AI writing patterns
Signs of AI writing
A tell is a habit of phrasing that language models fall into far more often than people do. Every tell here also appears in human writing; what the checker reports is how often, not who wrote it. Try a paragraph first, then pick where to start.
Try it on your own writing
The scan runs in memory on Numen's server and the passage is discarded with the response.
Where do you want to start?
52 tells
Sentence structure (10 tells)
- AnaphoraRepeated sentence openers“Nobody read it. Nobody fixed it. Nobody asked.” Three openers make one drumbeat.Moderate signal
- CataphoraCounted lead-ins“There are three key reasons” promises a list the paragraph still has to fill.Moderate signal
- Did not repetition“Did not X, did not Y” chains“Did not stall, did not leak, did not cost a day” carries one fact and three refusals.Moderate signal
- Final thoughtsWrap-up headings“Final Thoughts” tells you the piece is ending, never what the ending says.Moderate signal
- Mic dropMic-drop fragmentsOne line stands alone after the paragraph: “It matters.” Nothing above it earned that.Moderate signal
- No fluff no filler“No X, no Y” chains“No fluff, no filler” refuses two things the reader was never offered.Moderate signal
- On the x side“On the X side,” openers“On the funding side,” frames a whole clause before anyone says who did anything.Moderate signal
- Pseudo-cleft sentence“What matters is …” frames“What changed is the schedule” is a slower way to say “the schedule changed.”Moderate signal
- Rhetorical questionStacked rhetorical questions“Did it work? Compared to what?” The questions stop, and no answer ever arrives.Moderate signal
- Rule of threeEchoing sentence runsWhen each sentence ends on “made the shortlist this year”, the rule of three has stalled.Moderate signal
Rhetorical moves (17 tells)
- AntithesisStranded auxiliary contrast“The launch held; the budget didn’t.” The second half never names a verb.Moderate signal
- As we all knowBorrowed consensus“As we all know” hands you the agreement and keeps the evidence out of sight.Weak signal
- Don’t call it“Don’t VERB it … VERB it”“Don’t call it that. Call it this.” The verb returns and only the label changes.Strong signal
- Hypophora“The result? …” pivots“The result?” is a question that never waits for anyone else to answer it.Moderate signal
- Let’s be honestPerformative honesty“Let’s be honest” promises candor, then delivers a claim nobody doubted.Moderate signal
- MerismFalse ranges“Everything from onboarding to offboarding” names two stops, not two ends.Weak signal
- Negative parallelism“Not X. Not Y. Just Z.”“Not a scam. Not a bargain. Just an ordinary lamp.” The last line has to pay for both.Strong signal
- Not just but also“Not just X, but Y”“Not just a broken heater, but a shrug.” Conceding the heater costs the writer nothing.Weak signal
- Participial phraseParticiple sentence tailsCut everything after “, underscoring how” and see whether a claim goes with it.Moderate signal
- Paving the way forSignificance tails“Paving the way for what?” The tail claims a result nobody watched land.Moderate signal
- Serves as“Serves as” dodges“Serves as a catalyst” says no more than “is a catalyst,” and says it slower.Moderate signal
- Some might argueStrawman contrasts“Some might argue” hands the other side a line nobody on it ever said.Weak signal
- That’s where it lives“Where it actually lives”“That’s where the real work lives.” Nothing lives there, and no work got named.Strong signal
- Then something shiftedNarrative pivots“Then something shifted.” The sentence announces a turn and names nothing.Moderate signal
- Think of it asAnalogy openers“Think of it as a thermostat” hands you a comparison before the subject arrives.Moderate signal
- We call thisCoined concept labels“The onboarding trap” names a pattern the writer has seen exactly once.Moderate signal
- Weasel words“Experts argue”“Experts say” is a citation with the name taken out of it.Moderate signal
Stock phrases (25 tells)
- Cannot be overstatedUrgency inflation“Cannot be overstated” announces weight the sentence beside it never puts on the scale.Strong signal
- Despite the challenges“Despite these challenges”“Despite these challenges” names none of them, and the sentence moves on regardless.Strong signal
- “Don’t take my word for it” is a promise, and the next clause has to keep it.Strong signal
- Here’s the thing“Here’s the twist”“Here’s the thing:” promises a reveal, then hands over a line you already had.Strong signal
- In conclusionSignposted conclusions“In conclusion,” says the paragraph, before a single thing has been concluded.Weak signal
- In today’s fast-paced worldOpening clichés“In today’s fast-paced world” takes up a first sentence and says nothing.Moderate signal
- Is the entire game“Is the entire …”“Onboarding was the entire story.” The sentence never says what came second.Strong signal
- Is the whole game“Is the whole …”“Retention is the whole game.” The paragraph above it never argued that.Strong signal
- It is important to note“It’s important to note”“It’s important to note that” rates a claim before it has told you what the claim is.Moderate signal
- Let’s unpack that“Let’s unpack this”“Let’s unpack that” announces an explanation, then asks you to wait for it.Moderate signal
- Litotes“That’s not nothing”“That’s not nothing” is litotes that grants a result weight and never says how much.Strong signal
- Puffery“Fits in your head”“Fits in your head” says the hard part is over without naming what it was.Strong signal
- Shell noun“The problem is that …”“The catch is that” names a box; the sentence still has to fill it.Weak signal
- “Sit with that.” Three words stand in for a feeling the paragraph never named.Strong signal
- That’s the part“That’s the part …”“That’s the part.” The phrase promises a detail the sentence may not have.Strong signal
- That’s the whole point“That’s the whole …”“That’s the whole point.” Then the paragraph ends, and nothing arrives to prove it.Moderate signal
- That’s why it mattered“That’s why X mattered”“That’s why it mattered” closes a paragraph on a verdict the story never earned.Strong signal
- The entire point is“The entire … is”“The entire point is trust” reads like a finding until you check the line above it.Strong signal
- The only one I trust“The only X I trust”“The only one I trust” names a winner and never says who else was in the running.Strong signal
- “The punchline is” promises a reversal, then hands over a fact that needed no warning.Strong signal
- This is real“Is real … and / not”“The backlog is real, and it keeps growing” insists twice and measures nothing.Moderate signal
- Turns out“Turns out …”“Turns out” reports news, and nobody in the paragraph was ever in doubt.Moderate signal
- “It’s worth naming” promises a name, and the sentence ends before one arrives.Strong signal
- “The résumé is dead, long live the portfolio.” Nobody checked the first half.Strong signal
- “You already know the answer” credits you with a conclusion nobody ever showed you.Moderate signal
How to read AI writing patterns
A tell is a sentence shape that language models reach for far more often than people do. None of these shapes is new, because models learned to write from us. Schools still teach in conclusion as the way into a closing paragraph, and speechwriters have opened line after line with the same words, the device rhetoric calls anaphora, for as long as anyone has given speeches. Every lesson here quotes a person who used the move well.
So a tell is evidence about frequency, not authorship. Each lesson gives a rate measured against a fixed sample of pre-2023 books, news and documentation, which shows how often people used the pattern before chatbots were drafting for them. One match means very little. A paragraph that stacks rhetorical questions, leans on not just ... but also, and ends on a tidy antithesis is worth a second look, because each move is ordinary alone and the checker reports how densely they cluster, per 100 words, against that same sample.
That is also where the checker stops. A match raises the probability that a model drafted the passage, but it cannot tell a student who has always written by the rule of three from a chatbot that picked up the habit from students like her. The pattern is a fine habit in the right place. The checker counts, and the decision stays with you.