The 42 Tropes
Every AI writing pattern we scan for. Organized by severity, starting with the patterns that make readers stop reading and ending with the ones only editors catch. No single pattern proves anything. The score reflects concentration, and some flags will be coincidental.
Dead Giveaway
Tier 1 (6 patterns)Patterns that generate the strongest reader reactions. Any one of these appearing prominently can break trust on contact.
"It's not X, it's Y"
The single most commonly identified AI writing tell. A stray instance is forgivable, but ten in a blog post insults the reader. AI models love this rhetorical move because it sounds profound while requiring zero actual insight.
Em Dash Addiction
This text treats em dashes like oxygen. GPT-4o uses roughly 10x more em dashes than GPT-3.5, and readers now pattern-match them as a robot tell. A couple per page blend in, but a couple per paragraph announce the machine.
Vocab Hall of Shame
The word "delve" spiked 900% in academic papers after ChatGPT launched. A Chapman professor went viral saying if it shows up in your paper, there's a 99% chance AI wrote it. These words are fine in small doses. Their sudden ubiquity is the giveaway.
Leftover Artifacts
Oops. Someone forgot to proofread the AI output. "As a large language model" is the classic, but you also see placeholder brackets, table stubs, and half-finished instructions the model was supposed to fill in. This is the AI equivalent of leaving the price tag on your shirt.
Fabricated Citations
The rhetorical equivalent of a fake Rolex. "Studies show" without naming the study. "Experts say" without naming the expert. "Research suggests" without linking the research. AI loves these because they sound authoritative while committing to nothing.
Sycophantic Opener
Nothing screams "a chatbot wrote this" like opening with lavish praise of the question itself. "Great question!" "That's a really insightful observation!" Real humans just answer the question. They don't warm up with a compliment first.
Red Flag
Tier 2 (9 patterns)Patterns that reliably cause readers to disengage. Editors and hiring managers both flag these.
Fast-Paced World
The laziest possible opening for any article about technology, business, or really anything written after 1995. "In today's rapidly evolving landscape" is the AI equivalent of a throat-clearing noise. It says nothing and everyone skips it.
Ornate Metaphors
AI models reach for the same decorative metaphors like a college freshman reaching for a thesaurus. "Tapestry of innovation." "Beacon of progress." "Fabric of society." These are wallpaper, worn so smooth they carry zero meaning and surprise nobody.
Rhetorical Self-Answer
Ask a question, then immediately answer it. "The result? Devastating." "So what does this mean? It means everything." This is the literary equivalent of high-fiving yourself. Once per piece can work, but AI does it every other paragraph.
"Not Only... But Also"
AI adores parallel constructions. "Not only did they improve efficiency, but they also boosted morale." The balanced, polished feel is exactly the problem, because real writing has rough edges and this construction sands them all off.
False Suspense
Creating drama where none exists. "Here's the thing." "Here's where it gets interesting." "Let that sink in." If you have to tell the reader something is interesting, it probably isn't. This is the writing equivalent of a drumroll before opening a bag of chips.
Formulaic Conclusion
The "In conclusion" of it all. AI wraps up pieces like a fifth-grader ending an essay. "In summary." "Overall." "To wrap things up." Good writing just stops when it's done, without announcing the exit.
Excessive Hedging
AI hedges like a lawyer drafting a liability waiver. "It could be argued." "One might suggest." "This may potentially." It's trained to avoid being wrong, so it qualifies everything into meaninglessness. A paragraph with four "mays" is a paragraph that says nothing.
Verdict Language
Handing down a judgment from on high. "That's what leadership looks like." "And that changes everything." "This is what the future demands." AI loves to play the wise narrator delivering final verdicts. The reader didn't ask for a ruling.
Breathless Enthusiasm
Everything is "groundbreaking" and "transformative" and "remarkable." AI writes like a press release for a product launch that will change the world (it won't). Constant superlatives flatten the landscape until the reader stops believing any of them.
