We respect attorney-client confidentiality. No tracking pixels in our emails.
Legal research prompting is not like general AI prompting. Jurisdiction anchoring, issue framing, and adverse authority requests change the quality of AI research outputs dramatically. Here are 20 copy-paste templates.
2026/07/21
A bankruptcy attorney in Chicago asked a general-purpose AI to summarize the automatic stay provisions under 11 U.S.C. § 362. The AI returned a confident, well-organized summary — citing a Seventh Circuit case that does not exist, a law review article that was never published, and a Treasury regulation that was repealed in 2019. The attorney caught the errors before filing. A colleague in the same firm, working under deadline, did not catch a similar set of phantom citations in a different brief and had to file an embarrassing correction with the court.
The problem was not that the AI hallucinated — all current legal AI tools do sometimes. The problem was that the prompt gave the AI no constraint framework, no jurisdiction anchor, and no instruction to flag uncertainty. A better-structured prompt would have produced a better output and flagged the confidence level of each citation.
This guide explains why legal research prompting is structurally different from general-purpose AI prompting and provides 20 copy-paste templates organized by research task.
Legal AI research tools operate on a spectrum. On one end are general-purpose large language models (GPT-4, Claude) with no legal database grounding. On the other end are retrieval-augmented legal research tools (CoCounsel, Vincent AI, Westlaw Precision) where the AI's outputs are grounded in a specific legal corpus and citations link to verifiable source documents.
The prompt engineering techniques in this guide apply across the spectrum but serve different purposes. For grounded tools, good prompting improves research quality and relevance. For ungrounded general-purpose LLMs, good prompting is the primary safeguard against dangerously confident hallucinated output.
Legal research has unique requirements that distinguish it from general information retrieval:
Jurisdiction specificity: A correct answer in California may be wrong in Texas. A correct answer in the Ninth Circuit may be wrong in the Fifth. Every legal research prompt should anchor to a specific jurisdiction.
Citation format matters: When you ask for case law, you need citations in a format you can verify. An AI that says "courts have generally held" without citations is producing legal writing, not legal research.
Temporal relevance: Law changes. A prompt that does not specify a recency requirement may return overruled precedent. A prompt asking for "current" law should specify a review date.
Procedural posture: The standard of review changes the relevant authority. The same factual issue analyzed at 12(b)(6) versus summary judgment implicates different bodies of case law.
Adverse authority obligation: Under MRPC 3.3, attorneys have a duty to disclose directly adverse controlling authority. A research prompt that only asks for supporting authority produces one-sided research with potential ethics exposure.
Every legal research prompt should identify the jurisdiction at the outset. Jurisdiction anchoring does three things: it focuses the AI on the relevant body of law, it reduces the chance of receiving out-of-jurisdiction authority presented as controlling, and it gives you a verification anchor (you know which Westlaw database to check).
Weak: "What are the elements of tortious interference with contract?"
Strong: "Under Texas law, in the Fifth Circuit, what are the elements of tortious interference with an existing contract? Please cite controlling Texas Supreme Court and Fifth Circuit authority. Indicate the year of each case."
Frame the legal issue with the procedural context. A motion to dismiss brief requires authority on the pleading standard; a summary judgment brief requires authority on the evidentiary standard. Including procedural posture eliminates irrelevant authority and focuses the output on what you can actually use.
Template: "I am defending a Rule 12(b)(6) motion to dismiss in the [DISTRICT] District. The plaintiff alleges [CLAIM TYPE] under [STATUTE/COMMON LAW]. Under [CIRCUIT] Circuit precedent, what standard governs dismissal of [CLAIM TYPE] claims? Please provide citations with year and indicate whether each case is still good law."
Explicitly asking the AI to surface adverse authority is one of the most important techniques for producing defensible research. It fulfills your MRPC 3.3 obligation research function and produces a more complete legal analysis.
Template: "I am researching [LEGAL ISSUE] in [JURISDICTION]. First, provide the best three to five cases supporting the position that [YOUR POSITION]. Then, provide the two to three most significant cases adverse to that position that I should be aware of and address in my brief. For each case, provide the citation, a one-sentence description of the holding, and whether it is still good law."
