Step-by-step sharing on how legal professionals can make good use of AI — Prompt

Pan Yang is a partner at Merits & Tree Law Firm

Hello~ Hello everyone, this is Pan Shuai sharing how legal people can make good use of AI series, and we look forward to discussing the practical experience of legal technology, AI and digital law firms, and making friends with you.

As a legal technology explorer, based on the recent sharing and summary within the law firm, I have compiled a series of materials on the use of large AI models, and combined with my own experience, I wrote this article, many of which are derived from personal practice and the exploration of practical operations in the firm. It should be noted that AI technology is developing very rapidly, and the following views only represent personal experience at the current point in time.

In today's era of artificial intelligence, AI-related large language models have become indispensable assistants in all walks of life, and in order to truly realize human-machine collaboration, lawyers must first understand the advantages and disadvantages of AI, and at the same time recognize the strengths and weaknesses of the lawyer group, so as to better achieve the collaborative division of labor between lawyers and large models.

01 How lawyers and AI can work better together

What large AI models are good at

1. Information retrieval and collation:

The AI model can quickly retrieve relevant information from a large amount of data, such as the search of laws, regulations and cases, and extract and organize case-related materials.

2. Pattern Recognition and Prediction:

By importing historical data and reference information, and setting instructions, the AI model can predict the possible outcome of the case, such as the trend of judgments and the amount of compensation, to assist in formulating litigation strategies.

3. Automated Document Processing:

Large AI models can automatically generate and revise standardized texts and contracts, reducing lawyers' workload in drafting and revising documents.

4. Ability to multitask:

Large AI models can handle multiple tasks at the same time, without the limitation of time and physical strength, which can greatly improve the efficiency of processing basic tasks at the same time.

What large AI models are not good at

1. Legal Interpretation and Reasoning:

Large AI models can make some predictions and judgments based on historical information and data, but it is still difficult to deduce complex legal interpretations and arguments like professional lawyers. In particular, the capabilities of AI are very limited when it involves intersecting multiple areas of law or when it is necessary to interpret the law in a deep social context.

2. Understand morality and emotion:

Large AI models struggle to understand the complex emotions and motivations involved in a case, such as the multiple relationships that can arise between a couple in a divorce case.

3. Innovative or Personalized Service:

It is difficult for AI large models to provide personalized legal services based on the comprehensive needs of customers, because large models are based on preset data and rules, and cannot collect all the instant information of customers in a timely manner, and it is difficult to generate innovative and professional answers beyond the content of the corpus, so it is difficult to provide professional services to customers in a targeted manner, even if it is based on the same cause or case.

What lawyers are good at

1. Legal expertise: Lawyers have in-depth legal knowledge and are able to provide professional legal analysis and advice, such as professional analysis in securities fraud cases.

2. Communication and negotiation: In the process of communication and negotiation, lawyers can build trust, express opinions, and facilitate transactions with clients, counterparties, and other institutions.

3. Creative Problem Solving: Lawyers are able to provide compliance advice on emerging industries or regulatory gaps, such as developing legality guidance for new technologies.

4. Crisis response: As professionals, lawyers need to be able to make professional judgments and provide timely legal advice and solutions in emergency situations.

​

What the lawyer is not good at

1. Processing a large amount of information and data: In the case of processing a large amount of text and data, the efficiency of the lawyer's manual labor is very limited, such as data extraction and collation in a large number of document investigations.

2. Dealing with details: It may be difficult for a lawyer to remember all the facts and details in various types of cases, such as: time, person, amount, relationship, especially in complex cases.

3. Energy and emotion: When dealing with complex cases, lawyers may face pressure from emotions, energy, time, etc., which may affect professional judgment. Based on the above, the collaboration between lawyers and AI is not simply additive, but a relationship that leverages each other's advantages and strengthens each other. When lawyers use the powerful tool of large models, the most critical task is how to accurately ask questions and instructions according to different legal business scenarios to guide AI to maximize its effectiveness.

​

When dealing with information retrieval and sorting tasks, lawyers can guide AI to accurately capture key information such as relevant laws and regulations, precedent judgments, etc., so that they can quickly obtain the materials needed for case preparation and spend more time on legal analysis. When automated document processing is required, lawyers can guide AI to generate and modify standardized contracts. For example, give the prompt "Automatically generate a draft contract on matters XX based on the template provided, and mark out the key terms that need to be manually reviewed". In this way, the lawyer can significantly reduce the workload of drafting and revising documents while maintaining the quality of the contract. At the same time, in the handling of cases involving complex legal relationships and cutting-edge nature, lawyers need to rely more on their own professional analysis, give full play to their communication skills to obtain detailed information and promote the progress of the case. At this time, using AI to assist in sorting out the facts and evidence of the case can reduce a lot of work time.