Worth Noting
Tier 3 (16 patterns)These are harmless alone, but when three or more cluster in a few paragraphs, they signal an unedited draft.
Formal Transitions
Your blog post isn't a dissertation. "Moreover" and "furthermore" and "additionally" belong in legal briefs and academic papers, not a LinkedIn post about team culture. AI defaults to formal transitions because it learned from formal text. Real humans say "also" or just start a new sentence.
AI Vocab Cluster
Any one of these words is fine in isolation, but cluster a few in the same paragraph and your AI detector starts beeping. They sit at the exact center of the model's vocabulary distribution, safe and polished enough to appear in any context without friction.
Listicle Bullets
AI turns everything into a numbered list with bold lead-ins even when prose would work better. "1. **Improved Efficiency**: By streamlining..." Genuine lists — steps in a process, items in an order, things that are actually enumerable — are fine. This flags the AI tell: generic benefit-summary lists where the content would read better as connected sentences.
Punchy Fragments
Short sentence. For emphasis. That lands. Except when every paragraph ends this way, it stops landing and starts grating. AI learned this trick from copywriting courses and now deploys it with the subtlety of a car alarm.
Hollow Signaling
"It's worth noting" that if something were actually worth noting, you wouldn't need to announce it. Same with "interestingly" (let the reader decide if it's interesting) and "importantly" (ditto). These phrases are filler dressed up as emphasis.
Stakes Inflation
AI treats a SaaS feature update like the moon landing. "This fundamentally reshapes how we think about project management." It added a Gantt chart. When every feature is a watershed moment, the reader tunes out all of them.
Participial Overuse
Trailing "-ing" phrases piled onto sentences like luggage on a roof rack. "The team shipped the feature, improving performance, reducing costs, and enabling new workflows." One participial phrase is fine. Three in a row is AI stacking clauses because it can't commit to a period.
"From X to Y"
"From beginners to experts." "Everything from design to deployment." "From ideation to execution." AI loves this construction because it implies comprehensiveness without any specifics. The range gesture replaces actual detail every time.
Triplet Framing
Everything comes in threes. "Speed, scale, and simplicity." "Plan, build, and ship." "Clear, concise, and compelling." The rule of three is a real rhetorical device, but AI applies it on autopilot. When every sentence is a triad, the pattern becomes the message.
Anaphora Abuse
"They built the platform. They scaled the team. They changed the industry." Repetition at the start of consecutive sentences is a powerful rhetorical device when used deliberately (ask MLK). AI uses it like a toddler who just learned a new word.
Equivocation Seesaw
Make a claim, then immediately soften it. "This is a major breakthrough, though challenges remain." "The results are promising, but more research is needed." AI is allergic to taking a position. Every assertion gets an instant counterweight, leaving the reader exactly where they started.
Latinate Vocabulary
"Utilize" instead of "use." "Facilitate" instead of "help." "Commence" instead of "start." AI defaults to the fancier word because its training data is full of academic papers and corporate memos. Clear writing picks the shorter word almost every time.
Dramatic Countdown
"Not complex. Not costly. Just effective." Three short negations (or affirmations) building to a mic drop. AI borrowed this from ad copy and uses it constantly. It works in a Nike commercial because you hear it once. In AI text, it shows up every 200 words.
Colon Preface
"Here's the takeaway: start with the member." "The lesson is clear: invest early." AI loves to build a little runway before the actual point, using a colon as the launchpad. Once or twice is fine. But when every insight needs a setup clause, the text reads like a series of fortune cookies.
Elegant Variation
AI avoids repeating a word by cycling through synonyms for the same thing. A company becomes "the platform," then "this tool," then "the solution," then "the offering," all in the same paragraph. Each synonym tries to sound fresh, but the effect is the opposite: the reader notices the writer is dancing around repetition instead of owning it.