Specify the standard of review when the procedural context is relevant to the authority you need. Standards of review directly affect which cases are on point — a highly deferential abuse-of-discretion standard calls for different precedent than a de novo standard.
Template: "On appeal to the [CIRCUIT], what is the standard of review for a district court's denial of a motion for [RELIEF TYPE] in a [CASE TYPE] matter? Please cite the leading [CIRCUIT] Circuit case establishing this standard and any circuit splits on this question as of [YEAR]."
Always ask the AI to indicate its confidence in citations and flag areas where it is uncertain. This does not eliminate hallucination, but it concentrates your verification effort on flagged items.
Template ending: "For each case you cite, indicate: (1) whether you are highly confident this case exists and accurately states the holding, (2) any areas where you are uncertain about the citation or holding, and (3) whether you recommend I independently verify the citation before relying on it."
Scenario: Research for a First Amendment retaliation claim
You represent a public school teacher in the Sixth Circuit who was terminated after speaking at a public school board meeting on a matter of public concern. You need research for an opposition to a motion for summary judgment.
Prompt (complete example):
"I am opposing a motion for summary judgment in a 42 U.S.C. § 1983 First Amendment retaliation case in the Sixth Circuit. My client is a public school teacher who was terminated after speaking at a public school board meeting. The school district argues the speech was not on a matter of public concern.
Please provide: (1) The Sixth Circuit's current test for whether public employee speech on a matter of public concern is protected under Pickering/Connick. (2) Sixth Circuit cases from the last ten years addressing public school teacher speech at school board meetings. (3) The two most significant adverse cases — cases where courts found such speech was not protected — that I should address in my brief. (4) For each citation, indicate the year, the circuit or court, and your confidence level in the accuracy of the citation."
This structured prompt produces a research output that covers your primary authority, the directly adverse authority, and confidence flags for citation verification.
CoCounsel — Retrieval-augmented research on Westlaw corpus; less hallucination than ungrounded LLMs. Use the templates above to structure queries.
Vincent AI — Grounded in vLex database; strong for international research prompting.
Casetext — CARA A.I. analyzes brief-length inputs and returns relevant cases; complement with structured prompts.
Westlaw Precision — The gold standard for citation verification after AI research.
Paxton AI — Government-focused research assistant; useful for regulatory research prompting.
See also: Westlaw vs Casetext comparison.
Case Law Research
"Under [JURISDICTION] law, what are the elements of [CLAIM TYPE]? Cite the controlling case(s) establishing the test with year and court. Indicate if the test has been modified by recent decisions."
"In the [CIRCUIT] Circuit, what is the current test for [LEGAL STANDARD]? Provide the leading case, the year it was decided, and any subsequent refinements. Note any circuit split."
"Find [JURISDICTION] cases from [YEAR RANGE] where courts [HELD/DECLINED TO HOLD] that [SPECIFIC PROPOSITION]. For each case: citation, key facts, holding, outcome."
"What are the best three cases supporting the argument that [PROPOSITION] under [JURISDICTION] law? What are the two strongest counterarguments based on existing precedent?"
"In [JURISDICTION], what is the statute of limitations for [CLAIM TYPE]? Does the discovery rule apply? Cite the controlling case and statute."
Statutory Interpretation
"Interpret [STATUTE CITATION] section [SECTION NUMBER]: what does the plain text require for [SPECIFIC FACT PATTERN]? Does legislative history modify the plain text interpretation? Cite relevant committee reports or floor debate if available."
"Has the [CIRCUIT] Circuit interpreted [STATUTE CITATION] to [SPECIFIC QUESTION]? Provide the leading case and whether the interpretation aligns with the majority of circuits."
"Under the [AGENCY] regulations implementing [STATUTE], what is required when [FACTUAL SCENARIO]? Cite the relevant CFR section and any agency guidance documents."
