Find Legal AI Prompts and Practical Templates

Choose the task you want help with

Do not upload a real client document simply because a template asks for source text. Use your organization's approved systems and confidentiality rules.

Template 1 — Summarize a published case

Suitable input: A public judicial opinion that you are authorized to use, with court, date and link.

Read only the published opinion provided here: [SOURCE LINK OR PUBLIC TEXT]. Create a neutral summary with these headings: court and date; parties; question before the court; key facts stated in the opinion; ruling; reasoning; important qualifications. Quote no passage unless you can point to its location. Distinguish the court's words from your own summary. If information is absent, write “not established from this source.” Do not add outside cases or statutes.

Before use: Check the citation, procedural status, actual holding, omitted dissents or exceptions, and whether the decision remains good law. The summary is not a substitute for reading the opinion.

Template 2 — Check an AI-generated citation

Suitable input: A draft citation and a permitted public source URL. Do not ask the model to invent authorities.

Review this legal citation: [CITATION]. If a source is available, identify its court or issuing body, date, title, and the exact passage relevant to the accompanying statement: [STATEMENT]. If the source cannot be accessed or does not support the statement, say so clearly. Do not replace the citation with a guessed authority. Return a table: citation; source found; proposition supported; qualification; manual action needed.

Before use: Open the source yourself. Confirm page or paragraph references and check any treatment by later decisions or amendments.

Template 3 — Organize a fictional contract issue list

Suitable input: A fictional agreement or a document explicitly approved for the tool and task.

Read the supplied agreement. Identify where it discusses the parties, duration, termination, renewal, confidentiality, governing law, liability and dispute resolution. For each topic, provide the clause location and a neutral description. If a topic is not found, write “not located”—do not assume it is absent from every version. List ambiguities and questions for a qualified reviewer. Do not state whether the agreement is enforceable.

Before use: Compare with the complete signed or approved version, including attachments, definitions, amendments and jurisdiction-specific requirements.

Template 4 — Prepare a legal request intake outline

Suitable input: A fictional scenario without identifying individuals.

Create a short intake questionnaire for this type of request: [GENERAL CATEGORY]. Ask what happened, relevant dates, involved organizations, documents already available, location or jurisdiction, decision deadline and the result the requester is seeking. Avoid asking for unnecessary personal information. Flag emergency or time-sensitive questions for prompt human review. Do not provide a legal conclusion.

Before use: Adapt to your organization's data policy, intake categories and routing process. Have the responsible professional approve the questions before using them with clients.

Template 5 — Review an AI answer before sharing

Assess the draft below only for reviewable quality issues. Make a table containing: claim; cited supporting source; missing source; ambiguity; possible outdated information; confidentiality concern; recommended human check. If you cannot substantiate a claim, mark it “unverified.” Do not invent evidence and do not rewrite an unverified claim as fact. Draft: [APPROVED NON-SENSITIVE TEXT].

The reviewer—not the tool—decides what can be kept, corrected or removed.

What information should you never assume is safe to enter?

Before copying material into a prompt, check whether it includes client identities, privileged communications, employee records, private case files, trade secrets or sensitive personal data. A public example is not the same as a private client document. Product configuration and organizational agreements can affect permitted use.

The American Bar Association's discussion of AI in legal practice stresses professional accountability, accuracy, confidentiality and oversight. The NIST generative-AI risk profile offers additional context for evaluating AI use.

Keep a record of template changes

For every adapted template, record its purpose, date, owner, approved data types and review step. Mark instructions that refer to specific software capabilities as tool-specific, because they may not work the same way elsewhere. If a prompt fails to flag an important exception, revise it and repeat the test rather than assuming the next answer will be reliable.

Example: public case-summary review

Suppose an AI summary states a court decided a broad legal rule. The reviewer should open the opinion, identify the precise question decided, check whether the statement came from the majority, distinguish a court's holding from passing commentary, and record the result. The correct outcome may be to narrow or remove the claim.

Frequently asked questions

Do prompts replace a legal professional? No. They organize work; they do not take responsibility for source verification, legal judgment or professional duties.

Will one template work with Claude, ChatGPT and every other assistant? Not necessarily. Input formats, retrieval capabilities and connected data differ. Recheck the result in the specific tool and configuration.

Can a prompt guarantee accurate legal citations? No. Ask for evidence and manually confirm it.

Can these examples be used for a law firm's internal training? They may be adapted as educational examples, subject to the firm's review, confidentiality requirements and the final published use terms.

Choose a template and check the result

Start with one low-risk public or fictional example. Record the expected result, test the instruction, and review the output using the companion Legal AI Evaluation Checklists page (B02-007).