Best Gemini Prompts for Lawyers in 2026

Three properties separate useful prompts from sanction-risk prompts. **Scope discipline:** prompt asks Gemini to do a defined task (summarize, outline, extract), not to answer an open legal question. **Citation grounding:** prompt supplies the source text or requires Gemini to mark every assertion as derived-from-provided-text vs. general-knowledge. **Reviewer assumption:** output is draft input to a licensed attorney, never final advice to a client.

Per ABA Formal Opinion 512 (July 2024), Rule 1.1 competence covers any technology used in representation; attorneys must verify outputs. *Mata v. Avianca*, 22-cv-1461 (S.D.N.Y. 2023) sanctioned attorneys $5,000 for six ChatGPT-fabricated cases. *Park v. Kim* (2d Cir. 2024) extended the rule to appellate practice. The LegalAI Hallucination Tracker lists 31+ U.S. cases with AI-fabricated authority as of mid-2026.

**Prompt block:** "Summarize the attached opinion in 250 words. For every legal proposition you state, cite the page or paragraph of the opinion where it appears. Mark any statement you cannot ground in the attached text as [GENERAL KNOWLEDGE — VERIFY]. Do not extend the holding beyond what the court actually decided. Do not predict how other courts would rule. Structure: (1) procedural posture, (2) facts, (3) issue, (4) holding, (5) reasoning, (6) disposition."

**Safe-use note:** Upload the actual opinion. Never let Gemini retrieve case law from training data alone — that is the *Mata* failure mode. Per Stanford CodeX, unsourced legal queries hallucinate 58-82%.

**Sample output:** "(1) Posture: appeal from summary judgment, p. 1. (2) Facts: plaintiff alleged breach of fiduciary duty (p. 2-4). (3) Issue: whether a managing member owes fiduciary duties under Del. LLC Act §18-1101(c). [GENERAL KNOWLEDGE — VERIFY] Act amended 2018 to clarify default duties."

**Prompt block:** "You are a litigation associate drafting deposition outline questions. Based on the attached fact summary and produced documents, generate 40 deposition questions for [WITNESS NAME] organized into 6 topical buckets. Start broad, narrow into specifics. Flag any question that requires a legal conclusion from the witness — those go in a separate 'objection-likely' section. Do not invent facts not in the source material."

**Safe-use note:** Gemini sometimes assumes facts not in the record. Strike anything introducing unfamiliar dates, dollar amounts, or third parties. The drafting attorney owns every question per the duty of competence.

**Sample output:** "Topic 1 — Background (Q1-Q6). Topic 4 — Knowledge of the March 12 meeting (Q22-Q28). [OBJECTION-LIKELY] Q29: Did you believe the contract was enforceable?"

**Prompt block:** "I will paste the original contract clause and my proposed redline. Explain the rationale for each substantive change in plain English suitable for client communication. Identify any change that materially shifts risk allocation. Do not opine on enforceability. Do not state that a clause is or is not legally binding. Flag changes that may trigger renegotiation of related clauses."

**Safe-use note:** Useful for explaining your edits to a client or junior associate. Not a substitute for an enforceability opinion — that requires jurisdiction-specific analysis. Per ABA Opinion 512, opining on enforceability from model output without independent attorney analysis risks competence and supervision violations.

**Sample output:** "Change 1 (§4.2 indemnification cap): raised from 1x to 2x fees. Rationale: aligns with limit-of-liability section. Risk shift: increases counterparty exposure; expect pushback. Related: §7.1 (insurance) may need higher coverage."

**Prompt block:** "I will paste a draft brief paragraph with citations. For each citation, do the following: (1) state the case name, court, and year as written; (2) identify whether the cited proposition can be verified from the citation alone or requires the underlying opinion; (3) flag any citation that does not match standard Bluebook format. Do not assume any citation is accurate. Do not generate replacement citations. If a citation appears suspicious, output [VERIFY MANUALLY] with reasoning."

**Safe-use note:** Structurally safer than "check my citations" because it forbids Gemini from generating replacements — the source of nearly every reported sanction. Always verify on Westlaw, Lexis, or Bloomberg Law before filing. Per *Mata v. Avianca*, citation verification is non-delegable.

**Sample output:** "Citation 1: *Smith v. Jones*, 123 F.3d 456 (2d Cir. 2019). Format: Bluebook-compliant. [VERIFY MANUALLY] 2d Cir. 2019 F.3d volumes run ~920-960; vol. 123 is outside range — possible transposition or fabrication."

**Prompt block:** "I will paste a recording transcript or notes from a client intake call. Produce a structured summary: (1) parties involved, (2) chronology of events with dates, (3) documents the client mentioned, (4) potential causes of action or claims the client described in their own words (do not add legal theories the client did not raise), (5) deadlines or statute-of-limitations triggers the client mentioned, (6) open questions for follow-up. Do not provide legal advice. Do not characterize the client's claims as strong or weak."

**Safe-use note:** Fast triage of incoming matters. Do not let the intake summary become the case theory — that emerges from attorney analysis. Confidentiality: confirm firm AI policy permits sending client info to Gemini, and use Workspace Enterprise tier for data-protection commitments.

