Motion to Set Aside Copilot

What It Does

When a landlord files an eviction lawsuit in California, the tenant has 10 court days to respond. Thousands miss this window: improper service, hospitalization, language barriers, or simply not knowing. The court enters a default judgment. A sheriff posts a Notice to Vacate. For many tenants, this is the first moment they learn the lawsuit exists.

The tenant's only recourse is a Motion to Set Aside (MSA) the default judgment. Filing one requires reserving court dates, preparing a multi-document legal packet (Notice of Motion, Points and Authorities, a sworn personal Declaration, a Proposed Order, Proposed Answer, fee waivers), serving the landlord's attorney, and making a compelling case to the judge. The current fully-human workflow at legal aids takes 4–5 hours per tenant: 45–60 minutes for pre-screening, 30–60 minutes for the attorney interview, roughly 2 hours for paralegal drafting, and an hour for attorney review. Because of this, groups can only assist a small number of people who want help with an MSA. Everyone else is turned away.

The MSA Copilot is designed for those tenants: the ones legal aid cannot help. It is a public-facing, self-help tool (not an attorney-facing internal tool) that guides an unrepresented tenant (perhaps with a navigator or volunteer) through the full MSA process in five phases.

Phase-by-Phase Walkthrough

Phase 1: Intake. A four-step guided form collects personal information, case details, key dates, and what happened with service. The system captures how the tenant was served and why they didn't respond. These inputs become the foundation for all downstream document generation: a single source of truth populated once and carried through every phase.

Phase 2: Overview and Urgency Assessment. The system explains what a Motion to Set Aside does in plain language, then generates dynamic warnings based on the tenant's data: statutory deadline alerts (the six-month limit), imminent lockout risks, and time-sensitive procedural flags. This phase orients the user before they start gathering materials.

Phase 3: Evidence Gathering. A checklist of required court documents (summons, default judgment, proof of service) is paired with upload functionality. Evidence prompts are dynamically tailored to the tenant's situation: improper service, medical emergency, disability, minor child receiving papers. Uploaded files are automatically labeled as exhibits and annotated with user-provided descriptions. These exhibits flow into the final packet.

Phase 4: Document Preparation. Three components. First, a structured review where the tenant confirms their case details with validation indicators. Second, an AI-guided declaration drafting interview: a chatbot conducts targeted questioning to elicit the sworn personal narrative that makes or breaks the motion, then generates a court-ready declaration. Third, download of a complete document set: the motion, declaration, proposed answer, proposed order, notice documents, fee waivers (including FW-003), and a compiled exhibit index. All documents follow California pleading standards (28-line numbered paper).

Phase 5: Filing and Court Preparation. Step-by-step procedural guidance through a slideshow-style interface covering hearing scheduling, document preparation, service requirements, filing instructions, and what to expect at the court appearance. Includes deadline tracking, San Diego County-specific instructions, outcome tracking, and optional AI assistance for hearing preparation and responding to landlord opposition. (Built but not yet tested with users.)

The Declaration Problem

The team's key insight: the declaration is the hardest and most valuable problem to solve. Most MSA documents are standard legal forms that a smart form-filler can handle. The declaration is different. It's a sworn personal narrative, and getting it right requires eliciting a specific kind of story from a stressed, non-legally-trained person: why they didn't respond to the lawsuit, what their circumstances were, and what defense they would have raised. This is legally thorny and emotionally sensitive.

The AI operates as what the team calls a "stubborn" interviewer. It asks targeted questions, avoids repetition, and focuses on required legal elements (grounds for relief, evidence of diligence, meritorious defense). It will not generate a declaration until it has collected sufficient information, ensuring outputs are complete and grounded in user-provided facts rather than AI invention.

Design Decisions Worth Noting

Public-facing, not attorney-facing. The tool is designed for a tenant with no legal training, in crisis, under time pressure. Every screen uses plain language, large text, mobile-friendly layouts, and a calming visual design. Legal terms are simplified or explained inline. The phase-based workflow shows only what's needed at each stage. It could be used with a navigator, paralegal, or even a lawyer -- but should still be friendly to someone who is trying to understand things on their own.

UPL boundaries as a structural constraint. The tool provides information, not advice. It never tells a tenant they "qualify" or have a "strong case." Language uses "if you choose to" framing and presents consequences of options rather than recommending actions. The LASSD partner reviewed all language for unauthorized practice of law compliance. This boundary shaped the entire design.

AI verification step. A cross-check compares the generated declaration against user-provided facts before download, catching hallucinations before they reach the courthouse.

Jurisdiction-specific but adaptable. The tool is scoped to San Diego County procedures (SDSC forms, hearing reservation systems, filing locations). When a user enters a different county, the system displays a warning. A Los Angeles County practitioner reviewed the prototype and found the structure, logic, and UX strong enough to adapt without fundamental redesign, suggesting the pattern can travel across California counties with targeted procedural updates.

What They Tested

The team conducted four structured expert reviews:

Technical review (three reviewers, February 2026) praised the chatbot declaration drafting and exhibit management. Flagged missing input validation, unintuitive legal jargon, and no document verification step for hallucination detection.

Legal aid practitioner review confirmed the declaration drafting was "really on point." Identified critical procedural corrections: "Motion" needed to be "Notice of Motion," a Memorandum of Points and Authorities was missing, the Ex Parte flow was being shown to all users (only ~50% of MSA filers face imminent lockout), and the fee waiver section needed the FW-003 court order form.

Second legal review by another expert provided screen-by-screen feedback on precision: evidence should support "your motion" not "your case," court document names needed verification against what San Diego County actually uses, and Judicial Council form numbers should be included wherever applicable.

Cross-county legal review by another housing attorney evaluated from a Los Angeles County perspective and found the platform user-friendly, readable, and far less intimidating than paper-based MSA packets. Recommended clearer urgency communication, better separation of motion evidence from underlying defense evidence, and tailored prompts for common grounds (disability, hospitalization, mail delivery problems).

The team also created a synthetic test dataset of fictional user personas but notes it requires further validation before pilot deployment.

What Didn't Work (Their Own Account)

Phase 5 is built but untested. PII handling has no formal policy yet. County-specific procedural differences were deeper than anticipated. The California pleading paper formatting (evenly spaced 28-line numbered pages) is still being refined. And critically: no testing with actual self-represented tenants has been conducted. All reviews are from legal professionals and peers.

Risks and Limitations

AI hallucination in legal documents is the most critical risk. A fabricated fact in a sworn declaration could cause court rejection or worse. The verification step mitigates but doesn't eliminate this.

UPL exposure if language drifts toward advice. The team maintains continuous review with their legal partner.

Wrong jurisdiction instructions if the tool is used outside San Diego without adaptation.

PII exposure from sensitive personal and case information collected during intake. Formal privacy framework required before any public deployment.

No real-user testing yet. The tool is designed for tenants in crisis with limited legal and technology literacy. Expert validation is necessary but not sufficient.

What's Next

Move toward a Pilot with a legal aid team: 5–10 supervised sessions with tenants at the law library or courthouse, tracking document completeness, filing success rates, time to completion, and user confidence.

Privacy framework before any public-facing deployment.

Cross-county adaptation starting with Los Angeles County, mapping procedural variations and building jurisdictional logic.

Phase 5 testing for filing logistics and hearing preparation.

Accessibility testing for limited English proficiency and limited technology experience.