Two 2026 working papers found the same pattern in federal court records: the share of civil cases with a self-represented plaintiff climbed from 11.33% before generative AI tools were widely available to 16.94% after, and docket activity per self-represented case rose 38% in the first 180 days. The first of those papers also found the AI-drafted filings are no more likely to win. More filings, same outcomes, is the part that matters for your budget.
What the data shows
A preprint posted May 28, 2026 by Or Cohen-Sasson examined about 2.8 million federal civil filings from fiscal 2008 through 2025 and found the pro se plaintiff rate rose from 11.33% to 16.94%, a 5.61-point increase. It flagged 13.9% of non-form complaints in the post-AI period as showing AI-consistent drafting patterns.
A separate working paper by Anand V. Shah of MIT and Joshua Y. Levy of USC, summarized by ISBA Mutual in June 2026, reviewed more than 4.5 million federal civil cases filed between 2005 and 2026 and landed in nearly the same place: a long-steady 11% share rising to 16.8% in fiscal 2025. That study also measured the workload. Docket activity per self-represented case in the first 180 days rose 38%, total docket entries per court from those cases rose 158% by mid-2025, and entries in lawyer-represented cases rose 23%. Its AI-detection rate went from one document in 800 between 2019 and 2022 to 18% by early 2026.
Appeals point the same direction. Self-represented litigants filed 20,878 appeals in fiscal 2025, up 9% and half of all new cases in the regional courts of appeals.
Two caveats before you act on any of it. Both papers are working papers and have not been peer reviewed. And they count federal civil cases, while most landlord-tenant disputes are heard in state housing and small-claims courts, where comparable data is thin. What the federal numbers do capture is the category where a landlord carries federal exposure: Cohen-Sasson found the increases most pronounced in civil rights and other statutory cases, the bucket that holds Fair Housing, credit-reporting and debt-collection claims.
Volume without merit still costs money
The Cohen-Sasson paper reports that complaints flagged as AI-drafted showed no improvement in win rates. They were more likely to be dismissed and to terminate at earlier procedural phases.
That is the number to sit with. The filings are not getting better; there are simply more of them, and each one moves more paper. Benjamin Young, a litigation attorney at Williams Teusink, wrote in Multifamily Dive on September 15 that AI does not turn a self-represented tenant into a skilled attorney, but it "can make it easier for them to generate enough legal activity" to create delay and consume resources. He described one tenant filing 22 separate pleadings between a Friday and a Sunday, most of them duplicative.
A motion you win still costs you counsel hours, a continuance and a unit that is not turning. That cost lands whether or not the filing had any merit.
The errors run both ways
This is not a one-sided risk, and treating it that way is how operators get hurt.
MassLandlords reported in April 2026 that ChatGPT-drafted notices frequently carry missing or improper language, giving the example of a notice served mid-month telling a tenant they have 30 days to vacate, which does not square with Massachusetts timing rules. Peter Shapiro, a mediator counselor quoted in that piece, said judges sometimes require the landlord to start over from square one. Notice periods and required language vary by state; check your own statute rather than assuming the example applies.
Court filings are not safer. In March 2026, a Connecticut firm filed eviction documents that Yale law students found contained citations that were erroneous or computer-generated fiction. That case was still pending when MassLandlords published, and no finding against the firm had been reported.
Courts have started attaching money to the problem. At least $145,000 in sanctions for AI-generated fake citations was imposed in the first quarter of 2026, including $109,700 across Oregon matters and $30,000 in the Sixth Circuit's Whiting v. City of Athens, split evenly between two attorneys. Those penalties fell on licensed attorneys. That reporting does not describe fines levied against self-represented parties themselves, so do not assume a court will tax an opposing tenant for a bad filing.
What to do this week
This is general information, not legal advice. Notice rules, sanctions practice and eviction procedure vary by state; confirm anything here with local counsel before you change a process.
- Rebuild your notice templates from the statute. Pull your state's actual notice requirements and compare them line by line against whatever your team is sending. Pay attention to the timing math, which is where the Massachusetts example broke.
- Ask your eviction counsel, in writing, how they verify citations in anything AI helped draft. You are the client whose case gets dismissed and whose name appears in the docket.
- Tighten the written record before a dispute starts. Keep maintenance tickets, notices, ledgers and communications in one place with dates intact. Volume litigation is won on documentation, not argument.
- Tell on-site teams what unusual looks like. Repeated motions, duplicate filings, or a routine nonpayment case that suddenly generates a stack of paper should route to counsel immediately, not after the third filing.
- Price the fight against the alternative. Young's argument is that a negotiated move-out can beat a win that takes four months. Run that math per case rather than defaulting to litigation.
- Do not screen applicants on litigation history as a reaction to this. Using prior court involvement as a rental criterion raises Fair Housing and retaliation exposure and is restricted in some jurisdictions. Talk to counsel before touching screening criteria.
What we're watching
- Whether the two working papers survive peer review. They agree closely, which is encouraging, but neither has been through it yet.
- State housing court data. The federal record is a proxy. The courts where most of this actually lands are not publishing comparable numbers.
- Whether sanctions reach self-represented filers. The Q1 2026 penalties hit attorneys. If courts extend that, the calculus on defending thin filings changes.
- Detection rates. A jump from one flagged document in 800 to 18% in roughly four years suggests this has not plateaued.