipedia Signs of AI writing; Berens & Kobak, "Delving into ChatGPT usage in academic writing," Science Advances (2024), llm-excess-vocab; hardikpandya/stop-slop.


Macro Structure

These are document-level tells. Some are deterministic and cross-referenced to scripts/structure_scan.py; others require agent judgment and are not scanner-enforced.

Tell Enforcement Genre caveat
Both-sidesism Agent judgment — not scanner-enforced. Watch for "on one hand/on the other hand" that refuses to choose when the piece owes the reader a conclusion. Balanced journalism, judicial analysis, and literature reviews may need genuine opposing views.
Templated redemption arc Agent judgment — not scanner-enforced. Watch for a too-clean fall, lesson, transformation, uplift sequence. Memoir, sermons, and some case studies may intentionally use an arc.
Preview/recap symmetry Cross-check conclusion_coda and summary_sandwich; agent judgment decides whether the ending adds anything. Abstracts, executive summaries, and TL;DRs can legitimately restate the main point.
Over-determination Agent judgment — not scanner-enforced. Watch for explicit theme-stating that explains what the reader should infer. Explainers, tutorials, and accessibility-minded docs may need explicitness.
Elegant variation pointer Agent judgment — not scanner-enforced. Watch for one referent cycling through many labels to avoid repetition. Literary prose and journalism may vary references for rhythm, as long as clarity survives.
Uniform emotional register Agent judgment — not scanner-enforced. Watch for every paragraph landing at the same polished confidence level. Formal reports and reference docs may intentionally keep emotional texture low.

Related deterministic metrics: sentence_burstiness, bold_colon_listicle, one_line_staccato, connective_paragraph_openers, signpost_density, opener_repetition, and participial_closer_share.

Silhouette (discourse-level)

scripts/silhouette_scan.py scores idea arrangement one level above the surface scanner. All five tells are scanner-enforced; the composite silhouette_penalty flags at >= 1.0 against a committed human reference.

Tell What fires
Scaffold openers Body paragraphs opening on a discourse-cue class instead of their own claim — scanner-enforced (scaffold_opener_share).
Recap loop Opening vocabulary that vanishes mid-document and returns at the end — scanner-enforced (callback_content, strongest single tell).
Rotating opener roles Several distinct cue-opener classes rotating like a template — scanner-enforced (role_entropy_bits).
Preview-then-fulfill Intro content words reappearing as body-paragraph heads — scanner-enforced (preview_fulfillment).
Heading preview Section headings restating the intro's outline — scanner-enforced (heading_preview, retained under --genre docs).

Anti-gaming: a cue-deletion attack collapses the scaffold and role metrics, so silhouette must be scored jointly with structure_scan.py (staccato / opener_repetition) as the lower fence.


Quick Reference: Detection Regex

# Throat-clearing (case-insensitive)
(?i)(here's the thing|the uncomfortable truth|it turns out|let me be clear|the truth is|let's be real)

# Emphasis crutches
(?i)(full stop\.|period\.|let that sink in|make no mistake|read that again)

# "<abstract noun> is/are real" emphasis closer (spare literal "is it genuine?")
(?i)\b(?:the |that |this )?(?:struggle|stakes|pain|threat|risk|danger|fear|hype|magic|hustle|grind|stress|pressure|burnout|tension|mechanic)\s+(?:is|are|was|were)\s+(?:very |so |all too )?real\b

# Binary contrast pattern
(?i)(not because .+\. because|isn't the problem\.|feels like .+\. it's actually|\b\w+\s+(?:isn'?t|aren'?t|is not|are not)\s+[^.\n]+?\.\s+it'?s\b)

# Fragmentation
\. (That's it\.|And that's|That's the)
(?i)(?:^|[.!?]\s+|\n)\s*the\s+[a-z][a-z\-']*(?:\s+[a-z][a-z\-']*){0,2}\s+loop\.
(?i)\b\w+\s+things\.\s+one\s+thing\b

