Research report
Does Turnitin Detect AI-polished Human-written Research Papers with GPT-5.6 Sol?
Experiment: Benchmarking Turnitin Detection of AI-Polished Graduate Essays
Cite this research
In a benchmark of 500 AI-polished graduate essays totaling 132,275 words, Turnitin achieved a word-level accuracy of 47.54% only, correctly flagging 62,879 words as AI-generated.
According to Turnitin0 Research, the tool correctly identifies less than half of AI-polished words in graduate essays, with accuracy dropping to 0% in some majors.
- 500 articles tested
- Checker: Turnitin
- LLM: gpt-5.6-sol
- Corpus: PLOS
- 1 linked dataset
- Detection screenshots available
- Published July 29, 2026
Peng, J. (2026). Does Turnitin Detect AI-polished Human-written Research Papers with GPT-5.6 Sol?. Turnitin0 Research. https://www.turnitin0.com/research/reports/does-turnitin-detect-ai-polished-human-written-research-papers-with-gpt-5-6-sol
Licensed under CC0 1.0 Universal. You may copy, modify, distribute, and cite this research—including text, figures, tables, and datasets—without asking permission.
Overview
This benchmark evaluates Turnitin's ability to detect AI-polished human-written graduate essays. The dataset comprises 500 essays from the PLOS corpus, each polished by GPT-5.6-sol, totaling 132,275 words. The study measures word-level accuracy—the proportion of words correctly flagged as AI-generated.
- Overall word accuracy is 47.54% (62,879 correct words out of 132,275 total words).
- Domain performance varies: Business essays achieve 64.36% accuracy (809 correct words out of 1,257), while Humanities essays drop to 28.27% (834 correct words out of 2,950).
- Major-level extremes: Chemistry (94.10%, 271 correct out of 288) and History (95.37%, 268 correct out of 281) show high accuracy; Civil Engineering, Criminal Justice, and Mechanical Engineering each score 0% accuracy.
- Word count buckets: Essays around 400 words have 47.66% accuracy (62,686 correct out of 131,539), while the single 800-word essay achieves 26.22% (193 correct out of 736).
- Dataset: 500 graduate-level essays from the PLOS corpus, polished by GPT-5.6-sol using an AI-polished generation method.
- Domains: Business, Humanities, Psychology, and STEM; 18 majors represented.
- Word count: 132,275 total words (excluding references/bibliography), with essays ranging from 400 to 800 words.
- Evaluation metric: Word-level accuracy—percentage of words correctly flagged as AI-generated by Turnitin.
- Analysis dimensions: Overall, domain, major, and word count bucket.
- Scoring unit: accuracy is measured in words. First identify correct paragraphs under the expectation below, then sum the words in those paragraphs for the correct-word count; total words are summed over all counted paragraphs.
- Correct paragraph: a counted paragraph that was flagged as AI. Correct words are the words in those paragraphs.
- References excluded: from a References/Bibliography-style heading onward, those paragraphs are left out of word counts.
- Small sample sizes in some majors: Several majors have only 1–2 essays, limiting statistical reliability.
- Total articles: 500
- Correct words: 62,879 out of 132,275
- Word accuracy: 47.54%
- Business: 64.36% accuracy (809 correct out of 1,257 words)
- Humanities: 28.27% accuracy (834 correct out of 2,950 words)
- Psychology: 46.89% accuracy (5,058 correct out of 10,787 words)
- STEM: 47.90% accuracy (56,178 correct out of 117,281 words)
- Highest accuracy: History (95.37%, 268 correct out of 281), Chemistry (94.10%, 271 correct out of 288)
- Lowest accuracy: Civil Engineering, Criminal Justice, Mechanical Engineering (0% each)
- Other notable: Biology (59.26%, 41,711 correct out of 70,392), Computer Science (26.89%, 749 correct out of 2,785)
- 400-word essays: 47.66% accuracy (62,686 correct out of 131,539 words, 499 articles)
- 800-word essays: 26.22% accuracy (193 correct out of 736 words, 1 article)
Detection report screenshots
Turnitin marks suspected AIGC text in cyan.