02 How a lawyer writes a good prompt

1.什么是 Prompt ？

A prompt is information or questions that are provided to an artificial intelligence (AI) system to guide the AI to produce a specific response or perform a specific task. Just like the questions you ask in a chat, a Prompt is the starting point for starting a conversation or guiding the direction of the conversation.

For AI, a good Prompt can help it understand your needs more accurately and give more relevant and useful answers. In short, a prompt is like a word you say to an AI that tells it what you want it to do.

2. Suggested framework and format for Prompt

First: CRISPE

Capacity and Role（能力与角色）

Insight（洞察）

Statement（陈述）

Personality（个性）

Experiment（举例）

Example:

Capacity and Role，你希望它的角色和能力：

For example, you are a lawyer specializing in the field of civil and commercial law, and you are good at case studies, legal text searches, and case strategy analysis. By giving AI such a role, we can make more effective use of its data processing and pattern recognition capabilities, which can improve the productivity of lawyers.

Insight, which provides context and context:

For example, in a complex contract dispute case, we can provide AI with the key facts of the case, the relevant laws, and the background of the case. In this way, AI can understand the ins and outs of the case in advance when outputting answers, providing more accurate assistance to lawyers.

Statement，你希望AI做什么：

For example, make it clear what the AI is expected to accomplish. Taking a contract dispute case as an example, we can ask the AI to summarize the claims of both parties in the case, search for legal provisions, and predict the possible judgment outcome.

Personality，你希望AI以什么风格或方式回答你：

For example, it is specified to mimic the logical rigor and concise style of a senior lawyer. In this way, when AI provides information, it will be more in line with the communication and expression habits of professional lawyers.

Experiment, which asks the AI to provide you with several examples:

For example: "Please give at least three different litigation strategies for this case and analyze the advantages and disadvantages of each." "This kind of directive helps the lawyer to understand the information more comprehensively, and this kind of directive helps the lawyer to sort through multiple ideas and provide more ideas for solutions.

​

The second: PEMSSC

Personality（个性化的风格）

Example Inquiry（给参考、或一定的逻辑结构）

Multiple Angles（多个角度来思考）

Summarize（总结概括）

Separator (distinguish: quotation marks, separators, first, second, last, etc.)

Capacity and Role（能力或者角色）

Example:

Personality

For example, we can choose a style that is humorous and insightful, while incorporating someone's innovative perspective. Such a style not only makes boring legal analysis interesting and interesting, but also inspires clients to think about problems from a new perspective.

Example Inquiry

For example, when providing legal advice to clients, we can use logical structures such as SWOT analysis or the 4P principles. For example, use a SWOT analysis to assess the strengths, weaknesses, opportunities, and threats of a business contract to help clients close smarter deals.

Multiple Angles

When faced with complex legal issues, we need to think from multiple perspectives. For example, when analysing a commercial litigation, we can consider the litigation strategy from multiple perspectives such as market trends, competitor behavior, strategic planning, financial situation, and market outlook to ensure that we provide comprehensive and in-depth legal advice to our clients.

Summarize

When providing legal advice, we need to summarize the core points and considerations in clear and concise language. For example, we can summarize 5 core points about case handling and list 7 points that need special attention in the progress of the case.

Separator

To make sure our suggestions are clear and easy to understand, we can use quotation marks, separators, and conjunctions like "first, second, last" to organize Prompts. Not only does this make recommendations more organized, but it also gives you better information through AI.

Capacity and Role

For example, you are a lawyer specializing in the field of civil and commercial law, and you are good at case studies, legal text searches, and case strategy analysis. By giving AI such a role, we can make more effective use of its data processing and pattern recognition capabilities, which can improve the productivity of lawyers.

3. Prompt 方法总结

Format = [Set Roles + Objectives + Context and Background + (Positive Requirements) Detailed Needs and Details + (Negative Requirements) Restrictions and Unwanted Content + Language Style and Form of Answers]

4.Prompt 技巧

Be clear about the context and purpose

When asking AI questions, in addition to a clear description of the question, it is best to sort out the background information and the purpose of the question, which can help the AI better understand the context of the question and improve the accuracy of the answer. For example, when a lawyer is handling a traffic accident case, he can ask: "If you are given a traffic accident case fact xxx, how should the liability of party X be divided according to the xxx regulations?"

Learn to ask questions and how to improve the accuracy of your answers

Asking good questions is key to improving the accuracy of your AI's answers. This includes using clear, specific language and avoiding vague expressions. It's also important to understand how AI works and its limitations, so you can better design your questions so that they can provide useful answers.