Regulatory Analysis
"Under [REGULATION CITATION], does [FACTUAL SCENARIO] trigger reporting requirements? Please confirm the current regulatory text and whether any recent amendments changed this analysis since [DATE]."
"What penalties apply under [REGULATION] for [VIOLATION TYPE]? Are there safe harbor provisions? Cite the enforcement guidance and at least one enforcement action."
"Compare the [AGENCY A] and [AGENCY B] regulatory requirements for [ACTIVITY TYPE]. Where do they overlap and where do they conflict?"
Brief Writing Support
"I am arguing [POSITION] in a [COURT TYPE] brief. Identify potential counterarguments my opponent is likely to raise, and for each: (a) the counterargument, (b) the legal authority supporting it, (c) my best response."
"Draft a statement of the standard of review for a [MOTION TYPE] in [JURISDICTION], citing controlling authority. Include the most favorable formulation of the standard for my position as [MOVANT/OPPONENT]."
"What are the leading cases in [JURISDICTION] on attorney fee awards in [CASE TYPE] matters? What factors do courts weigh? Provide citation and year for each case."
Contract Interpretation
"Under [JURISDICTION] contract law, how do courts interpret [SPECIFIC CLAUSE TYPE] when [AMBIGUITY SCENARIO]? Does parol evidence apply? Cite the controlling rule and a recent application."
"In [JURISDICTION], is a [CLAUSE TYPE] clause enforceable when [SPECIFIC FACTS]? Identify any public policy exceptions that might invalidate it. Cite cases for and against enforceability."
Due Diligence
"What are the standard representations and warranties a target company provides regarding [SPECIFIC SUBJECT MATTER] in a [DEAL TYPE] transaction under [JURISDICTION] law? What disclosure schedules are typically required?"
"Identify the key regulatory approvals required in [JURISDICTION] for [TRANSACTION TYPE]. What is the typical timeline for each approval? Are there any recent regulatory changes that affect this analysis?"
Q: Do these prompt techniques work differently on retrieval-grounded tools versus general LLMs?
A: Yes. On retrieval-grounded tools like CoCounsel or Vincent AI, the jurisdiction anchoring primarily focuses the retrieval query. On general LLMs, it also serves as a guardrail against generating out-of-jurisdiction phantom citations. Apply confidence-level flagging and independent citation verification more aggressively when using ungrounded LLMs.
Q: How should I handle it when the AI flags low confidence on a citation?
A: Verify it in Westlaw or Lexis before using it in any document. A low-confidence flag means the AI is uncertain whether the citation is accurate. Do not use it in a brief until you have confirmed it exists and the holding is as described.
Q: Can I use these templates with tools like Harvey or CoCounsel that have their own structured interfaces?
A: Yes. These templates are designed to work in any text-input interface. For tools with structured research workflows, adapt the template language to fit the interface — the underlying principles (jurisdiction anchoring, adverse authority, confidence flagging) apply universally.
Q: Should I include client facts in prompts sent to external AI tools?
A: No, unless you have confirmed the tool's data processing terms and your jurisdiction's ethics guidance on client confidentiality. Strip identifying facts from research prompts and use generic placeholders. Research-focused prompts rarely need specific client information to generate useful general legal analysis.
Q: How many templates should I use per research task?
A: Use multiple templates for complex research tasks — one for primary authority, one for adverse authority, one for procedural posture. For straightforward factual legal questions, a single well-structured prompt is sufficient.
Legal research prompting is a discipline, not a casual skill. The five techniques — jurisdiction anchoring, issue framing with procedural posture, adverse authority requests, standard of review specification, and confidence-level flagging — produce meaningfully better research outputs and reduce the verification burden.
The 20 templates are starting points. Adapt them to your practice area, jurisdiction, and tool. The [BRACKETED] variables remind you to fill in the specific facts that make a prompt legally precise rather than legally general.
No prompt eliminates the need to verify citations before use. The goal is to reduce the time spent on research and increase the precision of outputs — not to replace attorney judgment about what the law means.
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: 2026-07-21.