Prompts IA pour avocats : bibliothèque de 168 prompts + guide 2026

Why Prompts Matter More Than the AI Model You Use

Every legal AI tool on the market, whether it is ChatGPT, Claude, Gemini, or a purpose-built platform like HAQQ, runs on the same fundamental technology: large language models. These models are pattern-recognition engines trained on vast amounts of text. They do not store facts. They predict what words should come next based on the instructions you give them.

This means that the quality of your output is overwhelmingly determined by the quality of your input. A vague prompt produces vague results. A structured, context-rich prompt produces structured, actionable work product. The difference between AI that wastes your time and AI that saves you hours comes down to one thing: how you write your prompts.

For lawyers, this is not a technical curiosity. It is an operational reality. The firms and legal departments that master prompting will outperform those that do not. This guide will teach you exactly how.

What Is a Prompt, Exactly?

A prompt is the instruction you type into an AI tool. It tells the model what to do, what context to consider, what format to follow, and what constraints to respect. Think of it as a brief to a junior associate: the more precise the brief, the better the work product.

In legal practice, prompts are not casual questions. They are structured instructions that define scope, jurisdiction, output format, and audience. A well-crafted legal prompt contains four elements: role, context, task, and format.

• Role: Define who the AI should act as. Example: 'You are a senior corporate lawyer specializing in M&A transactions.'
• Context: Provide the factual background. Example: 'You are reviewing a vendor agreement for a SaaS company under English law.'
• Task: State exactly what you need. Example: 'Identify all non-standard indemnity clauses and suggest alternative wording.'
• Format: Specify the output structure. Example: 'Present findings as a risk table with columns for clause reference, risk level, and recommended revision.'

How Large Language Models Process Your Prompts

Understanding how LLMs work helps you write better prompts. Unlike a database that retrieves stored answers, an LLM generates text by predicting the most likely next word based on everything it has seen in training and everything you provide in your prompt.

This has several practical implications for lawyers. First, attention: the model processes all parts of your input simultaneously, paying attention to every word. If you include examples of poor drafting, it may reproduce elements of them. For drafting tasks, always show good examples only. Second, probabilities: the model does not pick the same word every time. More structured prompts reduce variation and increase reliability. Third, task complexity: asking the model to handle a complex, multi-step task in one prompt will produce weaker results than breaking it into sequential steps.

The Prompt Formula: Intent + Context + Instruction

Thomson Reuters recommends a simple formula for well-structured prompts that applies across all legal AI tools: Intent + Context + Instruction. Start with a clear expression of what you are trying to achieve. Then provide the contextual background that anchors the AI's response. Finally, add the specific instruction telling the AI what task to perform.

For example, your intent might be: 'I need to assess whether this expert witness can be discredited.' Your context: 'The document contains all prior testimony of the expert in a medical malpractice case.' Your instruction: 'Does the document contain any contradiction inconsistent with the expert's current testimony?'

Seven Techniques That Transform Legal Prompts

Based on analysis of best practices from leading legal AI practitioners, here are seven techniques that consistently produce superior results.

1. Assign a Persona

Telling the AI to act as a specific type of legal professional narrows the scope of its response and improves relevance. Instead of a generic answer, you get analysis from the perspective of a specialist. Example: 'You are an experienced US-based data privacy lawyer. Explain the differences between a data processor and a data controller under GDPR.'

2. Provide Deep Context

Context eliminates ambiguity. Include the type of case, the jurisdiction, the parties involved, the relevant legal framework, and any specific constraints. The more context you provide, the less the AI has to guess. Example: 'You are reviewing a cross-border supply agreement between a US manufacturer and an EU distributor. The agreement is governed by German law.'

3. Break Complex Tasks Into Steps

LLMs produce significantly better results when you decompose a complex task into sequential steps rather than asking for everything at once. Instead of 'Draft a full board resolution,' try: Step 1: 'Outline the key sections of a board resolution authorizing a partnership agreement.' Step 2: 'Draft the recitals section.' Step 3: 'Draft the operative clauses.'