Dismantle the link, cut the process, and fall into a certain work detail

Before applying AI, you need to do a meticulous teardown of your workflow. This means breaking down complex tasks into smaller, more specific pieces so that AI can perform them more precisely.

For example, if you're using AI for data analysis, you may need to separate data cleansing, data extraction, model selection, model training, and result interpretation. The benefit of this is that you can optimize the performance of your AI for every part, while also making it easier to identify and fix issues.

There is also a way for complex issues where the lawyer can ask questions in a step-by-step and detailed manner. Start with a broad question, then refine or deepen the question based on the AI's responses. This approach helps the lawyer to gain a deeper understanding of all aspects of the issue step by step.

For example, when dealing with an IP infringement case, a lawyer can first ask, "Is the defendant infringing in this case?" and then ask further based on the AI's response: "If so, what is the type and extent of the infringement?"

What to refer to and learn from the AI allows him to understand the structure and learning, write the process, and write the knowhow

AI systems often require large amounts of data and examples to learn and understand the structure of the task. Providing high-quality reference materials and learning content is key to improving AI performance. This may include detailed how-to guides, industry best practices, case studies, and more. At the same time, it's important to write detailed processes and knowhows, not only to help AI better understand tasks, but also to provide guidance to human users. For example, in Automated Document Processing, you can write a detailed guide on how to work with different types of documents and how to use AI tools to be more efficient.

Utilize terminology in your area of expertise to guide

Use legal jargon in Prompt to steer the AI's answers. For example, when dealing with contract disputes, the lawyer can prompt: "Analyze the performance of the contract from three aspects: contract signing conditions, exclusive cooperation and liability for breach of contract." "This kind of guidance helps the AI provide the information it needs more precisely.

Validation & Feedback

There is a certain lag in the corpus of large models, and after using AI answers, lawyers must cross-verify the content to ensure the accuracy of the information. At the same time, when lawyers use AI, they should also combine their own professional knowledge to guide them. Screening and judging AI's responses are screened and judged through professional knowledge to ensure that they are in line with mainland legal ethics, legislative purposes, and practices.

More Prompt reference learning content: Through WeChat Official Account, Zhihu, Baidu, Secret Tower Search, Copilot, Search: How to Write a Good Prompt

03 律师常用 Prompt 场景

The art of Prompt is how to communicate effectively with AI with precision. In the legal realm, an effective prompt can lead AI to deliver more precise and relevant output.

Scenario 1: Case retrieval

Case retrieval is best done with AI products in the legal industry, as general-purpose AI either can't find out, the data is incomplete, or the generative content can't meet the needs.

1.Prompt Command Words:

Search for specific cases from the last five years on the "likelihood of confusion" criterion in trademark infringement cases, and provide a summary of the key takeaways from the three most similar cases.

2. Expected Effect:

The AI system will search relevant databases to identify cases that meet the requirements, and extract the key reasons and results of the "likelihood of confusion" test in these cases to form a summary report.

3. Other examples:

（1）Prompt指令词：

Search all cases involving software algorithm patent infringement in the past three years, analyze the comparison of technical features and the criteria for infringement determination in court judgments, and provide reference for upcoming patent infringement cases.

（2）Prompt指令词：

Compare the different judgments of courts in different regions on the calculation standards of overtime pay when handling labor dispute cases, summarize the trend of judgments in favor of employers, and provide customers with compliance guidance.

（3）Prompt指令词：

Research on environmental law related cases, especially legal liability involving industrial waste disposal, and provide clients with compliance treatment advice to reduce potential legal risks.

4.Promopt结构提示:

[Case field or type + clarify the key content that needs to be found + the purpose of finding the case + other things you want the AI to do]

Scenario 2: Search for similar cases

Case retrieval is best done with AI products in the legal industry, as general-purpose AI either can't find out, the data is incomplete, or the generative content can't meet the needs.

1. Prompt指令词：

An employee of a company posted an article criticizing the company's management on the company's intranet, causing dissatisfaction among the company's top management. The company dismissed the employee on the grounds that he had violated the company's rules and regulations. The employee believes that the company's actions violated his rights and takes the company to court.

Based on the above, please recommend me 5 similar cases

2.Prompt Command Words:

Based on the following cases, please recommend 5 similar cases to me.

3.Prompt结构提示：

[Extract case facts (you can also let the AI refine and summarize first) or send the case directly to the AI, and let the AI recommend similar cases based on the above facts]

Scenario 3: Reading legal texts

Text reading is based on the function of AI text uploading, and several mainstream companies already have similar functions. However, not all products have long or multi-text content.