**Prompt block:** "I will paste my time entries for the day as raw notes. Rewrite each entry as a billing narrative following these rules: (1) use active voice and specific verbs (drafted, reviewed, analyzed, conferred); (2) reference the matter and document by name; (3) avoid vague terms (worked on, looked at); (4) keep each narrative under 25 words; (5) preserve the time duration exactly as I provided it; (6) do not combine entries unless I instructed you to."

**Safe-use note:** Pure rewriting task — low hallucination risk because Gemini is transforming text you supplied. Still review for accuracy; clients reject vague entries and over-precise entries that exceed what you actually did.

**Sample output structure:** "0.4 — Drafted reply to opposing counsel email re: discovery deadline extension; reviewed prior correspondence; confirmed extension date with client."

**Prompt block:** "I will provide the procedural posture, the relief sought, the governing rule or statute, and a short fact statement. Outline a motion brief with these sections: (1) introduction (3-5 sentences), (2) statement of facts (cite to record), (3) standard of review, (4) argument (with proposed headings as questions), (5) conclusion. Do not write the argument — only the headings. Do not invent record cites; mark them as [RECORD CITE NEEDED]."

**Safe-use note:** Outline-only drastically reduces hallucination risk versus full-brief generation. The lawyer still owns the argument. Per Thomson Reuters' 2025 survey, outline-and-edit workflows score highest among legal AI use cases.

**Prompt block:** "I will paste the text of a produced document. Classify it across these dimensions: (1) document type (email, memo, contract, financial record, other), (2) date or date range mentioned, (3) people or entities named, (4) topics discussed (3-5 keywords), (5) relevance flag (HIGH / MEDIUM / LOW / IRRELEVANT) based on the case theory I provide below, (6) privilege risk (potentially privileged / not privileged / unclear). Provide reasoning for privilege risk in one sentence. Do not make a final privilege determination."

**Safe-use note:** First-pass triage only. Final privilege calls are non-delegable. Route HIGH-relevance or PRIVILEGED flags to attorney review immediately. Per the Sedona Conference Principles on TAR, AI-assisted review is now widely accepted when paired with attorney protocol sign-off.

**Prompt block:** "I will provide the case posture, our strongest 3 arguments, opposing counsel's strongest 3 arguments, and our settlement authority range. Draft a 1-page settlement position memo for internal use only that: (1) frames our opening number with rationale, (2) anticipates the opposing party's counter, (3) identifies our walk-away and the rationale, (4) lists 3 non-monetary terms that may unlock value. Do not characterize litigation outcomes as guaranteed."

**Safe-use note:** Internal-use only. Never share a Gemini-drafted position outside the firm without attorney rewriting. The model has no live knowledge of jury verdicts, opposing counsel history, or judge mediation patterns — those inputs come from you.

**Prompt block:** "I will paste a legal document (court order, contract, statute excerpt). Rewrite it as a plain-English explanation for a client at an 8th-grade reading level. Preserve every legally significant detail (dates, amounts, deadlines, conditions). Do not interpret the document. Do not predict outcomes. End with: 'I am happy to discuss this further on a call.' Tag any term that should remain in legalese for accuracy with [LEGAL TERM] notation."

**Safe-use note:** Translation, not analysis. Accuracy risk is losing nuance during simplification. Read original alongside before sending. Per ABA Opinion 512 and Rule 5.3, the attorney verifies that simplification did not introduce inaccuracy.

**Sample output:** "The court set a hearing for July 15 at 9:00 a.m. on the [LEGAL TERM: motion to dismiss]. You do not need to attend. If granted, the case ends. If denied, it continues to [LEGAL TERM: discovery]."

**Prompt block:** "I will paste a list of parties, related entities, and counsel from a potential new matter. Cross-reference against the list of current and former clients I will paste below. Identify exact name matches, near matches (likely typos or variations), and any indirect connections (parent-subsidiary, common counsel). Do not opine on whether a conflict exists. Output a table for human review only."

**Safe-use note:** Conflict determination is a judgment call under Rules 1.7-1.10; Gemini's role is pattern matching to surface candidates for review. Validate every match against the firm conflicts database.

**Prompt block:** "I will paste opposing counsel's brief. Identify and list: (1) unsupported assertions (claims without citation), (2) citations that lack a pin cite, (3) legal standards stated without authority, (4) factual claims that may be contested by our record evidence (which I will summarize), (5) logical gaps between premise and conclusion. Do not make a final argument. Do not rewrite their brief. Output a structured list for our reply."

**Safe-use note:** Starting list for attorney review. Gemini's scan is one input; pair with the lawyer's own re-read.

**Sample output:** "Unsupported assertion #1 (p. 4): 'Defendant has consistently demonstrated bad faith' — no record cite. Counter: cite May 3 and June 12 good-faith negotiation records."

Using Gemini as a legal advisor: produces fabricated citations, sanctions exposure, and competence-rule violations. Every reported AI-sanction case shares this pattern — attorneys treated model output as legal research rather than draft input.
Using Gemini as a drafting and triage assistant: scope-disciplined prompts + citation grounding + human review at the end. Per the ABA 2025 TechReport, this is the pattern producing the documented time savings without the documented sanctions.