AI Prompts for Lawyers: A Library by Practice Area

The prompts for lawyers are already a core professional skill: they distinguish the practitioner who gets usable drafts from the one who gets generic text. This library brings together the structure and templates we install at law firms when implementing Claude — adaptable to any serious model.

# What is legal prompting, and why does it matter?

Legal prompting is the skill of writing precise instructions so a language model can handle legal tasks with the level of detail, tone, and structure required by legal practice. It matters because the same model produces radically different results depending on the instruction: “draft a lease” produces a generic template; a prompt with a role, context, format, and constraints produces a draft an associate can polish in minutes.

# How do you structure an effective legal prompt?

The five parts, in order:

• 1 · Role: “Act as a Mexican litigator specializing in commercial law…”—sets the register and legal framework.
• 2 · Context: the working materials (facts, attached contract, applicable legal precedents). The golden rule: AI works with what you give it, not what it “knows.”
• 3 · Task: a specific verb and a limited scope (“identify risky clauses,” not “review this”).
• 4 · Format: the exact output structure (sections, table, maximum length).
• 5 · Constraints: “Don’t invent citations. If there’s no basis for it in the material provided, say so explicitly.” — the line that prevents a Mata v. Avianca.

# Which prompts should you use for each practice area?

# Civil / Commercial — contract review

Actúa como abogado mercantil mexicano. Te entrego un contrato de
suministro [PEGAR/ADJUNTAR]. Tarea: produce un informe en 5 secciones:
1) Objeto y partes en lenguaje claro
2) Obligaciones y riesgos por parte
3) Cláusulas críticas o desequilibradas que conviene renegociar
4) Cláusulas faltantes vs. estándar de mercado mexicano
5) Recomendaciones concretas para proteger a mi cliente [VENDEDOR/COMPRADOR]
Formato: secciones numeradas, máximo 150 palabras cada una.
Restricción: no cites legislación ni tesis que no estén en el material;
si un punto requiere verificación, márcalo como [VERIFICAR].

# Labor — specific clause

Actúa como abogado laboral mexicano. Redacta una cláusula de duración
para contrato individual de trabajo de un empleado residente en México,
con periodo de prueba de 90 días conforme a la LFT vigente [ADJUNTAR
ARTÍCULOS RELEVANTES]. Formato: cláusula lista para insertar + nota de
3 líneas explicando el fundamento. Restricción: si el supuesto no
admite periodo de prueba de 90 días, dilo antes de redactar.

# Litigation — witness statement analysis

Actúa como asociado de litigio. Te entrego [N] declaraciones
testimoniales [ADJUNTAR]. Tarea: 1) cronología unificada de hechos,
2) tabla de contradicciones entre declarantes (quién, qué, dónde
difieren), 3) hechos no controvertidos. Formato: cronología con
fechas, tabla de 3 columnas, lista final. Restricción: cita textual
de cada declaración al señalar una contradicción.

# Tax / Corporate — due diligence

Actúa como abogado corporativo en una operación de M&A. Te entrego
los contratos de la carpeta [X] del data room. Tarea: identifica por
documento: cambio de control, consentimientos requeridos, exclusividad,
no competencia, terminación anticipada y pasivos contingentes.
Formato: tabla por documento + semáforo de riesgo (alto/medio/bajo)
con justificación de una línea. Restricción: marca [ILEGIBLE/FALTA]
donde el documento esté incompleto en vez de asumir contenido.

# What mistakes ruin a legal prompt?

• Asking for open-ended research without providing material: that’s a recipe for fabricated citations. AI synthesizes what you give it; source research should be done in official databases.
• Tasks without a defined scope: “review this contract” produces generalities. “Identify early termination clauses and their risks” produces useful work.
• Ignoring the format: without a structured output, every result is different and can’t be standardized across the firm.
• Forgetting the anti-hallucination constraint: explicitly requiring it to flag uncertainty turns silent risk into visible red flags.

# How do you scale from one-off prompts to a firm-wide library?

The real productivity leap comes not from individual prompts but from a curated library: versioned templates by practice area, each with an owner, tested against the firm’s own forms, and integrated into onboarding for new hires. It’s one of the core deliverables when we implement Claude at a law firm — along with usage policies and the verification process.

# Frequently asked questions

# Do these prompts work in ChatGPT, or only in Claude?

The five-part structure works with any serious model. The templates are optimized for Claude because of its 200K-token context window — key for attaching complete contracts and case files — but they can be adapted.

# Can I ask AI to find applicable case law?

Only for materials you provide. Researching sources is done in the Semanario Judicial and official databases; AI synthesizes, compares, and organizes what it finds. Asking it for case law “from memory” invites fabricated citations.

# How do I measure whether a prompt is good?

By editing time: a good prompt produces a draft the lawyer can refine in minutes, rather than rewrite. If the team rewrites more than 50% of the output, the prompt needs more context or a better output format.

# Are prompts confidential to the law firm?

They should be: a library tailored to your mix of practice areas is a cumulative competitive advantage. Treat it like the firm’s form documents—an internal, versioned asset with controlled access.

Cluster readings: Complete guide to AI for lawyers · Is it safe to use AI at a law firm?