# Filler adverbs
(?i)^(interestingly|importantly|crucially|fundamentally),

# Significance inflation
(?i)(stands as a testament|pivotal moment|indelible mark|rich tapestry|cornerstone of)

# Promotional language
(?i)(nestled|boasts a|breathtaking|must-visit|in the heart of|world-class|state-of-the-art|a hidden gem)

# Superficial -ing analyses
(?i),\s+(highlighting|showcasing|underscoring|fostering|demonstrating|reflecting|signaling|paving the way)

# Vague attributions
(?i)(experts argue|industry reports|some critics|many believe|it is widely regarded)

# Copula avoidance
(?i)(serves as a|stands as a|represents a|constitutes a|functions as a|operates as a)

# Communication artifacts
(?i)(i hope this helps|certainly!|great question|happy to help|let me know if you need)

# Reader-steering + vague endorsement
(?i)(here's what's interesting|here's what caught my eye|here's what stood out|worth reading|worth paying attention to|worth a look|worth exploring|worth checking out|worth your time)

# Reasoning-chain artifacts
(?i)(let me think step by step|breaking this down|to approach this systematically|here's my thought process|working through this logically)

# Acknowledgment loops
(?i)(you're asking about|to answer your question|the question of whether)

# Novelty inflation
(?i)(introduced a term|coined the phrase|a concept nobody's naming|a problem nobody talks about|the insight everyone's missing|what nobody tells you about)

# Knowledge-cutoff disclaimers
(?i)(as of my last|as of my knowledge cutoff|based on my training data|based on available information)

# Generic positive conclusions
(?i)(the future looks bright|exciting times lie ahead|only time will tell|one thing is certain|continues to evolve)

# Negative parallelisms
(?i)(not only .+? but also|it's not just about|not merely .+? but)

# Contrastive definitions (soft; exempts imperative/numeric corrections + authentication idiom)
(?im)(?:^|[.!?]\s+)(?![a-z]*\b(?:use|run|write|keep|drop|make|ship|add)\b)[a-z][^.!?\n]{10,}(?<!real)(?<!genuine),\s+not\s+(?:a |an |the )?[a-z][^.!?\n]{0,45}[.!?]
(?i)\b(?:isn'?t|aren'?t|wasn'?t|weren'?t)\s+(?:a |an |the )?[^,.!?\n]{2,45},\s+(?:it'?s|that'?s|they'?re)\s+[^.!?\n]{1,45}[.!?]

# Slogan & template cadence (soft; fragment_template fires only at 2+ occurrences)
(?im)(?:^|[.!?]\s+)\b(?:ship|emit|build|deploy|render|run|write|export|scan|track|save|load|push)\b(?:\s+[^\s.!?]+){1,3}[.!?]\s+\b(?:ship|emit|build|deploy|render|run|write|export|scan|track|save|load|push)\b(?:\s+[^\s.!?]+){1,3}[.!?]
(?im)(?:^|[.!?]\s+)[a-z][\w'-]*s\s+that\s+[^.!?\n]{1,40}[.!?]
(?im)(?:^|[.!?]\s+)one\s+[^,.!?\n]{1,30},\s+(?:two|three|four|five|\d+)\s+[^,.!?\n]{1,30}[.!?]
(?i)\b(?:loop|story|experience|magic|feedback|workflow)\s+ships?\s+(?:inside|within|with)\b|ships?\s+(?:inside|within)\s+(?:the|your|a|an)\b

# Standalone slogan/spec fragment lines & headers (soft; whole-line only — embedded counts stay clean)
# slogan_fragment: "N X, one Y." ("Four presets, one input.")
(?:^|\n)[ \t]*(?:#{1,6}[ \t]*|>[ \t]*|[-*+][ \t]+)?(?:\*\*)?(?:one|two|three|four|five|six|seven|eight|nine|ten|\d+)\s+[^,.!?\n]{1,40},\s+one\s+[^,.!?\n]{1,40}[.!?](?:\*\*)?[ \t]*(?=\n|$)
# spec_fragment: "N noun-phrase, past-participle ..." ("Eight criteria, scored 1 to 5.")
(?:^|\n)[ \t]*(?:#{1,6}[ \t]*|>[ \t]*|[-*+][ \t]+)?(?:\*\*)?(?:one|two|three|four|five|six|seven|eight|nine|ten|\d+)\s+[^,.!?\n]{1,40},\s+(?:[a-z]+ed|built|written|drawn|made|split|set|done|shown|given|taken)\b[^.!?\n]{0,40}[.!?](?:\*\*)?[ \t]*(?=\n|$)

