🔍 Read the full analysis: Six Patterns That Can Make AI Writing Less Pleasant To Read on ThorstenMeyerAI.com
Get business pricing on office and shipping supplies
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
TL;DR
A preprint by three linguists identifies six writing patterns that appeared more often in text from advanced AI models than in human comparison material. The researchers argue the patterns may make sentences harder to process, but they did not test reader comprehension directly, and the paper has not been peer reviewed.
Three linguists have identified six language patterns that appeared more often in text from advanced AI models than in comparable human writing, in a preprint posted October 21. The study compares 14 language models with human-authored material and argues that features such as dense noun strings and unusual word pairings may make some AI writing harder to process; it did not directly test whether readers understood the texts 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 model output in three settings: 120 casual chats with a simulated user, 45 coding sessions based on GitHub projects and 980 short persuasive essays.
The six patterns they tracked were rare word combinations, long gaps between words that belong together, omission of the optional word “that,” strings of three or more nouns, nouns used as verbs and descriptions in which inanimate objects appear to act like living things. The researchers compared the model texts with human material including subtitles, StackExchange posts, student essays and Wikipedia.
According to the paper’s summary, all nine frontier or near-frontier models used more rare word combinations and more descriptions of inanimate things acting like living things than the human comparison texts. The five smaller open models, which had lighter post-training, wrote more like people on some measures. The source article’s before-and-after rewrites are illustrations by its author, not quotations from the study; examples quoted from models are identified in the paper.
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 Dense AI Sentences Slow Reading
The study gives editors and AI users a set of specific patterns to check rather than relying on a general impression that a passage “sounds like AI.” If the patterns do create extra work for readers, small edits—moving a verb closer to its subject, unpacking noun strings or replacing an unusual phrase—could make text easier to follow without changing its underlying point.
The authors also connect the results to how models are trained. They argue that systems optimized for tasks such as math, coding and reasoning may be rewarded for dense, information-heavy answers rather than clear prose. That is an interpretation of the findings, not a demonstrated cause: the study does not establish that training objectives produced the patterns or that editing them improves comprehension for every reader.
The findings also bear on efforts to identify AI-written text. The source article says the researchers favor editing for clarity over relying on automated detectors, which have a weak record and have sometimes flagged human writing. The study itself, however, evaluates linguistic features; it does not test detector accuracy or prescribe a universal editing method.
Products Worth Considering
Logitech MX Mechanical Full-Size Wireless Mechanical Keyboard, Tactile Quiet Switches, Backlit Keys, Bluetooth, USB-C, Multi-Device, Graphit
Logitech MX Mechanical Wireless Illuminated Keyboard, Tactile Quiet, Graphite
HP Sprocket Photo Booth Instant Color Printer Bundle
As an affiliate, we earn on qualifying purchases.
How Researchers Compared the Texts
The researchers assessed output from 14 models, including nine they classified as frontier or near-frontier and five smaller open models from the Llama and Olmo families. Human comparison material was matched to the task where possible: subtitles for casual conversation, StackExchange posts for coding, student essays for persuasion and Wikipedia text.
One comparison involved translating sentences into Japanese and then back into English. The authors used this round trip as a control: if a feature became less common after translation and back-translation, they reasoned that the original model had selected a less ordinary wording even when a plainer alternative could preserve the meaning. The paper’s appendix reports $251.19 in API fees for data collection.
The paper appeared on arXiv on October 21, 2026. The source material describes it as a preprint that has not yet been peer reviewed. Its results concern English text and selected tasks and comparison datasets, not every model, language or writing situation.
“Why is AI-generated language so hard to understand?”
— The researchers, in the paper title
Limits of the Reading Evidence
The researchers did not test real readers’ comprehension, so the claim that the identified patterns make text harder to understand remains an interpretation informed by earlier studies. The supplied material does not report a direct experiment measuring reading time or comprehension across the study’s model and human texts.
The work is a preprint, not yet peer reviewed, and covers English and a limited set of models, prompts and comparison sources. The source summary also says a test on one small model is a proof of concept, not proof that changing training or post-training causes clearer writing. It remains uncertain how consistently the six patterns appear in other languages, tasks or later model versions.
Peer Review and Reader Testing
The immediate next step is for the paper to undergo peer review and for other researchers to examine whether the methods and comparisons support its conclusions. A direct study with readers could test whether the six patterns actually increase reading time or reduce comprehension, rather than inferring difficulty from linguistic features.
Until then, writers and editors can treat the list as a practical editing checklist, not a proven test of AI authorship or a guarantee that revised text will be clearer. The source material does not identify a scheduled follow-up study or a specific review date.
Key Questions
What six patterns did the study identify?
The list is rare word combinations, long gaps between words that belong together, omission of “that,” strings of three or more nouns, nouns used as verbs and inanimate objects described as if they were alive.
Did the study prove that AI writing is harder to understand?
No. The authors identified patterns they say may make text harder to process, but they did not test reader comprehension directly. The paper also has not yet been peer reviewed.
Which models used the patterns more often?
According to the study summary, all nine frontier or near-frontier models used more rare word combinations and more descriptions of inanimate things acting like living things than the human comparison material. The five smaller open models wrote more like people on some measures.
What can a writer do to make AI text easier to read?
The source article recommends checking for unusual pairings, moving verbs closer to their subjects, keeping “that” when it clarifies a sentence, breaking up noun strings and replacing awkward noun-to-verb conversions. These are editing suggestions, not proof that a passage was AI-generated.
Source: ThorstenMeyerAI.com
