A Stanford HAI study found AI-assisted drafting in up to 72% of self-represented litigant filings in California courts — up from near zero in 2022.
A Stanford HAI study of Santa Clara County courts found that AI-assisted drafting now appears in up to 72% of documents filed by self-represented litigants — up from near zero in 2022. The findings are intensifying a national debate courts have been slow to address: should filings drafted with AI assistance require mandatory disclosure?
From Near Zero to Majority Practice in Four Years
For most of legal history, courts could safely assume that professional-looking filings came from lawyers. Self-represented litigants — people filing without an attorney, often in landlord-tenant disputes, small claims, or family court — historically produced rough, procedurally imperfect documents that were easy to identify.
AI changed that. Tools like ChatGPT, Claude, and legal-specific AI platforms now let anyone generate professional-looking complaints, motions, and responses in minutes. The Stanford study documents how thoroughly this shift has taken hold.
Analyzing filings in Santa Clara County across multiple case types from 2022 through 2026, the researchers found:
- AI-drafting indicators near zero in 2022 among all filer categories
- Over one-third of all complaints showed AI-drafting markers by 2026
- Up to 72% of self-represented litigant filings in certain case categories show AI-assisted drafting
- Nearly 900 US court rulings now reference AI-flagged filings
The researchers used linguistic markers, structural patterns, and metadata signals to detect AI-drafted content — though they acknowledge that improving AI output quality will make identification progressively harder over time.
The Error Problem Has Changed — and That's the Real Issue
Early AI legal drafting produced obvious errors: hallucinated case citations, non-existent statutes, wrong courts. Judges and clerks learned to flag these quickly. The Stanford study documents a more serious shift: a new class of subtle errors that look like legitimate filings.
These include:
Get this in your inbox.
Daily AI intelligence. Free. No spam.
- Jurisdiction mismatches — professionally formatted motions filed in courts that lack authority over the dispute
- Procedural timing errors — proper-looking filings submitted outside statutes of limitations
- Authority errors — real cases cited whose holdings don't apply to the jurisdiction or the facts at hand
These errors require substantive review to catch — the kind courts often can't provide for the volume of pro se filers they handle on crowded dockets.
"The challenge isn't that AI-assisted filings are automatically bad," the study notes. "It's that the error patterns have changed in ways that our systems aren't designed to catch."
Access Expanded, But the Tradeoff Is Real
The Stanford team is careful not to frame this as pure threat. AI has genuinely expanded legal access for populations that historically had almost none: low-income renters fighting eviction, workers filing wage claims, individuals navigating administrative processes without counsel. For these groups, AI-assisted drafting represents a meaningful improvement in their ability to participate in legal proceedings.
The problem is that the same tool expanding access is also introducing a category of errors that is hard to detect at scale. When a self-represented plaintiff files a 20-page complaint that reads like attorney work, the opposing side must respond substantively — even if the underlying claims are flawed in ways neither party initially recognizes.
That imposes real costs across the system: on opposing parties, on courts, and ultimately on the litigants themselves when their professionally-looking AI-drafted documents don't survive scrutiny.
Courts Are Responding Ad Hoc
The nearly 900 court rulings now referencing AI-flagged filings reflect a system doing triage without a playbook. Some judges have issued standing orders requiring disclosure of AI use. Others are ruling on motions to strike AI-assisted filings. No consistent national standard exists for how courts should handle, evaluate, or respond to AI-drafted documents.
The burden falls unevenly. Clerks and judges handling high-volume pro se dockets must decide how much additional scrutiny to apply to filings that look professional. Opposing parties must decide how aggressively to challenge documents they suspect are AI-generated. The result is inconsistency that serves nobody well.
What to Watch
The federal judiciary's Advisory Committee on Civil Rules is expected to take up AI disclosure standards in early 2027. New York, Texas, and Illinois have pending legislation requiring disclosure of AI use in legal filings. How California responds — given the Stanford study was conducted there — will likely set the template for other states.
The core policy question isn't whether to allow AI in legal drafting. That ship has sailed. The question is whether mandatory disclosure can give courts enough information to apply appropriate scrutiny, without penalizing litigants for using the tools that are now available to everyone.
What was near zero in 2022 is now a majority practice among certain filer types. The legal system didn't see it coming fast enough. Catching up is now the work.
Did this help you understand AI better?
Your feedback helps us write more useful content.
Get tomorrow's AI briefing
Join readers who start their day with NexChron. Free, daily, no spam.