Paragraph Machine

Every paragraph follows the same template: topic sentence, two supporting sentences, transition. "Identical bricks, identical height, forever." If 30%+ of paragraphs open and close the same way, it's likely machine-written. Fourier analysis can detect the periodicity — AI repeats rhetorical forms every 50-100 words.

Read enough AI output and you start to feel it before you can name it. A rhythmic monotony. Every paragraph the same weight and shape.

Here's what's happening structurally: topic sentence states the point, two or three supporting sentences elaborate, a transitional sentence bridges to the next paragraph. Repeat. The paragraphs land at roughly the same length -- 4-6 sentences each, give or take nothing. Human writing lurches. It digresses. It drops a one-sentence paragraph for emphasis, then sprawls for half a page. AI writing does none of that.

We think "periodicity" is the right name for the structural tell. Run Fourier analysis on the rhetorical features of AI text and you find repeating patterns every 50-100 words. Human text doesn't do this. It's our contribution to the pattern catalog -- others have called AI writing "formulaic," but nobody had named the specific frequency.

One AI paragraph can pass. Five in a row and trained readers feel the metronome, even if they can't say why.

Inside Higher Ed called it "identical bricks, identical height, forever." Pangram's detection research put a number on it: 30%+ of paragraphs sharing the same structural open-and-close pattern flags likely machine authorship.

The template, repeated Furthermore, the team implemented several key changes. These changes led to improved outcomes across all metrics. Moreover, the results exceeded initial projections.

This success demonstrated the value of the approach. Additionally, stakeholder feedback was overwhelmingly positive. The findings suggest a clear path forward.

Looking ahead, several opportunities remain. The organization plans to expand these efforts. Ultimately, these developments point to continued growth.
Transition word pattern Moreover, the analysis revealed significant trends. These trends align with previous research. Furthermore, the data supports broader conclusions.

Additionally, several factors contribute to this outcome. First among these is the quality of implementation. It is also worth noting the role of timing.

Consequently, organizations should consider these findings carefully. The implications extend beyond the immediate context. In conclusion, this represents a meaningful development.
Human variation (for contrast) The code worked.

I mean — it compiled. Whether it actually worked was another question entirely, one I spent the next three hours discovering the answer to, mostly by staring at log files and questioning my career choices.

But the point is: the template approach failed here.
30%
structural repetition threshold for likely AI authorship (Pangram)
50-100
word interval at which AI repeats rhetorical forms (Fourier analysis)
0.15-0.30
AI writing burstiness score vs 0.60-1.00+ human (GPTZero)
4x
fewer unique words in AI text vs human (VU Amsterdam)

The "Identical Bricks" Phenomenon

Inside Higher Ed published "Ways to Distinguish AI-Composed Essays" (July 2024), coining the "identical bricks" image. Their observation: AI paragraphs are structurally interchangeable. Swap paragraph 3 with paragraph 7 and nothing breaks. Human paragraphs have positional logic. They build on what came before, and they fall apart if you rearrange them.

Recognizing AI Structures in Writing

Michelle Kassorla's "Recognizing AI Structures in Writing" documented the template at the paragraph level: topic sentence, support, support, transition. She found it in over 80% of AI-generated student submissions. The skeleton is hard to miss once you know what to look for.

The 30% Structural Threshold

Pangram put a number on it. Over 30% of paragraphs following the same structural opening-closing pattern flags likely AI authorship. Below that threshold, the text might just be a mediocre writer. Above it, the machine is showing.

Burstiness and Paragraph-Level Uniformity

GPTZero's burstiness metric puts numbers on the intuition. Human writing scores 0.60-1.00+ (high variation in sentence complexity). AI writing: 0.15-0.30. Mechanical uniformity. Low burstiness is what The Paragraph Machine feels like when you quantify it.

Fourier Analysis: A Novel Approach

Fourier analysis for detecting rhetorical periodicity shows up in academic stylometry papers, but nobody in the public AI-detection conversation has named it as a distinct tell. We think it should be.

Student Essay Detection

Writing instructors spotted this before anyone else. Michelle Kassorla documented how AI-generated student essays all shared the same internal rhythm, even when the vocabulary varied. The words changed from essay to essay. The skeleton didn't.

Michelle Kassorla →

CNET AI Article Errors

CNET published 77 AI-written articles in 2023. The math errors got the headlines, but the structural uniformity was just as damning. Every section followed the same paragraph pattern. Swap the intro from the credit card article into the mortgage article and it reads fine. That's the tell.

Washington Post →

LinkedIn Content Mills

Originality.ai analyzed LinkedIn and estimated 54% of long-form posts were AI-generated. Paragraph rhythm was one of the detection signals. Scroll through LinkedIn's thought-leadership feed and the structural uniformity is hard to miss -- post after post with the same cadence, the same arc from setup to takeaway.

Originality.ai →