"Despite Challenges" Pivot
A rigid three-part formula: acknowledge something positive, list challenges or concerns, then pivot to vague optimism about the future. "Despite these challenges, the potential remains significant." AI uses this structure to handle any topic that involves tradeoffs, and it always lands in the same noncommittal place.
Subtle Tell
Tier 4 (8 patterns)Casual readers miss these, but editors and publishers who evaluate writing professionally catch them every time.
Consensus Middle
Every word is the safest possible pick from the center of the vocabulary distribution. The adjectives are predictable and the verb choices generic. The whole thing reads like a committee wrote it, because statistically, a committee did (a committee of billions of training tokens all averaged together).
Uniform Sentence Length
Human sentences range from 3 words to 40. AI sentences cluster between 15 and 25 like they're observing a speed limit, and the variance is unnervingly low. Real writers alternate between terse bursts and winding, clause-heavy builds. AI just cruises at one speed.
Missing Specifics
"Various sectors" instead of naming three companies. "Numerous studies" without citing one. "Significant impact" without a number. AI is allergic to specifics because specifics can be wrong, and being wrong is the one thing it's trained to avoid. The result is prose that gestures vaguely at reality.
Treadmill Effect
The text moves its legs but goes nowhere. The same point gets restated in slightly different words, paragraph after paragraph. "Innovation drives growth. By innovating, organizations can grow. Growth-oriented organizations prioritize innovation." You just read three sentences that said one thing.
Third-Person Detachment
"One might consider the implications." "It could be observed that." AI writes like an anthropologist studying humans from behind a two-way mirror. Real people say "I think" or "you should try." The clinical distance is a safety mechanism, and readers feel it.
"Serves As" Dodge
"Is" becomes "serves as." "Works as" becomes "stands as." "Shows" becomes "represents." AI avoids direct statements the way a diplomat avoids direct answers. "The dashboard serves as a window into performance" just means "the dashboard shows performance." Say that instead.
Importance Adverbs
"Quietly orchestrating." "Deeply transformative." "Remarkably consistent." AI sprinkles these adverbs like seasoning to make bland sentences taste important. If you have to tell the reader something is deep or remarkable, the sentence itself probably isn't doing the work.
Uniform Tone
The politeness level and formality stay constant from first word to last, the emotional register perfectly flat. The writer never gets angry or funny. Human writing has texture because human moods have texture, and AI flattens all of it out.
Deep Cut
Tier 5 (5 patterns)Statistical and structural patterns, detectable through analysis rather than casual reading.
Low Perplexity
Every word in this text is the statistically safest choice given the words before it. That's literally how language models work, picking the highest-probability next token. The result reads smoothly but predictably, like a song where you can always guess the next note, and it never takes the weird left turn that makes human writing feel alive.
Low Burstiness
Burstiness measures how much sentence length varies. Humans are bursty. We write a 6-word zinger, then a 40-word monster. AI flattens this curve into a gentle hum. The standard deviation of sentence lengths in AI text is measurably, consistently lower than in human text.
Perfect Grammar
Zero typos, zero fragments, zero sentences starting with "And" or "But," and not a single comma splice or stylistic risk anywhere. This sounds like a compliment until you realize that real writers break rules on purpose, constantly. Perfect grammar in a casual blog post is like wearing a tuxedo to a barbecue.
Style Consistency
The comma habits hold steady from paragraph one to paragraph fifty, the capitalization choices never waver, and the formatting conventions show zero drift. Humans are inconsistent creatures who sometimes use an Oxford comma and sometimes don't, even in the same document. AI's unwavering perfection is itself the tell.
Hollow Sensory Details
"The aroma of fresh coffee filled the room." "Sunlight streamed through the window." "A gentle breeze carried the scent of pine." AI sprinkles generic sensory details like stock photos in a PowerPoint. They're technically descriptive but weirdly impersonal, because the model has never smelled coffee or felt a breeze.