We respect attorney-client confidentiality. No tracking pixels in our emails.
From research to cite-checking, discover how AI fits every stage of motion practice—with prompt templates, tool picks, and real time-savings benchmarks for litigators.
2026/07/31
In March 2025, a federal district judge in the Northern District of California sanctioned a law firm $15,000 after its motion to dismiss cited three cases that did not exist—hallucinations generated by an AI research tool that no attorney had verified before filing. The firm's mistake was not using AI. The mistake was skipping the verification step that experienced litigators know is non-negotiable.
That case crystallized a reality every litigator now faces: AI tools have become genuinely useful across the motion lifecycle, but only when integrated into a disciplined workflow. Partners at AmLaw 100 firms report cutting first-draft time by 40–60% on standard motions. Associates at boutique litigation shops describe finishing opposition briefs over a single weekend that once required two weeks. The productivity gains are real—and so are the risks when process breaks down.
This guide walks through the complete motion practice workflow, stage by stage, with specific tool recommendations, prompt templates, and the quality-control checkpoints that separate firms winning sanctions from those winning on the merits.
Motion practice has always been labor-intensive. A fully briefed summary judgment motion in a complex commercial case can involve weeks of case law research, dozens of hours of record review, multiple drafts, and meticulous cite-checking before a single page reaches the court. For associates, motions are the crucible of legal training. For partners, they are often the most billable—and most scrutinized—work product the firm produces.
The arrival of large language models in legal practice beginning in 2023 created immediate pressure to apply AI to briefing. Early experiments were mixed. Tools trained on general web data hallucinated citations at alarming rates. Firms that rushed AI into production without process controls faced embarrassing corrections and, in some cases, sanctions.
By 2025, a second generation of legal-specific AI tools had matured significantly. Harvey AI and CoCounsel built retrieval-augmented systems that ground responses in actual case databases rather than generating citations from model weights alone. Westlaw Precision integrated AI research assistants directly into its verified database, dramatically reducing hallucination risk. Briefpoint emerged as a specialized tool for brief drafting that pulls from actual record documents.
The lesson from 2023–2025 is that AI in motion practice is not a single tool—it is a workflow integration problem. The firms extracting the most value have mapped AI assistance to specific stages of the motion lifecycle and built quality-control checkpoints between each stage.
Understanding prompt engineering has become a practical skill for litigators, not just technologists. How you instruct an AI model determines whether you get a useful research memo or a plausible-sounding fiction. The gap between a good prompt and a poor one is routinely the difference between two hours of useful work and three hours cleaning up hallucinations.
This guide is structured around five stages: research, outlining, drafting, cite-checking, and revision. Each stage has specific tool recommendations, prompt templates calibrated to that stage's output requirements, and the quality-control steps that protect both the work product and the attorney's professional responsibility.
AI research tools have become genuinely reliable when used within verified databases. Westlaw Precision and CoCounsel both operate as retrieval-augmented generation (RAG) systems—they retrieve actual cases from verified databases before generating summaries, which dramatically reduces fabrication risk compared to general-purpose LLMs.
Effective research prompts are specific about jurisdiction, standard, and posture. Compare these two prompts:
Weak: "Find cases about breach of contract damages."
Strong: "In the Ninth Circuit, what is the standard for awarding lost profits as consequential damages in a commercial lease breach case where the plaintiff failed to mitigate? Provide cases addressing both the standard and the mitigation analysis, with holdings and relevant quotes."
The second prompt specifies circuit, damages type, factual context, and the specific legal issue. It requests holdings and quotes—output that can be directly verified against the source. Always request verbatim quotes from AI research tools; paraphrased holdings are harder to verify and more prone to distortion.
For opposition research, Trellis and Westlaw Precision allow you to pull an opposing counsel's prior filings and brief history. Prompt: "Summarize the arguments opposing counsel made in their prior motions to dismiss in [case type] matters. Identify any positions that are inconsistent with the arguments they are making in this case." This is one of AI's highest-value uses in motion practice—it surfaces impeachment material against opposing counsel's legal positions that manual research would rarely catch.
After research, outlining is the stage where attorney judgment is most critical. AI can suggest structures, but the strategic choices about which arguments to lead with, which to include, and which to omit belong to the attorney.
Useful outlining prompt: "I am drafting a motion for summary judgment in a Title VII retaliation case. The defendant's strongest argument is temporal proximity. The plaintiff's strongest response is a three-month chain of documented complaints. Draft a three-section argument outline with the strongest order of presentation, including the key sub-issues under each heading."
Review the outline critically before proceeding to drafting. AI tends toward exhaustive outlines that include every conceivable sub-argument. Part of attorney judgment is deciding what to exclude—the arguments that dilute rather than strengthen.
Draft with AI by feeding it your research memo and outline, not by asking it to research and draft simultaneously. The two-step process produces substantially better output and maintains clearer attorney control over the research foundation.
Prompt template for argument sections: "Draft the argument section for [heading] using only the following cases and authorities: [paste research memo]. The key facts supporting this argument are: [paste relevant facts from record]. Use formal brief style. Do not add citations not included in the materials I have provided. Target length: 500–700 words."
The instruction to use only provided authorities is critical. It prevents the model from supplementing your research with hallucinated cases.
Briefpoint is purpose-built for this stage—it allows you to upload record documents and drafts argument sections grounded in your actual record, reducing the hallucination risk that comes with open-ended drafting prompts.
