AI rewrites its own text to evade detectors — study of 19,000 academic abstracts

Full text

AI-shaped vocabulary is becoming harder to separate from ordinary human writing

The analysis found a sevenfold rise in AI-associated vocabulary use

Rewritten AI text can enter published research under human authorship

Fırat Mıhcı, a computational linguist and founder of HumanizeMy.ai, argues that AI can correct its own writing style until common detection signals disappear entirely.

The analysis claims that a second AI model can rewrite an original AI draft until identifying traits become far less visible to reviewers.

Once a human editor approves that rewritten draft, it gets published under a real name instead of being labelled as machine-generated.

A study of nearly 19,000 abstracts

The study behind this claim examined 18,989 abstracts spanning three separate academic fields, published across the years 2019 through 2026.

The analysis compares word frequency trends before and after the release of widely used AI writing tools in late 2022.

In computational linguistics, AI-associated word use rose from 1.9 to 14.0 instances per 10,000 words after late 2022, roughly a seven-times increase.

Neuroscience abstracts showed a similar pattern, climbing from 1.7 to 8.7 instances per 10,000 words, close to a five-times increase.

Mathematics abstracts barely changed, moving only from 0.9 to 1.3 instances, which researchers treated as a control group for comparison.

The study noted that overlapping confidence ranges in mathematics made the small increase statistically indistinguishable from no change at all over the full study period.

One tracked term, "delve," showed up in 2.75% of surveyed abstracts during 2024 before falling to just 0.12% within an incomplete 2026 data set.

A feedback loop that confuses detectors

Mıhcı shared the findings as an open study meant to expose a blind spot in current AI detection methods used across many academic publishers today.

According to the study, this pattern creates a cycle in which AI wording gradually becomes accepted as ordinary human vocabulary.

Detection tools trained on older writing samples risk falsely flagging real authors whose natural vocabulary now resembles AI-generated text more closely than before.

AI tools trained on newer papers instead risk absorbing AI-influenced writing as a normal baseline, making future AI text harder to catch.

Words such as "intricate," "showcase," and "nuanced" were also tracked, alongside "underscore," a term commonly linked to AI-generated phrasing.

Editors and publishers reviewing these abstracts reportedly did not indicate that the wording carried any artificial origin before the manuscripts were accepted for publication.

"AI only needs its rewritten output to be accepted once as human. After that, the disguise becomes part of the answer key," said Fırat Mıhcı

The research does not claim every abstract containing these words was written by artificial intelligence rather than a human author.

It instead argues that AI-associated language has entered published writing in ways that make word-based detection increasingly unreliable over time.

Because the 2026 sample remains partial, the reported decline in words like "delve" may still shift once more data arrives.

If AI-shaped phrasing keeps entering the published record undetected, the very baseline used to define human writing may keep moving.

This pattern raises significant questions for any field relying on writing style alone to separate human output from machine output.