HAQQ

Prompt Engineering for Lawyers: 7 Principles That Hold Up

By Issam Amro · · 9 min read · guides

Seven prompt engineering principles for lawyers — context, role assignment, few-shot examples, verification — that produce accurate, defensible legal output.

There is a new competency in legal practice, and it belongs alongside legal research, drafting, and advocacy: the ability to communicate precisely with AI. Prompt architecture — the art and science of crafting instructions that guide AI tools to useful, accurate, and defensible outputs — is fast becoming as important as knowing where to find the law.

Key facts

  • "Uncritical reliance on AI output without independent verification may breach the duty of competence" (ABA position as summarized in the article).
A lawyer must know what they want, before they can ask the AI for it, in order to get the best answer. Vague questions produce vague answers. Ambiguous instructions produce unreliable outputs.

Why Prompt Architecture Matters in Law

Legal work demands a level of precision that generic prompting cannot achieve. A prompt that produces an adequate response for a marketing team may produce a dangerously incomplete one for a litigator preparing submissions. Skilfully constructed prompts generate more accurate, efficient, and defensible results across research, drafting, document review, and discovery workflows.

They also help manage compliance risk — because a well-constructed prompt tells the AI not just what to produce, but how to produce it, in what format, and with what caveats.

1. Establish Context First

AI tools do not know your matter unless you tell them. Always open a prompt by establishing the relevant jurisdiction, area of law, type of matter, and the role you want the AI to play.

You are a legal research assistant. The jurisdiction is England and Wales.
The area of law is employment law.
I am representing a claimant in an unfair dismissal matter.

2. Be Specific About the Output You Need

Specify the format, length, and level of detail required. Do you need a structured memo? A list of key authorities? A draft clause? A risk summary? The more precisely you define the output, the more useful the response will be. Generic requests produce generic answers.

3. Provide the Relevant Facts and Documents

Do not ask an AI to analyse a situation it cannot see. Upload the relevant contract, judgment, or statutory provision. Tell the AI the material facts. AI performs best when it works from the actual documents in front of you, not from its general training data.

4. Use Role Assignment

Assigning the AI a specific expert role — "Act as a senior barrister reviewing this statement of case for procedural weaknesses" — significantly improves output quality. Role assignment activates relevant training patterns and encourages the AI to respond with appropriate domain-specific rigour.

5. Apply Iterative Refinement

Do not expect perfection from a first prompt. Evaluate the initial response, identify what is missing or imprecise, and refine. Ask follow-up questions. Probe inconsistencies. This back-and-forth approach — known as iterative refinement — is one of the most powerful techniques available to legal AI users.

6. Use Few-Shot Examples

For complex or nuanced tasks, provide the AI with one or two examples of the type of output you want before asking it to produce its own. This few-shot learning technique is particularly effective for drafting specific clause types, identifying patterns in case law, or analysing contract language with a particular standard in mind.

7. Always Verify the Output

Prompt architecture is not about outsourcing judgment — it is about directing it. Every AI output must be reviewed and verified by a qualified lawyer before it is relied upon, filed, or communicated to a client. The ABA has made clear that uncritical reliance on AI output without independent verification may breach the duty of competence. The lawyer signs the document; the AI does not.

  • Specify jurisdiction and governing law in every research prompt
  • Define the audience — is the output for a client letter, internal memo, or court submission?
  • Ask for sources — instruct the AI to cite the authorities it relies on, then verify them independently
  • Break complex tasks into steps — rather than one long prompt, use a structured sequence of focused prompts
  • Develop prompt templates for your firm's most common tasks — contract review, research memos, due diligence checklists
  • Review the data-handling process of a tool before uploading privileged material
At HAQQ, we believe that access to powerful legal AI is only half the equation. Knowing how to use it — precisely, ethically, and strategically — is what separates the firms that will lead the next era of legal practice from those that will be left behind.

FAQ

What is prompt architecture for lawyers?

Prompt architecture is the structured discipline of designing AI prompts so they produce accurate, consistent, defensible legal outputs. It goes beyond ad-hoc 'prompt engineering' tips: role assignment, context setting, structured output requirements, citation requirements, verification steps and iterative refinement.

What are the seven principles of legal prompt architecture?

The core seven are: 1) assign a precise legal role, 2) set jurisdiction and matter context up front, 3) provide the source material rather than asking the model to recall, 4) specify the output structure, 5) require explicit citations, 6) ask for the model's uncertainty, 7) refine iteratively rather than expecting the first answer to be final.

Is prompt architecture different from prompt engineering?

Yes. Prompt engineering is the tactical layer (the wording of a single prompt). Prompt architecture is the strategic layer (how prompts are designed across a workflow, with verification, source grounding and output structure baked in). Lawyers benefit more from architecture than from clever wording tricks.

Can I use the same prompts across ChatGPT, Claude and Gemini?

Mostly yes, with small adjustments. The seven principles transfer across models. Claude tends to be more literal with instructions, GPT more conversational, Gemini more concise. Test the same prompt on your target model before relying on it for client work.

Do prompt architecture principles work for purpose-built legal AI like HAQQ?

Yes - and they matter less because the platform handles much of the architecture for you. A purpose-built legal AI bakes role, jurisdiction, citation grounding and output structure into the product. The remaining lawyer skill is precise question framing and review of the output.

How long does it take to learn legal prompt architecture?

A focused lawyer can internalise the seven principles in an afternoon and reach professional fluency within a few weeks of daily use. The hardest part is not the techniques - it is the discipline of verifying every citation and never shipping AI output without lawyer review.

← All HAQQ articles