AI Motion and Factum Drafting Software for Ontario Litigators

Disclosure: DROZlegal publishes this guide and builds the defence_drafting and motion_preparation agents, the litigation form registry, and the DOCX-to-PDF-to-CaseLines packaging pipeline described below, as part of its practice-automation product for Ontario law firms. The AI-citation and court-filing statistics cited are drawn from courtready.ca's tracking database, Cassels LLP's case commentary, Canadian HR Reporter, and the Ontario Superior Court of Justice's own published notice — not ours. This article is general information about Ontario litigation drafting workflow, not legal advice.

What "AI motion and factum drafting" actually means here

Two agents do the drafting: defence_drafting for a statement of defence or responding materials, and motion_preparation for a motion record. Both sit among DROZlegal's 22 registered agent classes, and both are manual-dispatch and propose-only — a lawyer starts the job, and the output lands as a draft, never as something already sent anywhere.

Neither agent works from a blank page. Each pulls from the matter's own parties, dates, and facts already sitting on the case file, then routes the draft through the same form registry, DOCX pipeline, and citation linker described below — defence_drafting to a Form 18A statement of defence, motion_preparation to a Form 37A notice of motion plus its supporting Form 4D affidavit. A factum runs through that identical DOCX-to-PDF pipeline too, generated directly from the registry rather than through either named agent.

The drafting-to-filing pipeline, step by step

Six stages sit between "an agent drafted something" and "a lawyer has a document to file," each backed by its own piece of the litigation module (backend/app/litigation/):

Every step before filing is mechanical or review-gated, never autonomous submission. Filing itself is a permanent hard ceiling — one of six actions across the whole platform that never gets automated no matter how much autonomy a firm has earned elsewhere, alongside trust money movement and settlement acceptance. A lawyer submits the document to the court; the software's job ends at a filing-ready PDF.

What's actually generatable in a 241-entry registry

The registry behind step two is built from ontariocourtforms.on.ca, 241 entries deep, and split into seven archetypes. Six of the seven are fully generatable, top to bottom.

The SPECIALIZED bucket is the outlier, and deliberately so. A NOTICE or an AFFIDAVIT follows a small number of predictable content shapes; a SPECIALIZED form doesn't, and each one needs its own hand-authored content schema before the system can generate it. Of 51 SPECIALIZED entries, five have that schema built today — the Costs Outline and the Rule 49 offer-to-settle forms. The other 46 stay browse-only — findable and readable in the registry, not yet auto-drafted.

Why the citation linker matters right now

Once a document is drafted, the DOCX pipeline's citation linker hyperlinks every case-law and legislative reference it contains out to CanLII or e-Laws — a live record a lawyer can open before signing anything. That isn't a promise the model checked its own citations against a database; it's a bridge to a source the lawyer can verify without leaving the draft. In 2026, that check is not optional.

Canadian courts and tribunals have now documented 247 decisions involving AI-hallucinated case citations, with 372 fictitious citations turning up across 206 of them. Source: courtready.ca AI-hallucination tracking database, updated August 26, 2026 — a legal-tech tracking resource, not a court or law-society publication.

Ontario has already produced two of the clearest cautionary examples this cycle. In Ko v. Li, 2025 ONSC 2766, Justice Myers found a factum containing citations that could not be verified through any legal database, writing that "AI is ubiquitous and yet its risks and weaknesses are not yet universally understood." Source: Cassels LLP, "Fantasies in the Footnotes," published May 23, 2025. In Mazaheri v. Law Society of Ontario, 2026 ONLSTH 112, a suspended lawyer's unsupervised use of generative AI to draft motion materials containing fabricated citations was treated as a significantly aggravating factor, resulting in a $31,150 costs order. Source: Canadian HR Reporter, published June 18, 2026.

Neither case involved a citation-linking step. That's the gap the feature is built to close: a citation that doesn't resolve to a real CanLII or e-Laws entry is visibly, immediately wrong in the draft — before it reaches a factum with a lawyer's name on it, not after a judge finds it.

The honest caveat: no LibreOffice, no PDF

Step four's PDF conversion runs on LibreOffice headless, and that dependency is real. If LibreOffice isn't installed on the host, the generate endpoint returns the drafted DOCX only and fails fast with a clear error, rather than a silent or broken PDF. It's a plain technical limit worth knowing before you lean on the pipeline for a same-day filing, not something to discover at 4 p.m. on a deadline.

Proven on one file, not just a demo

This pipeline isn't only a product-tour claim. A responding factum was produced end-to-end through this suite — drafted, converted, and packaged through the same steps above — on a live Ontario matter. We don't name the matter or the client; the point isn't the case, it's that the sequence ran start to finish under a real deadline, not just in a demo environment.

The same case-detail page that carries the generate step also carries the offers, undertakings, and service-records tabs covered in our guide to litigation discovery and settlement tracking — settlement acceptance sits behind its own permanent hard ceiling there, the same way filing does here. For a broader look at what AI document drafting means across a firm, not just litigation, see AI document drafting for law firms; for the full 241-form registry alongside the deadline engine that runs next to it, see Law Firm Automation in Ontario. The full agent roster, including defence_drafting and motion_preparation, is on the AI Agents page; the module itself lives on the Litigation product page.