Drafting Motions and Briefs Without Hallucinations

Brief drafting is where AI most directly intersects with the work product courts actually read — and most directly intersects with the sanctions risk that has dominated the legal AI news cycle through 2025 and into 2026. US courts imposed more than $145,000 in AI hallucination sanctions in the first quarter of 2026 alone, including Oregon's $110,000 penalty and Nebraska's first license suspension. The lesson is not to avoid AI for briefs. The lesson is to use it correctly.

This lesson focuses on motions and briefs at the trial court and appellate levels. We will cover the full workflow from outlining through final cite-check, with explicit defenses against hallucinations.

What You'll Learn

• A four-phase drafting workflow that combines AI speed with human accuracy
• How to prompt for argument structure rather than fake authority
• How to use Clearbrief, Spellbook, and Briefpoint for verified drafting
• A pre-filing checklist that catches the failures that lead to sanctions

Why Hallucinations Happen

Hallucinations happen because foundation models predict plausible next tokens, not facts. When a model does not have a real authority for a proposition, it produces something that sounds like one. The fix is not "tell the model not to hallucinate." Models do not have stable self-control over hallucination. The fix is to architect the workflow so the model never has the chance to invent authority unchecked.

Three architectural changes make this work:

1. Use grounded tools. Westlaw Precision AI and Lexis+ AI with Protégé retrieve citations from real databases. They are not perfect, but they hallucinate far less than ungrounded chatbots.
2. Use citation verification tools. Clearbrief, in particular, verifies every citation against the actual record and against the actual cited source. This is a separate step from drafting.
3. Use a human verifier. No exceptions, ever, until the verification step is checked off by a human.

The Four-Phase Drafting Workflow

This is the workflow used by competent AI-leveraged litigators in 2026.

Phase 1 — Outline. Start in your grounded research tool. Ask for the legal framework, the elements of each claim or defense, and the leading authorities in your jurisdiction. The output is a research memo, not a brief.

Sample prompt:

For a motion to dismiss a {claim_type} claim in the {jurisdiction},
produce the legal framework including:
- Pleading standard
- Elements of the claim
- Leading recent cases in this jurisdiction
- The most-cited adverse authority
Provide citations only from primary law or treatises.

Replace \{claim_type\} and \{jurisdiction\} with specifics. The output is your skeleton.

Phase 2 — Argument structure. Move to a drafting tool like Spellbook (for transactional-flavored work) or a direct LLM in Tier B with no need for invented authority. Ask for an argument structure for each section. Specify: do not invent citations; flag every place where a citation is needed with [CITATION NEEDED].

This produces an outline with prose but explicit citation placeholders.

Phase 3 — Fill in citations. Return to your grounded tool. For each [CITATION NEEDED] placeholder, do a focused research query. Insert the verified citation into the draft.

Phase 4 — Polish, verify, and finalize. Use Clearbrief (or equivalent) to:

• Verify every citation exists
• Verify every quotation is verbatim from the source
• Verify the record citations match the actual record
• Check the proposition matches the cited section

Then a human attorney reads the entire brief and signs.

Working with Spellbook, Clearbrief, and Briefpoint

These three tools serve different stages.

Spellbook integrates into Microsoft Word. Best for drafting clauses, contract sections, and structured argument blocks. Strong for transactional documents; useful for litigators primarily in drafting discrete sections of briefs (e.g., factual recitations, standard-of-review sections).

Clearbrief integrates into Word as well. Its core feature is citation verification. Drop a draft in, and it will compare every citation to the underlying source and flag mismatches, broken quotes, and missing pinpoint cites. Topped Nevada State Bar's AI work group rankings with 40.5/50 in 2025. The most important AI tool you can adopt if you file briefs.

Briefpoint specializes in discovery — requests for production, interrogatories, and responses. It can reduce drafting time on those documents by roughly 87%, according to the company's own metrics, and the workflow is genuinely faster in practice. For litigators with heavy discovery loads, Briefpoint pays for itself in a single matter.

In a typical 2026 brief-drafting matter, you might use all three: Spellbook for sections, Clearbrief for citation verification, and Briefpoint for the underlying discovery responses that fuel the factual record.

Prompting for Structure, Not Authority

The fastest way to get a hallucinated citation is to ask for one. Avoid prompts like:

Write me a paragraph arguing that the contract was a unilateral
contract, citing two New York cases.

The model will produce a paragraph with two plausible-looking citations, at least one of which is likely to be invented. The correct version splits the task:

Step 1 — argument structure only:

Draft a paragraph arguing that the contract at issue was a
unilateral contract. Use [CITATION NEEDED] wherever a New York
case citation would support a proposition. Do not invent citations.

Step 2 — go to your grounded research tool and find the cases. Step 3 — insert verified citations.

This is slower than a one-shot prompt but produces a real, defensible draft.

The Pre-Filing Checklist

The single most important habit you can build is a pre-filing checklist that you actually use. A version that has prevented countless sanctions in 2026:

1. Citation existence. Every cited case has been opened in Westlaw, Lexis, or Bloomberg Law. None of them came from "I asked the AI for cases on X."
2. Holding match. For every cited case, the proposition you cite for matches the actual holding or reasoning in the cited section.
3. Quotation verification. Every direct quote has been verified against the source. Quotation marks come with line numbers or pin cites.
4. Record cites. Every record cite resolves to the actual record. Page numbers match.
5. Good law. Every case has been Shepardized or KeyCited; any negative treatment is acknowledged.
6. Signed-by-human. A human attorney has read the entire brief and signs the filing.

The Stanford research finding that Westlaw AI hallucinates 33% and Lexis+ AI 17% of the time is what makes step 1 non-negotiable — even grounded tools produce wrong citations often enough that one-shot trust is malpractice in 2026.

Appellate Briefs

Appellate briefs deserve special caution. The record on appeal is fixed; record citations are heavily scrutinized; standard of review is everything. Use AI to:

• Generate the statement of facts from the trial record
• Identify standard of review for each issue
• Surface every case in the jurisdiction on the standard of review and the substantive law
• Draft cross-references and tables of authorities

Do not use AI to draft the argument structure of an appellate brief without significant human re-architecture. Appellate strategy involves a tight ordering of points and concession choices that AI is not yet good at.

Key Takeaways

• Hallucinations are an architectural problem; solve them with grounded tools, citation verifiers, and human review.
• Split brief drafting into four phases: outline, argument structure, citation fill-in, and polish-and-verify.
• Spellbook, Clearbrief, and Briefpoint solve different parts of the workflow — use them together.
• Never ask an ungrounded LLM for citations. Ask for structure with [CITATION NEEDED] markers.
• Use a pre-filing checklist every time. The 2026 sanctions wave hit lawyers who skipped this step, not lawyers who used it.