Cite-checking AI-generated drafts is non-delegable and mandatory. Every citation must be verified against the primary source. This is where the sanctions cases arise—not from using AI, but from filing without verification.
Workflow: Export the AI draft. Run citations through Westlaw Precision or Casetext to confirm the case exists, the holding is accurate, and the quoted language appears verbatim. Flag any case where the AI's characterization of the holding differs even slightly from your read of the actual opinion.
CoCounsel includes a built-in cite-checking module that automates much of this workflow. It verifies citations against its database and flags discrepancies. Even so, attorney review of flagged items is required—automated cite-checking reduces risk, it does not eliminate it.
AI is particularly valuable for local rules compliance checking—an area where many firms still rely entirely on human memory and experience. Prompt: "Review this motion for compliance with Local Civil Rule 7.1 of the Southern District of New York, including word limits, formatting requirements, and required certifications. Identify any non-compliance."
Harvey AI has built local rules databases that allow it to check briefs against specific court requirements. This is low-hallucination-risk because local rules are discrete, checkable requirements—not legal analysis.
A mid-size litigation boutique handling a commercial lease dispute used the following workflow for a motion for summary judgment:
Day 1 – Research: The associate used CoCounsel to pull cases on the specific damages standard, then ran Westlaw Precision searches on each case to verify holdings and identify subsequent history. Research memo completed in 4 hours versus an estimated 8–10 hours manually.
Day 2 – Outline: Attorney reviewed the research memo, identified three core argument sections, and used Harvey AI to draft an outline. The attorney revised the outline substantially—AI had included two weak sub-arguments that were cut.
Days 3–4 – Drafting: Associate used the prompt template above to draft each argument section, feeding AI the relevant portion of the research memo and key record citations. First complete draft generated in 6 hours.
Day 5 – Cite-check: Every citation in the 28-page brief verified against source. Three cases required correction—two had inaccurate pinpoint citations, one had a characterization that required revision.
Day 6 – Revision: Partner revised for voice, strategy, and argument order. Local rules compliance check run via Harvey AI. Final brief filed same day.
Total time: approximately 30 attorney hours for a 28-page summary judgment motion. The firm's prior average for comparable motions was 55–65 hours.
CoCounsel – Best-in-class for research and cite-checking within a verified database. RAG architecture substantially reduces hallucination risk. Compare CoCounsel vs Harvey AI for your firm's specific use case.
Westlaw Precision – Integrated AI research within the industry-standard legal database. Strongest for citation verification and subsequent history.
Harvey AI – Strong for drafting, local rules compliance, and opposition research. Requires firm-level deployment with appropriate data handling agreements.
Briefpoint – Specialized brief drafting tool that grounds output in uploaded record documents. Reduces hallucination risk on factual sections.
Casetext – Solid research and cite-checking alternative; compare Casetext vs CoCounsel for research-focused use cases.
Q: If I use AI to draft a brief section, can I bill the full time I would have spent drafting manually?
A: No—most state bar ethics opinions require billing only for time actually spent. If AI reduced a 5-hour drafting task to 2 hours, bill 2 hours. Some firms have developed value-based billing arrangements that address AI efficiency gains directly.
Q: How do I handle a situation where AI-generated research conflicts with my own research?
A: Treat it as any conflicting authority—investigate the discrepancy. AI tools can surface cases you missed; they can also mischaracterize cases. Never default to AI's characterization over your own careful reading of the source.
Q: Which AI tools are safe to use with confidential client information?
A: Tools with zero-data-retention agreements and documented data handling policies—Harvey AI, CoCounsel, and Westlaw Precision all offer enterprise agreements with appropriate confidentiality protections. Review your firm's approved vendor list before inputting client data.
Q: Should associates or partners be running the AI tools in motion practice?
A: Both, at different stages. Associates are well-suited for research and first-draft stages with appropriate supervision. Partners should review AI output critically, revise for strategy and voice, and personally verify any argument section before filing.
Q: What do I do if opposing counsel uses AI-generated arguments and I suspect hallucination?
A: Verify every case they cite in their brief against primary sources. If you find fabricated or mischaracterized citations, raise it with the court promptly—most jurisdictions now have clear obligations to address opposing counsel's citation errors.
AI has made every stage of motion practice faster, but it has not made any stage less demanding of attorney judgment. The firms succeeding with AI in motion practice have not simply handed briefs to AI—they have built structured workflows that place AI assistance at specific stages while preserving the quality-control checkpoints that professional responsibility requires.
The five-stage workflow—research, outlining, drafting, cite-checking, revision—provides a reliable framework. The non-negotiable rule is that cite-checking is always done by a human attorney against primary sources. The practical upside is that the same attorney can handle more matters, produce higher-quality first drafts, and spend more time on the strategic work that clients actually pay for.
Start with the stages where AI risk is lowest: local rules compliance and research summarization within verified databases. Build confidence in those stages before extending AI use to open-ended argument drafting. And maintain a prompt engineering log—your best prompts from one motion are assets for the next.
Understanding AI hallucination risk is foundational. It does not go away with better tools—it is managed through workflow discipline.
This article reflects independent editorial analysis. LawyerAI does not accept payment for editorial coverage. Tool scores are based on methodology described in Our 5-Dimension Methodology. Last reviewed: 2026-07-31.