1. Operation action:

Select one or more documents to upload, AI will parse them, and then ask questions based on what you know

2. Examples:

Prompt Command: Who designed the drawing?

Prompt Instruction Word: Who is responsible for the design and construction of this project?

Prompt Instruction Word: What is the loan term?

Prompt Dictator: Who Pays Attorney Fees?

Prompt Instruction: Please summarize the plaintiff's claim and factual grounds?

Prompt Instruction: According to the civil reply, does the defendant Zhang San have a better defense proposal?

You can also use instructions to splic, and send the complete instructions to the large model.

3. The splicing template is as follows:

Answer my questions based on the following material content [each material content starts with a numeric serial number + file name]

You need to refer to the content of the material as much as possible: [Document 1 contents] + [Document 2 contents] + ... + [Document n contents] +

My question is: refer to the Prompt directive word above

Scenario 4: Research and interpretation of laws and regulations

Regulatory search is best done using AI products in the legal industry, and general-purpose AI either can't find out, or the data is incomplete, or the generative content can't meet the needs.

1. Prompt 指令词：

According to the latest revision of the Data Security Law, the main principles that enterprises should follow when processing personal information are analyzed, and specific operational suggestions are given.

2. Expected Effect:

Based on the latest provisions of the Data Security Law, the AI assistant will analyze the principles that enterprises must follow in processing personal information, and provide detailed operational guidelines or compliance recommendations based on the actual situation.

3. Other examples

（1）Prompt 指令词：

Analyze the specific impact of the new Cybersecurity Law on e-commerce platforms, especially the provisions on data protection and user privacy, and provide customers with compliance guidelines.

（2）Prompt 指令词：

Enter a specific legal provision such as international trade regulations, which is the latest changes in international trade regulations to provide customers with compliance guidance for cross-border transactions, especially in terms of tariffs and import and export restrictions.

（3）Prompt 指令词：

Interpret the impact of the newly revised tax law on SMEs, provide advice on tax planning and tax avoidance strategies, and ensure that clients optimize their tax liabilities within the legal scope.

4.Promopt结构提示:

[Requirement or purpose + according to a specific law or a law + specific needs or specific research details]

Scenario 5: Legal opinion writing or non-litigation transaction documents

1.Prompt Command Words:

For the copyright dispute case involving our party, enter [existing evidence materials] + [relevant legal provisions] to write a preliminary legal opinion to demonstrate the reasonableness of our claims and the possibility of winning the lawsuit.

2. Expected Effect:

Based on the case background, evidence materials, and applicable laws and regulations, the AI will automatically write a preliminary legal opinion, which includes a review of the facts of the case, legal analysis, and conclusions.

3. Other examples

（1）Prompt 指令词：

For a trade secret leakage case, analyze the possible legal liability and compensation scope, write legal opinions, and provide clients with coping strategies.

（2）Prompt 指令词：

Advising a company planning to go public on its corporate governance structure, including shareholder protection and board responsibilities, to ensure compliance with the requirements of the Companies Act and securities regulations.

（3）Prompt指令词：

Providing legal advice on a consumer rights protection case, analyzing the liability and consumer rights protection methods that merchants may face, and formulating countermeasures for clients.

（4） Prompt指令词：

Drafting an equity transfer agreement for a company's equity transfer transaction, ensuring that the agreement details the information of the transferor and the transferee, the share of the equity transfer, the transfer price, the payment method and timetable, the conditions precedent to the equity transfer, and the rights and obligations of both parties after the completion of the transaction. At the same time, it proposes provisions on confidentiality, liability for breach of contract and dispute resolution.

（5）Prompt指令词：

Drafting an agreement involving the transfer of equity in a start-up company, specifying the conditions, price, and payment method for the transfer of equity by the founding shareholder to the new investor, as well as the corporate governance structure and decision-making process after the new investor joins.

（6）Prompt指令词：

Drafting an agreement for a transaction involving the transfer of shares of a listed company, including a detailed description of the shares, the transfer price, payment arrangements, regulatory requirements for the transfer of shares, and the impact on the control of the company after the transfer.

（7）Prompt指令词：

Drafting an agreement involving the internal transfer of equity in a family business, ensuring that the terms cover the distribution of equity among family members, how the transfer price is determined, the adjustment of the governance structure of the family business, and the protection of the rights and interests of family members after the equity transfer. At the same time, it proposes how to deal with the succession and long-term development plan of the family business.

