# Stanford study finds AI-assisted drafting in over a third of Santa Clara court complaints

> Researchers say signs of AI-assisted drafting rose from near zero in 2022 to more than a third of examined Santa Clara complaints in 2026.

- Status: official (From the company, authors or authority themselves, with a link to their own page.)
- Topic: AI in Practice
- Published: 2026-11-19T16:07:46.314655+00:00
- Updated: 2026-11-19T16:08:43.674080+00:00
- Web page: https://ai-news.ikangai.com/story/stanford-study-finds-ai-assisted-drafting-in-over-a-third-of-santa-clara-court-c/
- Seen in: Stanford HAI
- Change: research (Research)
- Applies to: Santa Clara County complaints

## What changed

Stanford researchers estimate that signs of AI-assisted drafting rose from near zero in 2022 to more than a third of examined Santa Clara County complaints in 2026, and they propose the PhantomDocket benchmark and ScoreMix calibration method for detecting such use.

Stanford researchers estimate that signs of AI-assisted drafting rose from near zero in 2022 to more than a third of examined Santa Clara County complaints in 2026. The Stanford Institute for Human-Centered AI and its Regulation, Evaluation, and Governance Lab (RegLab) published the findings. For 2026 non-debt-collection cases, prevalence was 38% in Santa Clara County and 36% in San Francisco County, compared with 17% and 16% in Tulsa and Oklahoma counties in Oklahoma.

AI use also varies by who is filing. In Santa Clara, the share of self-represented complaints showing signs of AI use rose from 19% in 2024 to 72% in 2026. Among attorney-filed complaints, the share rose from 5% to 34%. Use also depends on case type. Employment and business complaints in California show the highest rates. Debt-collection complaints are templated and make up nearly half of filings in some courts, yet they show almost no AI use. Within a filing, AI-associated text appears more often in the statement of facts and causes of action.

Measurement is hard, and the researchers say off-the-shelf detectors are unreliable for legal writing. Legal language uses a narrower vocabulary than everyday speech and leans on repeated boilerplate, so a detector has less signal to work with. Commercial training data may also have left out legal writing. The results are sensitive to settings: one AI detector flagged anywhere from 10% to 98% of the same set of 2026 Santa Clara complaints as AI-written, depending on how it was configured.

To work around this, the team examined filings where AI use is not in doubt. Damien Charlotin's AI Hallucinations Cases Database, established in April 2025, tracks cases in which judges accuse or admonish parties for citing fictitious authorities. The researchers combined those cases with a set of controls into a benchmark they call PhantomDocket. They also developed a calibration method called ScoreMix, which estimates the share of filings more similar to the hallucinated ones than to pre-ChatGPT writing. The authors describe the resulting figures as calibrated, robust and conservative. They also report that false positive and false negative rates remain nontrivial even for the best detectors.

Because of those error rates, the researchers say detection is most reliable for estimating how common AI use is across a population of filings, not for deciding whether a particular filing used AI. The study raises open questions about whether AI helps litigants who cannot afford attorneys, or whether it floods courts with inadequate lawyering. The authors point to more than 1,500 U.S. cases documented to include legal hallucinations. They argue for better disclosure policies across courts so that AI assistance in filings can be tracked more accurately in the future.

## Why it matters

Courts and firms that rely on AI detectors for professional documents may get wildly inconsistent results, so detector output alone is weak evidence about any single filing. The findings also suggest that AI-drafted material is rising fastest among self-represented litigants.

## What you can do

Do not rely on off-the-shelf AI detectors to judge individual legal or other professional documents; pair any detection with calibration and clear disclosure rules.

## Original source

- [AI Might Be Flooding State Court Houses, but Proving It Is Another Matter | Stanford HAI](https://hai.stanford.edu/news/ai-might-be-flooding-state-court-houses-but-proving-it-is-another-matter)

---

AI wrote this summary from the sources listed, and nobody reviewed it individually before publication. Check the original source before you rely on it.

Cite as: IKANGAI AI News, "Stanford study finds AI-assisted drafting in over a third of Santa Clara court complaints", 2026-11-19, https://ai-news.ikangai.com/story/stanford-study-finds-ai-assisted-drafting-in-over-a-third-of-santa-clara-court-c/
