AI tell · grade E
Short fragment paragraphs aren't the universal AI tell they're sold as
Short, fragment-length paragraphs get listed everywhere as proof a chatbot wrote the page: a whole idea landing in three words, standing alone as its own paragraph. Our own logs do not support that as a shared habit. Three of the twelve models we sampled did it as a matter of course: two share an opening habit, a clipped affirmation fragment in 37 percent of one model's turns, carried over into that same model's reasoning variant, and a third model closes turns the same way instead. The other nine did not reach for it at all. A pattern three models share and nine skip is a house style, not a chatbot universal, so underslop's checklist leaves it out of the scored list entirely.
What it is
The pattern is a paragraph collapsed to one short beat: a sentence, sometimes a fragment, standing alone between two longer ones. It reads as punctuation for the eye, a place for a reader to breathe or a writer to land a point hard. Style guides for journalism and advertising have taught this rhythm for decades. A fragment on its own line carries none of the context clues that make other tells legible, no bold label stacked over it, no sourceless claim propping it up. It is just short.
How often it shows up
Two entries in our own model logs document this pattern, and neither one is a cross-model habit in the broad sense. One (entry emp-013) found a clipped affirmation fragment opening 37 percent of a specific engine's replies, 41 of 110 turns; that habit is shared with the same model's reasoning variant but absent from every other engine we tested, so it accounts for two of the twelve models sampled, not one. The other (entry emp-026) found a third, unrelated model, tencent/hy3, closing turns on one-to-five-word fragment beats standing alone as their own paragraph, from a bare 'Done.' up to a full clipped sentence like 'Everything else is bonus.' Added together, the two entries cover three of the twelve models, and the other nine never reach for the pattern. Compare that to a genuinely universal tell like the validation opener, present in all 12 of 12 models we sampled (entry emp-001), and the gap is still the whole argument: this pattern never earned a place on the SCORED list because it never showed up as shared behavior across more than a handful of models.
Why readers flag it
Readers flag it anyway because the rhythm is genuinely common online, in headline writing, in ad copy, in personal essays that lean on short lines for effect. That popularity is exactly the problem with treating it as an AI signal: a real writer who picked up the habit from a magazine profile or a marketing brief gets read as a machine, while the three models that actually produce the pattern also produce other, more reliable tells right alongside it. Flagging the fragment alone catches the wrong group more often than the right one.
The fix
Do not edit a short paragraph on its own. Read the sentences around it. If the fragment sits next to an opener that praises the question, a stack of bold-label bullets, or an unsourced authority claim, fix those, the actual scored patterns, and the fragment usually reads fine once they are gone. If the fragment is standing next to ordinary prose with nothing else off, leave it. Short is not, on its own, evidence of anything.
Examples
Good question. The answer is simple. Speed matters. So does trust. So does clarity. Here's the thing: most teams get this backwards.
Speed matters. So does trust. So does clarity. Most teams get this backwards.
Studies show short paragraphs boost engagement. That's not a guess. That's the format every high-performing newsletter uses now. Done.
Short paragraphs keep readers reading. It's not speculation—it's the pattern every newsletter that performs well has already adopted.
Questions
What is a fragmented paragraph?
A paragraph reduced to a single short sentence or a fragment, often just a few words, set off on its own line rather than folded into the paragraph around it. It is a formatting choice about pacing, not a grammatical category.
What is an AI short paragraph?
Advice pages use the term for a chatbot habit of ending or opening a reply with a clipped one-line beat. In our own logs the habit is real but narrow: three of twelve models did it as a matter of course, and the rest did not reach for it.
Should I stop writing short paragraphs to avoid sounding like AI?
No, not on this evidence. Only three of the twelve models in our sample used the pattern consistently, so treating every short paragraph as suspect mostly punishes writers who already write that way, journalists, copywriters, anyone who learned pacing from print. Look at what sits around the fragment instead of the fragment itself.
Which AI models actually write this way?
Three, in the sample we logged. Two share an opening habit, a clipped affirmation fragment in over a third of one model's turns, carried over into that model's reasoning variant. A third, unrelated model closes turns the same way instead, on a one-to-five-word declarative standing alone. Neither pattern showed up in the other nine models we tested.
Is a one-word paragraph always a red flag?
No. A single word standing alone works the same way in human writing it always has, as emphasis. The signal worth checking is whether the paragraphs around it carry the patterns that actually are well-evidenced: a validation opener, a bold-header list, a claim propped up by a source that is never named.
underslop marks this tell and the rest of the consensus set in any draft you paste, using the same open lint that scores it here. The free tier caps at 500 words a run, and your text is held for 0 seconds.