AivexaNewsSearch
AI news for builders and product teamsChecked every hour

Researchers find six patterns that make AI text so exhausting to read

Collected Oct 26, 2026

Researchers at New York University and the University of Texas at Austin have identified six linguistic patterns that frontier language models overuse, making their output systematically harder for people to parse. The findings, reported by The Decoder, also note a contrast: smaller models write more like humans than the largest ones do.

The researchers attribute the overuse to training on code and math, where being correct is rewarded but clarity is not. In practice, that suggests models are optimised for passing checks rather than for readable prose, and the habit carries over into ordinary writing. This is a plausible reading rather than a reported result, but it fits the well-known pattern that benchmarks reward verifiable answers.

The work names six recurring patterns but does not detail them here; developers evaluating model output should treat readability as a separate axis from accuracy. A likely trade-off is that scaling up and tuning on reasoning-heavy data buys correctness at some cost to how comfortable the text is to read.

Why it matters: teams shipping LLM-generated copy, docs or chat interfaces can expect the most capable models to be the least pleasant to read, based on these findings. Inference: if the cause is code and math training, prompt-level fixes may only partly help, and product teams may need style filters or smaller models for reader-facing text.

Read at The Decoder

Based on reporting from the original publisher. Visit the source for full context and later updates.

Publisher excerpt

A study from New York University and the University of Texas at Austin shows that frontier language models systematically overuse linguistic patterns that are hard for humans to parse. The researchers trace the problem to training on code and math, where correctness gets rewarded but clarity doesn't. Smaller models, by contrast, write more like humans. The article Researchers find six patterns that make AI text so exhausting to read appeared first on The Decoder .