The CRAFT Prompting Framework for Legal Work

Why Most Legal Prompts Fail

The single biggest reason attorneys get poor results from Claude AI is vague prompting. Typing "summarize this contract" into Claude produces a generic overview that no attorney would use. The difference between a useless AI response and a litigation-ready work product comes down to prompt structure. The CRAFT framework, developed specifically for legal professionals, solves this problem with a repeatable five-step method that works across every practice area.

What Is the CRAFT Prompting Framework?

CRAFT stands for Context, Role, Ask, Format, Tone. It is a structured approach to writing prompts that give Claude AI enough information to produce substantive, practice-ready output. Each element addresses a specific failure mode in legal AI prompting:

• Context eliminates ambiguity by grounding the task in jurisdiction, practice area, and facts.
• Role activates domain-specific reasoning by telling Claude what kind of attorney it should emulate.
• Ask prevents scope creep by specifying exactly what deliverable you need.
• Format ensures the output is structured for immediate use — memo, table, checklist, or redline.
• Tone sets the professional register so the output matches its intended audience.

Step 1: Context — Ground the Task

Every legal task exists within a specific context. Claude cannot infer your jurisdiction, the governing law, or the client situation. You must state it explicitly. A strong context statement includes the jurisdiction (e.g., "Delaware corporate law"), the transaction or matter type (e.g., "Series B preferred stock financing"), and relevant constraints (e.g., "the company has existing SAFE notes converting in this round").

Example: "We represent a Delaware-incorporated SaaS startup in a $15M Series B round. The lead investor is requiring a full-ratchet anti-dilution provision. California governs the side letter."

Step 2: Role — Activate Domain Expertise

When you tell Claude to "act as a senior corporate attorney specializing in venture capital transactions," the model draws on a deeper, more specialized subset of its training. This produces output that reflects the conventions, terminology, and analytical frameworks of that specific practice area. Be precise: "litigation attorney" is weaker than "senior commercial litigation partner with 15 years of breach-of-contract experience in federal court."

Step 3: Ask — Specify the Deliverable

State exactly what you need. Not "review this contract" but "identify every provision in this SPA that allocates indemnification risk to the buyer, flag any that deviate from ABA model terms, and recommend specific redline language for each." The more specific your ask, the more actionable Claude's output.

Step 4: Format — Structure the Output

Legal work demands structure. Tell Claude whether you need a formal memo with IRAC analysis, a bullet-point risk summary, a comparison table, a redline draft, or a client-ready email. If you need section headers, specify them. If you need the output in a specific order (e.g., highest risk first), say so.

Step 5: Tone — Match the Audience

A brief filed in federal court demands different language than a client advisory email. Tell Claude who the audience is: "persuasive tone suitable for a motion to dismiss," "plain English suitable for a non-lawyer board member," or "analytical tone for an internal research memo." This final step ensures the output is immediately usable without extensive editing.

CRAFT in Action: A Complete Example

Here is a full CRAFT prompt for AI-assisted contract review:

"Context: We represent the buyer in a $50M asset purchase of a manufacturing business in Ohio. The seller has provided a 60-page APA with Ohio choice-of-law. Role: Act as a senior M&A attorney with experience in manufacturing sector acquisitions. Ask: Review the attached APA and produce a risk matrix identifying every seller representation and warranty, rate each for risk (high/medium/low), and flag any that are narrower than ABA model provisions. Format: Structured table with columns for Section, Rep/Warranty Summary, Risk Rating, Issue Description, and Recommended Redline. Tone: Analytical and precise, suitable for a partner review memo."

Common Mistakes to Avoid

• Skipping Context: Without jurisdiction and facts, Claude defaults to generic legal reasoning that may not apply.
• Being too vague on Ask: "Help me with this contract" is not a prompt — it's a wish.
• Forgetting Format: Claude will choose its own format if you don't specify, and it rarely matches what you need.
• Ignoring Tone: A motion brief and a client email require fundamentally different registers.

For more prompting techniques and practice-area-specific examples, see How to Write Legal Prompts That Actually Work. If you're just getting started with Claude, our complete guide for attorneys covers the fundamentals.

Get the complete guide with 50+ CRAFT-formatted prompts across 11 practice areas.