AI Detection Study Finds 15% of Top Law Review Articles in 2026 Show AI Use

A new empirical study has documented the rapid adoption of AI writing tools in elite legal scholarship, finding that roughly one in six articles published in top-tier law reviews this year contain detectable AI-generated content.

Julian Nyarko, a law professor who specializes in empirical legal studies, analyzed 571 articles from the flagship law reviews of the top 14 U.S. law schools, comparing publications from 2020 (before GPT-3's release), 2025, and 2026. The findings, first reported on The Volokh Conspiracy legal blog, show AI use quintupling in a single year.

The methodology

Nyarko fed 17.3 million words of legal scholarship through Pangram 3.3.2, an AI detection tool with a claimed false-positive rate of 0.02% for academic writing. He used AI coding assistants to segment each article into abstract, body text, and footnotes, then manually verified every flagged passage to eliminate artifacts that might trigger false positives.

The 2020 baseline—133 articles published before GPT-3's May 29 launch—showed zero AI signal, suggesting the detection method produces minimal false positives on legal writing.

Sharp increase from 2025 to 2026

In 2025, just 9 of 273 articles (3.3%) showed any AI signal. By 2026, that figure jumped to 25 of 165 articles (15.2%)—a statistically significant increase with a p-value of 0.000012.

The intensity of AI use also increased. In 2026, nine articles contained more than 5% AI-generated words, and one article registered as 64.6% AI-written. The overall share of AI-generated words across all 2026 articles reached 1.10%, up from 0.20% in 2025.

Most AI signals appeared in article body text and abstracts rather than footnotes. Twelve of the 25 flagged articles showed AI content in multiple sections.

Why it matters

The findings arrive as legal academia grapples with disclosure norms around AI assistance. Unlike empirical research—where analysis often precedes writing—doctrinal legal scholarship typically develops arguments through the writing process itself. Whether AI assistance undermines intellectual rigor or simply automates tedious tasks remains contested.

Nyarko's study cannot determine whether authors, student editors, or both introduced the AI content, nor can it distinguish between light editing assistance and wholesale AI drafting. But the data suggests AI tools are becoming standard in legal scholarship production, even at the most selective journals.

The research also highlights a detection challenge: Nyarko noted that Pangram struggles with AI-generated legal contracts, which often reproduce human-written templates verbatim. Similar limitations may affect law review detection accuracy.

Nyarko disclosed that he used Claude AI extensively throughout his own data collection and analysis—a transparency practice he suggests should become standard. The findings were published by Orin Kerr, a Stanford Law professor, on The Volokh Conspiracy.

"While I remain somewhat agnostic about AI-assisted writing," Nyarko wrote, "I am less agnostic about having to read" what he termed "Claude-slop"—AI-generated prose heavy on filler phrases and light on substance.