🔍 Read the full analysis: A Reader’s Guide To Less Tiring AI Text: Six Patterns To Change on ThorstenMeyerAI.com
Get hardware and tech essentials delivered free with Prime
- Fast, free delivery on millions of items
- Prime Video, Amazon Music and more included
- Member-only deals all year
TL;DR
A preprint by three linguists identifies six writing patterns used more often by advanced AI models than by human writers, including noun stacks and unusual word combinations. The authors argue these patterns can make text harder to process, but they did not test reader comprehension directly; their suggested remedy is editing, not relying on AI detectors.
A preprint posted October 21 by three linguists compares writing from 14 language models with human text and identifies six patterns that the models used more often, including unusual word combinations and long gaps between related words. The authors connect those features to earlier research on reading difficulty, but say they did not test whether readers actually understood the model-written passages less well.
The paper, titled Why is AI-generated language so hard to understand?, was written by Thomas Hikaru Clark of New York University and Matias B. Martin and Kyle Mahowald of the University of Texas at Austin. The researchers examined outputs from 14 models across casual chats, coding sessions and short persuasive essays, comparing them with human writing such as subtitles, StackExchange posts, student essays and Wikipedia.
The six patterns are rare word combinations; long distances between words that belong together; omission of optional “that”; strings of three or more nouns; nouns used as verbs; and descriptions in which objects take actions more commonly associated with living things. In the paper’s examples, a sentence about a guitar “teaches you zero chords,” while another says code “swallows” an error. The article’s alternative phrasings are illustrations, not direct quotations from model outputs.
The authors report that all nine frontier or near-frontier models used rare word combinations and descriptions of objects acting like living things more often than human writers did. Five smaller open models, from the Llama and Olmo families, wrote more like people on some measures. The study also translated sentences into Japanese and back into English as a test of whether less unusual wording could preserve the meaning.
The study at a glance
14 models, 3 settings, 6 patterns
Three linguists compared AI text with human text of a similar kind, then checked which hard-to-read patterns the models chose on their own.
The models
Frontier and near-frontier models are highlighted.
The comparison
- Chat vs. film and TV subtitles
- Code vs. StackExchange posts
- Essays vs. 2,934 high-scoring student essays (grades 6 to 12) and Wikipedia
- Control: every sentence translated into Japanese and back. If a pattern gets rarer, the model chose it over a plainer option.
- Cost: $251.19 in API fees for the whole data collection.
Held for all nine
More rare word pairs and more objects that act like living things than human writers.
Grew with training
In the Olmo family the effects rose from base to instruct to the reasoning model, Olmo Think.
Fixable
Claude Opus 5.5 scored lower than Opus 5 on 19 of 21 difficulty comparisons.
Source: Clark, Martin & Mahowald, Why is AI-generated language so hard to understand?, arXiv preprint, 21 October 2026 (not peer reviewed). Cost figure from the paper’s appendix A.6.
Powered by Thorsten Meyer AI
Why Editing May Beat Detection
The study gives editors and people using AI a set of specific revision targets, rather than treating “AI-sounding” prose as a single vague quality. A writer can check whether a sentence delays its main verb, packs several nouns together or uses an unusual metaphor, then rewrite it in more direct language. The source article recommends editing the text instead of using a detector to decide whether it is AI-generated.
That distinction matters because the paper examines features of language, not authorship as a reliable yes-or-no test. A sentence with one of these patterns is not proof that AI wrote it, and a human writer can use the same constructions. The article also notes concerns about detectors flagging human writing. The findings offer possible clues for improving readability, not a method for establishing who wrote a passage.
The authors suggest that training priorities may help explain the patterns: systems rewarded for dense answers in areas such as maths, coding and reasoning may favor compact or unusual phrasing over plain prose. That is the authors’ proposed explanation, not a cause established by the comparisons. For readers, the practical point is narrower: clarity can often be improved by making relationships between words easier to follow.
Products Worth Considering
Apple MacBook Pro 16-inch Laptop with M5 Pro, 18-Core CPU, 20-Core GPU, 48GB Memory, 2TB SSD, Space Black
LG 24G414B-B 24-Inch UltraGear Full HD IPS Gaming Monitor
Logitech MX Keys S Wireless Keyboard, Graphite
As an affiliate, we earn on qualifying purchases.
How the Researchers Compared Text
The study drew on three kinds of model output: 120 simulated casual chats, 45 coding sessions on real GitHub projects and 980 short persuasive essays. The human comparison material varied by task, including film and television subtitles for chat and StackExchange posts for coding. The researchers also used 2,934 high-scoring essays by US students in grades 6 to 12 and Wikipedia text.
To test whether a model’s wording was unusually specific when a plainer option was available, the researchers translated sentences into Japanese and back into English. They treated patterns that became less common after this round trip as evidence that the original wording could have been expressed more ordinarily. According to the paper’s appendix, the data collection cost $251.19 in API fees. The authors cite earlier studies linking some of the six features to slower processing, but this paper itself did not measure how real readers performed.
“Why is AI-generated language so hard to understand?”
— The paper’s title
What the Study Cannot Establish
The paper is a preprint and has not yet been peer reviewed, according to the source article. It covers English-language text, so the findings do not establish whether the same patterns appear in other languages. The researchers also did not test whether readers understood the passages less accurately, took longer to read them or found them less satisfying.
The reported differences do not show that every output from an advanced model is difficult to read, or that every use of the six patterns makes a sentence unclear. The study’s comparisons identify tendencies in the sampled material; the source does not provide evidence that the proposed training explanation is confirmed. It is also unclear from the material provided how the patterns vary across all tasks, prompts and model versions.
Peer Review and Reader Testing
The next reported milestone is peer review; the source article does not give a timetable or say whether a journal submission is underway. Further research could test the same patterns with readers directly, measuring comprehension and reading time, and examine whether the results hold across languages and additional models.
For now, the paper’s findings support a practical but limited approach: revise sentences that delay key relationships or use dense, unfamiliar phrasing, while treating each edit as a judgment about clarity rather than proof of AI authorship. Whether those revisions measurably improve comprehension remains to be tested.
Key Questions
What are the six patterns identified in the paper?
They are rare word combinations, long gaps between related words, omitted optional “that,” strings of three or more nouns, nouns used as verbs and descriptions of objects acting like living things.
Did the researchers prove that AI writing is harder to understand?
No. The authors compared writing patterns and drew on earlier research linking some features to processing difficulty, but they did not test readers’ comprehension of the texts in this study.
Which models showed the patterns most often?
The paper reports that all nine frontier or near-frontier models in its sample used rare word combinations and descriptions of objects acting like living things more often than human writers did. The study also included five smaller open models.
Can these patterns prove that a passage was written by AI?
No. The patterns are features of writing, not a reliable authorship test. Human writers can use them too, and the source article cautions against relying on AI detectors as a substitute for editing.
Has the study been peer reviewed?
Not according to the source article. It describes the paper as a preprint, posted on arXiv on October 21, 2026.
Source: ThorstenMeyerAI.com
