🔍 Read the full analysis: Can You Make AI Writing Less Work To Read? Six Patterns To Tackle on ThorstenMeyerAI.com
Get the latest gadgets 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 compares writing from 14 language models with human text and identifies six language patterns more common in some AI outputs. The authors say editing for clearer phrasing may help, but they did not test readers’ comprehension, and the study has not been peer reviewed.
Three linguists have identified six language patterns that appeared more often in writing from advanced AI models than in comparable human text, in a preprint posted October 21. The study compares outputs from 14 language models with human writing across several settings; its authors argue that the patterns may make text harder to process, but they did not test whether readers actually understood it 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. It examines text from 14 models in casual chats, coding sessions and short persuasive essays, using human-written material of similar kinds as comparisons.
The six features are uncommon word combinations, long distances between words that belong together, omitted optional “that,” noun strings of three or more words, nouns used as verbs, and descriptions that give inanimate objects actions associated with living things. The source article says all nine frontier or near-frontier models in the study used more uncommon word combinations and more object-as-living-thing descriptions than human writers did. Smaller open models with lighter post-training were described as writing more like people.
The researchers also translated sentences into Japanese and back into English as a control. The source article says this altered wording while retaining meaning, allowing the authors to check whether a pattern became less common when a plainer alternative was available. Its examples illustrate possible edits, such as replacing a dense noun stack with a direct phrase or moving a verb closer to its subject. Those examples are the article writer’s illustrations, not quotations from the models.
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 AI prose can feel effortful
The findings offer a set of specific editing targets for people who use AI-generated text in emails, documentation, essays or customer-facing material. Instead of treating “AI-sounding” writing as one vague quality, writers can look for compressed noun strings, unusual word pairings and sentence structures that make readers hold information in memory while waiting for the sentence to resolve.
The authors’ proposed explanation is that training models to perform well on tasks such as mathematics, coding and reasoning may reward dense answers more than readable ones. That is an interpretation offered by the researchers, not a cause established by the study. The practical advice in the source article is to edit for clarity rather than rely on AI detectors, which it says have a weak record and have sometimes flagged human writing incorrectly. The paper itself, however, does not test editing advice against reader outcomes.
Products Worth Considering
Samsung 32-Inch Odyssey OLED G8 (G80SH) 4K Gaming Monitor
Jackery Explorer 300 Portable Power Station, 292Wh
Samsung 27-inch Odyssey OLED G6 QD-OLED QHD Gaming Monitor
As an affiliate, we earn on qualifying purchases.
How the researchers compared writing
The study combines three types of generated text: 120 simulated casual chats, 45 coding sessions on real GitHub projects, and 980 short persuasive essays. The comparisons include film and television subtitles, StackExchange posts, 2,934 high-scoring essays by US students in grades 6 to 12, and Wikipedia. These sources provide reference material for different kinds of writing rather than a single universal standard for human prose.
The 14 models included nine the source article classifies as frontier or near-frontier, alongside five smaller open models from the Llama and Olmo families. The source identifies Claude models, models in OpenAI’s GPT-5.6 family, GLM-5 and Kimi K2.5 among the larger group. The paper appeared on arXiv as a preprint, so it has not yet completed peer review. Its reported data collection cost was $251.19 in API fees, according to the paper’s appendix, as relayed by the source article.
““Why is AI-generated language so hard to understand?””
— The study’s authors, as summarized by ThorstenMeyerAI.com
What the preprint cannot establish
The work is one preprint and has not been peer reviewed. It covers English text, and the authors did not measure whether real readers read more slowly, understood less, or preferred edited versions. The source material says earlier studies have linked some of the identified features to slower reading, but the current paper’s comparison does not itself demonstrate that these model outputs impair comprehension.
The evidence also does not show that every AI response contains all six features or that every instance makes writing harder to understand. The model groups, tasks and human comparison texts set limits on how broadly the results can be applied. The source article characterizes the small-model test as a proof of concept, not proof that post-training is the cause. Further research would be needed to establish which features affect readers and whether the proposed edits improve comprehension.
Peer review and reader testing
The next evidence to watch for is whether the paper is peer reviewed and whether researchers test the six patterns with readers directly. Such work could compare reading time, comprehension and responses to original and edited passages, while checking whether results hold across genres, languages and model versions. No peer-review outcome or follow-up test is provided in the source material.
For now, the study supplies a diagnostic list rather than a validated editing standard. Readers and editors can use it to spot potentially dense phrasing, but claims that the edits make AI writing easier to understand remain to be tested. The source article’s recommendation to revise rather than depend on detectors is practical advice, not an outcome measured by this study.
Key Questions
What are the six patterns identified in the study?
The paper identifies uncommon word combinations, long gaps between related words, omitted optional “that,” strings of three or more nouns, nouns used as verbs, and descriptions that make inanimate objects seem to act like living things.
Did the researchers prove that AI writing is harder to understand?
No. The study compared language patterns in model-generated and human writing, but the authors did not test real readers’ comprehension. The paper’s claims about processing difficulty require further reader testing.
Has the study been peer reviewed?
No peer review is reported in the source material. The paper was posted to arXiv as a preprint on October 21, 2026.
What does the study suggest writers do?
The source article recommends editing for direct wording: clarify noun-heavy phrases, keep related words close together and prefer familiar expressions. These are practical suggestions; the study did not test whether the edits improve reader comprehension.
Source: ThorstenMeyerAI.com
