The next AI filing problem may be a document full of allegations that never clearly identifies the case the court must decide.
An Arizona federal judge has dismissed a self-represented plaintiff’s 67-page employment complaint with permission to amend, ordering her to write the replacement personally and without AI assistance.
The October 7 filing in George v. Rollins, before U.S. District Judge Krissa Lanham, addresses a problem that extends beyond fabricated citations: AI-assisted drafting that produces volume without a coherent pleading.
The problem was the pleading’s structure
Shelly George’s complaint contained 199 paragraphs. The judge said it appeared AI-generated because it resembled other AI-generated complaints the court had encountered. That was an assessment of its apparent origin, rather than a definitive finding about the tools used.
More concretely, the court identified repetitive factual allegations, discussions of governing law, references to unattached exhibits and uncertainty about which claims George intended to advance.
The order described shifting references to retaliation, hostile work environment and a combined theory. Its concern was that extensive allegations were not clearly connected to particular causes of action.
The court applied Federal Rule of Civil Procedure 8’s requirement for a short and plain statement showing entitlement to relief. It characterized the document as an impermissible “shotgun pleading.”
The remedy requires personal drafting
George must submit an amended complaint of no more than 25 pages by October 19. She must personally draft the relevant allegations and identify the claims she wishes to pursue. AI drafting of that amended complaint is prohibited.
The order also imposes requirements on parties using generative AI in other filings in the case. They must provide a declaration disclosing its use, identify any incorporated AI output, and certify personal review and verification through independent judgment. Inaccurate or undeclared outputs may attract sanctions.
The dismissal did not resolve the underlying employment allegations. The order provides an opportunity to submit a compliant pleading, while directing entry of dismissal with prejudice if no amended complaint is filed by the deadline.
Why citation checking is insufficient
The order exposes a gap in approaches that treat legal AI verification primarily as an exercise in checking authorities.
Every cited case could exist, yet a filing could still fail to explain what happened, which claim arises and why the requested remedy follows.
Pleading requires selection. Someone must decide which facts matter, separate distinct claims and remove material that obscures the dispute. More generated text can increase the burden of those decisions.
For Canadian litigators, the order is a comparative example rather than binding guidance. It shows how a court can combine ordinary pleading requirements with a tailored restriction on AI drafting.
A better review process
Lawyers reviewing AI-assisted pleadings should ask four questions:
- Can each claim be identified without reconstructing it from scattered paragraphs?
- Are the material facts connected to that claim?
- Are exhibits and factual references accurate and available?
- Does each paragraph help define the dispute?
Disclosure and verification requirements must also be checked against the rules and orders governing the particular proceeding.
The Arizona restriction is specific to this amended complaint. It should not be described as a general prohibition on AI in litigation.
Its broader lesson concerns professional judgment: producing a document and formulating a case are separate tasks.
A pleading earns its length by making the dispute clearer.
Canadian courts are confronting related drafting problems. See our coverage of poor AI filings and access to justice in B.C. and AI errors and junior-lawyer training.