🔍 Read the full analysis: Why AI Text Can Be A Struggle To Read: Six Patterns And Solutions 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 researchers at New York University and the University of Texas at Austin identifies six language patterns that appeared more often in text from advanced AI models than in human writing. The authors say the patterns may make text harder to process, but they did not test readers’ comprehension, and the paper has not been peer reviewed.
Researchers from New York University and the University of Texas at Austin have identified six language patterns that appeared more often in writing from advanced AI models than in comparable human text. Their October 21 preprint argues that the patterns may make AI responses harder to process, while stressing that the study did not directly test whether readers understood them less well.
In the paper, “Why is AI-generated language so hard to understand?”, Thomas Hikaru Clark, Matias B. Martin and Kyle Mahowald examined writing from 14 language models across casual chats, coding sessions and short persuasive essays. The set included nine frontier or near-frontier models and five smaller open models. Human comparisons included subtitles, StackExchange posts, student essays and Wikipedia text.
The six patterns were uncommon word pairings, 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 that give inanimate objects actions associated with living things. The authors report that all nine larger models used more rare word combinations and more descriptions of objects acting like living things than human writers did. Smaller models with lighter post-training were more similar to people on the patterns overall.
To test whether a model had chosen an unusual construction when a more ordinary alternative was available, the researchers translated sentences into Japanese and back into English. They treated a pattern becoming less common after this process as an indication that the original wording was not the only way to express the meaning. The paper’s appendix puts data collection costs at $251.19 in API fees. The article’s before-and-after editing examples are illustrations, not quotations from the study.
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
How These Patterns Affect Reading
The findings give editors and AI users a practical way to describe why some fluent, grammatically correct answers can still feel effortful to read. A reader may have to hold a subject in memory while a long clause intervenes, untangle a noun stack, or pause to interpret a rare pairing. The authors connect these features to earlier research on processing difficulty, but this study itself does not establish that any one pattern causes misunderstanding.
The paper’s suggested response is editing for clarity: move verbs closer to their subjects, unpack dense noun strings, retain “that” where it helps show sentence structure, and choose familiar wording. These changes can also help human-written material. The study does not show that every model response needs editing or that the six patterns account for all difficult AI prose.
The authors also caution against treating automated AI detection as a substitute for editing. The source article notes concerns about detectors misidentifying human writing; the study’s central contribution, however, is its comparison of language patterns, not a test of detection tools.
Products Worth Considering
Dell UltraSharp U2725QE 27-Inch 4K UHD IPS Black Monitor
Logitech MX Master 4 for Mac, Ergonomic Bluetooth Mouse with Haptic Feedback, Ultra-Fast Scrolling, USB-C Charging, for MacOS, iPadOS, Windo
FLEXISPOT 32-Inch Electric Standing Desk Converter with Keyboard Tray
As an affiliate, we earn on qualifying purchases.
How Researchers Compared the Text
The researchers sampled text in three settings: 120 casual chats with a simulated user, 45 coding sessions on real GitHub projects, and 980 short persuasive essays. To make comparisons more relevant, they used human material matched to the types of writing, including subtitles for conversational text and StackExchange posts for coding.
The study is a preprint on arXiv, not a peer-reviewed journal publication. Its authors describe the translation round trip as a way to examine whether unusual wording was necessary to preserve meaning. It is a limited test of word choice, not a direct measure of how quickly people read or whether they understood a passage.
““Why is AI-generated language so hard to understand?””
— The paper’s title
Limits of the Reading Evidence
The results come from one preprint that has not been peer reviewed and cover English-language writing. The researchers did not ask readers to rate the samples, measure reading speed, or test comprehension, so the claim that the patterns make text harder to understand remains an interpretation supported by earlier studies rather than a result directly measured here.
The comparison also depends on the selected models, prompts and human-text samples. The findings do not establish why every model uses these forms or show that training for coding, mathematics or reasoning caused them. The authors propose that optimization for dense, information-rich answers may favor wording that is less clear, but their test of a small model is described as a proof of concept, not proof of that explanation. The study also does not establish how common these patterns are in everyday AI use.
What Further Testing Can Show
The next step is independent review and replication across more models, writing tasks and languages. Direct reader studies could test whether the six patterns change comprehension, reading time or perceived clarity, and whether editing them improves those outcomes without changing meaning.
Until that evidence is available, the findings support a modest practical approach: treat the patterns as editing prompts, not a definitive test of authorship or readability. The paper does not provide a validated checklist or show that removing every instance will make a passage easier for every reader.
Key Questions
What are the six patterns identified in the paper?
They are rare word combinations, long gaps between related words, omission of “that,” noun stacks of three or more words, 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 linked to processing difficulty in earlier research, but they did not test readers’ comprehension or reading speed in this study.
Which models showed the patterns?
The researchers examined 14 models. They report that all nine frontier or near-frontier models used more rare word pairings and descriptions of objects acting like living things than the human comparison texts. Smaller models with lighter post-training were more similar to human writing overall.
What can a writer do to make an AI draft easier to read?
Consider replacing unusual pairings with familiar words, moving verbs closer to their subjects, unpacking noun stacks, and using a clarifying “that” where it helps. These are editing suggestions, not fixes proven by reader testing in this paper.
Has the paper been peer reviewed?
No. The paper was posted on arXiv on October 21, 2026, and the source describes it as a preprint that has not yet been peer reviewed.
Source: ThorstenMeyerAI.com