# Headline slogan cadence (soft; DOCUMENT-LEVEL — fires only at 3+ line matches)
(?i)(?:^|\n)[ \t]*(?:[a-z0-9][\w'-]*(?:[ ,]+[a-z0-9][\w'-]*){0,4}[.!?][ \t]+){1,3}[a-z0-9][\w'-]*(?:[ ,]+[a-z0-9][\w'-]*){0,4}[.!?][ \t]*(?=\n|$)

# Tool anthropomorphism (soft)
# tier A reflexive self-agency, fires anywhere; adjacency + "...out" lookahead spare "cleans up after itself" / "sorts itself out"
(?i)\b(?:the|these|those|its)\s+(?:suite|scanner|gates?|tool|system|pipeline|bench|rules?|tests?|code|models?|loop|harness)\s+(?:\w+ly\s+)?\w+\s+(?:(?:itself|themselves)(?!\s+out\b)|its\s+own\s+\w+)
(?i)(?:^|[.!?]\s+|\n)it\s+(?:\w+ly\s+)?\w+\s+(?:(?:itself|themselves)(?!\s+out\b)|its\s+own\s+\w+)
# tier B strongly-volitional verb, standalone-line only (excludes reads/runs/checks/returns/learns)
(?i)(?:^|\n)[ \t]*(?:#{1,6}[ \t]*|>[ \t]*|[-*+][ \t]+)?(?:\*\*)?(?:(?:the|its|this)\s+(?:suite|scanner|gates?|tool|system|pipeline|bench|rules?|tests?|code|models?|loop|harness)\s+|it\s+)(?:decide|hunt|want|know|believe|care|refuse|judge|think)(?:s|es|d|ed)?\b[^.!?\n]{0,55}[.!?](?:\*\*)?[ \t]*(?=\n|$)

# False concession structures
(?i)((while|although) .+?, .+? (remains|is still) .+?(challenge|concern|open question|limitation|constraint))

# AI vocabulary
(?i)\b(delve|garner|interplay|intricate|tapestry|underscore|multifaceted|paramount|burgeoning|notwithstanding|henceforth|aforementioned|pertaining to|resonates?|sheds light|strikes a balance|paints a picture|double-edged sword|substrate|load-bearing|wedge)\b

# Importance pre-announcements
(?i)(why this is load-bearing|why this matters)

# False ranges
(?i)(from .+? to .+?, from .+? to|spanning everything from|ranging from .+? to)

# Rhetorical question openers
(?im)(^\s*(?:but\s+)?what does this mean for|^\s*so why should you care\??|what's next\?)

# Parenthetical hedging
(?i)(\((?:and,?\s+)?(?:perhaps|more importantly|increasingly|more precisely|arguably)\b[^)]*\))

# Numbered list inflation
(?im)(^(?:here are|these are|the top)\s+\d+\s+(?:reasons|things|takeaways|lessons|ways)\b|three key takeaways|five things to know|top seven)

# Em-dash overuse (HIGH PRIORITY)
\u2014              # Any em-dash usage (flag for review)
\u2014[^\u2014]*\u2014  # Multiple em-dashes in same paragraph

# Colon reveals
(?i)(the (?:answer|secret|key|trick|truth|reality|problem|solution|takeaway|lesson|difference|reason) (?:is|was|isn't|remains))\s*:

# Commonly missed tells
(?i)(it's no secret|speaks volumes|the elephant in the room|it begs the question|sends a clear message|at the intersection of|resonates with|here's why\b|let's dive in|let's unpack|the bottom line|the key takeaway|it's clear that|what's clear is|buckle up|food for thought|raises the bar)

# Quote exemptions
# By default, quoted spans, markdown blockquotes, and code fences are masked before scanning.

# Elegant variation (requires manual review - look for entity cycling)
# "The company... the firm... the organization... the enterprise"