4. Specify the Output Format

Explicitly state whether you want a table, a memo, a numbered list, a redline comparison, or a narrative summary. Use placeholder patterns like [mm/dd/yyyy]: [description] to show the AI exactly what format you expect. This alone can transform unusable output into work-product-ready deliverables.

5. Set Guardrails

Define what the AI must do, not just what it should avoid. Positive instructions outperform negative ones. Instead of 'Do not include irrelevant information,' say 'Only include analysis relevant to indemnity and liability clauses under English law.' Also specify what to do when uncertain: 'If a clause is ambiguous, flag it as requiring human review rather than interpreting it.'

6. Iterate and Refine

No prompt is perfect on the first try. Treat AI interaction as a conversation. Start with your initial prompt, review the output, then refine: 'Expand on the data protection section.' 'Rewrite this for a non-legal audience.' 'Add three more examples of enforcement actions.' Each iteration gets you closer to the exact output you need.

7. Use Prompt Reinforcement

For critical instructions, repeat them. Place the most important directive at both the beginning and the end of your prompt. This combats the 'lost middle' bias, where LLMs tend to pay less attention to information in the center of long prompts. If accuracy on a specific point is essential, reinforce it.

Common Prompting Pitfalls Lawyers Must Avoid

Understanding what not to do is as important as knowing what to do. Here are the most common mistakes that degrade AI output quality.

• Information overload: Adding too much irrelevant data confuses the AI. Unlike a human, an LLM cannot distinguish between relevant and irrelevant information and will consider everything you give it.
• Primacy and recency bias: LLMs tend to forget information in the middle of long prompts. Keep your most important instructions at the beginning or end.
• Majority label bias: Words or concepts that appear more frequently in your prompt receive disproportionate weight. Use this to your advantage with reinforcement, but be aware it can unintentionally skew results.
• Lumping topics together: Separate different tasks, topics, or projects into distinct prompts or conversations. Most AI tools offer ways to segregate contexts.
• Assuming the AI will assume: Do not expect the AI to fill in gaps. Be explicit about every element of your request.

Ready-to-Use Legal Prompts by Practice Area

Here are practical, copy-ready prompts organized by practice area. Each follows the Role → Context → Task → Format structure.

Contract Review & Drafting

• 'You are a senior commercial lawyer. Review this supplier agreement and identify risks related to indemnity, liability, termination, and exclusivity. Explain them in business-friendly language.'
• 'Draft a limitation of liability clause for a SaaS agreement, capping liability at 12 months of fees paid and excluding indirect damages.'
• 'Redline this NDA to ensure mutual protection, prohibit reverse engineering, and extend trade secret protection indefinitely.'
• 'Generate a short, plain English summary of the key terms in this statement of work for the project manager.'

Data Privacy & Compliance

• 'Create a CCPA/CPRA compliance checklist for responding to consumer data access, deletion, and opt-out requests.'
• 'Explain the difference between a controller and a processor under GDPR, using examples relevant to a SaaS company.'
• 'List the first steps an in-house team should take in the first 48 hours of a suspected data breach involving customer data.'
• 'Draft a short internal memo outlining rules for using AI tools while protecting personal and confidential data.'

Legal Research & Analysis

• 'Conduct legal research on force majeure clauses in commercial contracts. Summarize the relevant case law, statutes, and regulations. Provide analysis and conclusions.'
• 'Provide an overview of the legal precedents in employment law for wrongful termination. Identify key cases and the key legal arguments.'
• 'What are the governing statutes and regulations relating to data portability in the European Union?'
• 'Draft a legal brief arguing that the non-compete clause is unenforceable based on applicable case law and legal principles.'