4.Prompt Command Words:

Step 1:

Provide [case description], I would like you to understand the circumstances of the case

Step 2:

According to the following legal opinion template, generate a legal opinion for the above content, and provide [document template]

5.Promopt结构提示:

[Case background + description of facts and details + position and purpose + key reference to the law or a certain point of view + text format]

Other scenarios to try: Sentencing Prediction (Criminal)

1.Prompt Command Words:

For a criminal case involving commercial bribery, the possible sentencing results under different defense strategies are simulated, and the best defense strategy is compared and recommended.

2. Expected Effect:

AI will predict the possible verdict under different strategies based on the verdicts of similar cases in the past, the current criminal law regulations, and a variety of possible defense strategies, and help lawyers formulate the optimal defense plan.

3. Other examples:

（1）Prompt指令词：

For an online fraud case, we simulated the sentencing outcome under different defense strategies, including the possibility of plea negotiation and not guilty defense, to provide the best defense plan for the client.

（2）Prompt指令词：

Designing a litigation strategy for a commercial contract dispute case, including selecting the appropriate court, gathering key evidence and preparing witness statements to improve the chances of success.

（3）Prompt指令词：

Simulate a court debate involving intellectual property infringement, analyze the possible arguments and rebuttal strategies of both the plaintiff and the defendant, predict the possible outcome of the case, and provide strategic advice to clients.

Other scenarios to try: litigation strategies

1.Prompt Command Words:

For a case involving trademark infringement, analyze the scope of the plaintiff's trademark rights and the defendant's potential infringement, and propose litigation strategies, including the asserted claims, the focus of evidence collection, possible legal defenses, and the possibility of settlement or mediation.

2. Expected Effect:

AI will provide litigation strategies for trademark infringement cases based on the relevant provisions of trademark law and case law. This will include a detailed analysis of the plaintiff's trademark rights, a legal assessment of the defendant's infringement, recommendations for the collection of key evidence, possible legal defences, and strategies for settlement or mediation where appropriate. The goal of AI is to help lawyers construct a comprehensive litigation plan to improve their chances of winning the case and maximize the interests of their clients.

3. Other examples:

（1）Prompt指令词：

Developing litigation strategies for a patent infringement case, analyzing the validity of the patent, the defendant's infringement and possible defenses, and making recommendations on how to prove infringement and calculate damages.

（2）Prompt指令词：

For an employment contract dispute case, analyze the rights and interests of employees and the responsibilities of the employer, and propose litigation strategies, including the rights and interests claimed, the focus of evidence collection, and how to prove the employer's breach of contract.

（3）Prompt指令词：

In a litigation involving intellectual property licensing, analyze the terms of the licensing agreement and the rights and obligations of the parties, and propose litigation strategies, including how to prove the breach of the licensing agreement, calculate damages, and possible conditions for termination.

Other scenarios to try: Moot courts

1.Prompt 指令词：

Simulate a court trial involving the breach of a commercial contract, analyze the plaintiff's claim for breach of contract liability and the defendant's defense of liability, predict the court's possible judgment, and make suggestions on how to optimize the court's presentation and evidence presentation.

2. Expected Effect:

The AI will simulate the court proceedings in accordance with the relevant provisions of the Commercial Contract Law and case law. The AI will provide an analysis of the possible arguments, evidence, and legal basis for both the plaintiff and the defendant, predict the likely outcome of the court's decision, and give recommendations on how to more effectively present the case, present evidence, and conduct cross-examination in court. This will help the lawyer prepare a more persuasive court strategy with a view to achieving the best possible outcome for the lawsuit.

3. Other examples:

（1）Prompt指令词：

Simulate a courtroom debate involving intellectual property infringement and analyze how the plaintiff can prove the validity of their intellectual property rights and the defendant's infringement, as well as the lawful use or fair use defenses that the defendant may raise.

（2）Prompt指令词：

In a corporate law litigation involving shareholders' interests, mock court arguments are conducted to analyze how shareholders can prove misconduct by the company's management and how the company can defend the legality of its decision-making process and results.

（3）Prompt指令词：

Simulate a labor law lawsuit and analyze how an employee can prove that an employer has violated an employment contract or labor regulations, and how the employer can defend that his or her actions are in accordance with the law or the contract.

PS: Please simulate [Plaintiff] and [Defendant] to express their views separately

Due to the rapid development of AI, as users, we may not be able to use AI and learn AI tools as fast as AI iteration, so it is very likely that the above sharing will be replaced by advanced technology in a short period of time, but the thinking, scenarios and logic should be able to give you some inspiration.

When I was studying in law school, I read a sentence that I thought was true, which just confirms the current era of law + AI:

"The law is nothing but a record of our will. —Rousseau, The Social Contract

— To be continued —

#让AI触手可及#