Research scaffolding that never supplies a citation
Legal research is a top AI use case at 74% — and the source of nearly every AI sanctions headline, because general-purpose chatbots fabricate convincing case citations. The safe version flips the task: use AI to map the issues, generate search strategies, and outline the memo, while explicitly forbidding it from naming authority. The citations come only from Westlaw or Lexis, pulled and read by you.
You are a legal research planning assistant. HARD RULE: you must not name, cite, or quote any specific case, statute section, regulation, or secondary source. If you are about to name one, write [RESEARCH — verify in Westlaw/Lexis] instead. Everything you produce is a hypothesis for me to verify, not authority. Research question: {{research_question}} Jurisdiction: {{jurisdiction}} Known facts: {{known_facts}} Produce: 1. The question restated as 2-4 precise sub-issues. 2. For each sub-issue: the elements, factors, or tests likely at play, phrased as questions to verify rather than statements of law. 3. Search strategy: 3-5 Boolean search strings per sub-issue suitable for Westlaw/Lexis, plus the categories of authority to look for (e.g., "the controlling statute," "the leading state supreme court test," "a recent intermediate appellate application") described generically. 4. The strongest counterarguments opposing counsel would raise, framed as risks to research. 5. A memo outline skeleton (QP, brief answer, discussion headings) with [RESEARCH] placeholders where authority belongs. Do not predict what the law is. Flag any sub-issue where the answer likely varies by jurisdiction.
Fill in your details and the prompt updates live — then copy.
Sub-issue 1 (restated): Whether the general contractor qualifies as a "statutory employer" entitled to workers' compensation exclusivity against a subcontractor's employee. Questions to verify: (a) Does the state's exclusivity provision extend up the contracting chain? [RESEARCH — verify in Westlaw/Lexis] (b) Does the answer change when the GC directly controlled site safety? Suggested searches: adv: "statutory employer" /p subcontractor /p exclusiv! — filter to Georgia appellate courts, last 10 years. Authority to look for: the exclusivity statute; the leading state supreme court test [RESEARCH — verify in Westlaw/Lexis]; recent appellate applications involving general contractors. Counterargument risk: an intentional-act or independent-duty exception.
The full workflow
- Run the prompt to map issues and generate search strings before opening Westlaw or Lexis.
- Run every search in a real legal database and pull the actual authority.
- Read every case you intend to cite — full text, not the AI's description, checking it's still good law.
- Draft the memo into the outline, replacing each [RESEARCH] placeholder with verified, Shepardized/KeyCited authority.
- Before filing anything, re-verify every citation independently — including ones you're sure about.
Watch out for
Never file a citation an AI produced without pulling and reading it in Westlaw or Lexis. Roughly 1,490 court decisions worldwide now involve hallucinated authority, and sanctions have escalated from Mata v. Avianca's $5,000 to $15,000 per attorney and indefinite bar suspensions — courts no longer treat 'the AI told me it was real' as mitigation.
Check your court's standing orders: a growing number of judges require certification that filings were checked for AI-generated content, and ABA Formal Opinion 512's competence duty means you're expected to know how these tools fail.
Asking the chatbot to confirm its own citations doesn't work — that's exactly what the Mata lawyers did. Verification must happen in an authoritative legal database.
Where this comes from
Every use case on this site is grounded in real reports from working lawyers — not invented by us.