Corporate Governance & M&A

• 'Write a one-page overview of directors fiduciary duties under Delaware law, covering duty of care, duty of loyalty, and the business judgment rule.'
• 'Draft a board resolution authorizing the CEO to execute a partnership agreement with a third-party provider.'
• 'List key legal due diligence questions for a potential acquisition in the software sector, focusing on IP, contracts, litigation, and employment.'
• 'Explain what a no-shop clause is and how it protects parties in an acquisition.'

The Security Problem With Generic AI Tools

Before discussing where to prompt, we must address the elephant in the room: security. Generic AI tools like ChatGPT, Claude, and Gemini are powerful, but they are not designed for legal-grade confidentiality.

Anything pasted into a public AI system may be processed, stored, or used for training outside your control. This creates serious risks when client data, counterparty information, or privileged communications are involved. A federal judge has already ruled that documents generated using public AI tools are not protected by attorney-client privilege.

• Data exposure risk: Your prompts and the documents you paste may be stored, logged, or used to train future model versions.
• No legal context: Generic models lack knowledge of your playbooks, templates, and risk thresholds.
• No consistency: Each session starts fresh. You re-explain your role, your standards, and your objectives every time.
• No exportable deliverables: Output is raw text that requires significant reformatting before it can be sent to a client.

If you must use a generic AI tool, always remove sensitive information first. Replace names, dates, amounts, and company details with placeholders. Never paste full contracts or regulated content into public AI platforms.

Why Purpose-Built Legal AI Changes the Prompting Equation

Purpose-built legal AI platforms like HAQQ fundamentally change how prompting works. Instead of starting from zero every time, you work within a system that already understands legal context, maintains matter memory, and applies your standards automatically.

With HAQQ, you do not need to specify your role in every prompt because the platform already knows you are a lawyer. You do not need to re-explain your risk thresholds because they are built into the system. You do not need to worry about data security because the platform is designed with sovereign, encrypted infrastructure that keeps your data under your control.

The result is that your prompts become shorter, faster, and more effective. You focus on the task, not the setup. And the output you receive is not raw text but structured, exportable legal work product ready to send to clients.

Building Your Prompt Library

The most productive legal teams do not write every prompt from scratch. They build libraries of tested, refined prompts for recurring tasks. A prompt that works well for NDA review today will work well tomorrow, next week, and next month, with minor adjustments for each specific matter.

HAQQ offers a free Prompt Library with over 168 ready-to-use legal prompts across practice areas including contract review, due diligence, compliance, risk analysis, and legal research. Each prompt is structured using the Role + Context + Task + Format framework and can be customized for your specific needs.

From Prompts to Practice: Making AI Part of Your Legal Workflow

Mastering prompts is not the end goal. It is the beginning. The real transformation happens when AI-driven prompting becomes embedded in your daily practice, not as an experiment but as a core part of how you deliver legal services.

Start with low-risk, high-frequency tasks: summarizing contracts, drafting standard clauses, checking compliance checklists. Build confidence. Refine your prompts. Then expand to more complex work: multi-document analysis, risk assessment across portfolios, due diligence coordination.

The firms that will lead in the next decade are not those with the biggest teams. They are those that combine legal expertise with AI mastery. Prompting is the skill that bridges the two.

Key Takeaways

• Prompt quality determines AI output quality. Invest time in crafting structured, context-rich prompts.
• Use the formula: Role + Context + Task + Format for every legal prompt.
• Break complex tasks into sequential steps instead of single monolithic prompts.
• Iterate and refine. No prompt is perfect on the first try.
• Avoid generic AI tools for sensitive legal work. Use purpose-built platforms with proper security.
• Build a prompt library for recurring tasks. Reuse and refine over time.
• Explore HAQQ's free Prompt Library with 168+ tested legal prompts to get started immediately.

Related reading

• we ran the same NDA prompt on ChatGPT and a purpose-built legal AI
• how large language models actually work (no CS degree required)
• we tested 3 models on 100 real legal questions
• 3,000 graded answers from 10 frontier models