AI tell · grade E
One ChatGPT model's sentence structure is a macro-template, not a single shape
Ask what ChatGPT's sentence structure looks like and most answers point at one sentence: short, declarative, maybe an em dash in the middle. That misses the actual pattern. In one model in our own logs, a rigid three-slot macro recurred almost identically across five unrelated conversations, a finance question and a personal scheduling question among them. Slot one validates the question and pivots into content, often with a dash or a stock opener. Slot two turns the body into bare capitalized labels, forty-six of them across the sample, each launching a cluster of bullets, 288 in total, with zero bold formatting anywhere. Slot three closes with an offer to do more or a forced-choice question sending the exchange back. No individual sentence in that shape looks wrong. The tell is that the whole shape repeats regardless of the subject.
What it is
A response macro has three named slots. The open pairs a validation of the question with a pivot, sometimes a dash, sometimes a stock phrase that announces a short version is coming. The body does not stay in prose; it breaks into a run of bare capitalized labels, each one launching a small bullet cluster, more label-and-bullet pairs than the content usually needs. The close hands the exchange back with an offer to do more or a question that forces a choice among options the model just listed. None of the three slots is unique to one model in isolation, but the exact combination, assembled at this rigidity and recurring near-identically across unrelated topics, is what makes it a template rather than a style.
How often it shows up
Two of the three slots are close to universal by themselves: some form of structural scaffolding over a casual or emotional exchange showed up in all twelve models we logged, and a dash used to pivot out of an opener showed up in nine of twelve. The full three-slot macro assembled at this rigidity, capitalized labels with zero bold and a closing offer-menu, was a single-model finding: one model out of the twelve, recurring near-identically across five separate conversations on unrelated subjects. A parallel pattern in the same model's sentences, stringing three parallel items for rhythm, showed up roughly twenty-four times in one sample and appeared in four of the twelve models overall, broader across the corpus than the macro's single-model footprint. We did not track how rigidly that triad pattern repeats within any one model the way we tracked the macro's five-conversation recurrence, so the two counts are not measuring the same kind of consistency.
Why readers flag it
Read any one reply from this macro on its own and it looks like reasonable, well-organized writing. What gives it away is reading two or three replies back to back on completely different subjects and noticing the identical shape underneath: the same validate-and-pivot open, the same label-and-bullet body, the same offer-menu close, regardless of whether the question was about money, a schedule, or something else entirely. A person composing fresh answers to different questions does not reach for the same architecture every time. A template does.
The fix
Break the macro deliberately rather than trying to fix any single sentence in it. Vary which slot opens a reply, sometimes start with the direct answer and skip the validation clause entirely. Convert the label-and-bullet body into connected prose whenever the content is not a genuine, itemizable list. Delete the closing offer-menu and the forced-choice question when the answer you already gave is complete; neither one adds information, and both exist to keep the exchange going rather than to finish it.
Examples
Totally fair — here's the breakdown. Options for weekend trips: budget stays under $200, drive time stays under three hours, and dog-friendly lodging is available — so state parks usually win. Would you like me to pull together a shortlist?
Here's the breakdown. Weekend trips with a $200 budget, under three hours of driving, and dog-friendly lodging usually point toward state parks. Want me to put together a shortlist?
Short answer: it depends on your risk tolerance, but here's the thing, most people get this wrong. A stock buyback isn't wasted cash, it's returned capital, and it can help holders, boost the share price, and signal confidence.
It depends on your risk tolerance, but most people get this wrong. A stock buyback is not wasted cash. It is returned capital, and for the people who hold the stock, it can be good: it can lift the share price and signal that management thinks the stock is undervalued.
Questions
How does ChatGPT structure sentences?
Any single sentence usually looks ordinary. The pattern worth naming sits one level up: a repeating three-part reply shape, an opener that pivots into content, a body forced into labeled bullet clusters, and a close that hands the exchange back to you. We only confirmed this exact assembly in one model out of the twelve we logged, recurring near-identically across five unrelated conversations inside that model.
What is the commonly used AI sentence structure?
There is no one sentence shape to point at. What we found in our own logs was a response-level template that recurs across unrelated topics inside a single model, not a signature sentence type.
Why does ChatGPT use short sentences?
Short, declarative sentences read as confident and are easy to scaffold into bullet clusters, which fits the label-and-bullet body slot in the macro we logged. The length is a side effect of the template, not the template itself.
What are the four types of sentence structure?
Simple, compound, complex, and compound-complex, the standard grammar categories. None of the four is itself an AI tell. The pattern we measured sits above sentence grammar, in how an entire reply is assembled and repeated.
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.