The legal sphere is undergoing a digital transformation. Today, lawyers not only consult clients but also create their own services: bots for preliminary document analysis, contract generators, and automated consultants. The Botman.one platform allows you to implement these ideas without deep programming knowledge, using a low-code approach and artificial intelligence capabilities.
However, the key to the success of any AI service is not only the technology but also the quality of the requests (prompts) you send to the neural network. An error in a prompt can cost you your reputation or lead to legal risks.
Based on the fundamental principles of prompt engineering, we have prepared a guide for lawyers on how to correctly compose requests for legal services on Botman.one.
1. Set Direction
The neural network cannot read minds. If you simply ask it to "draft a contract," it will produce an averaged result that may not account for the specifics of your jurisdiction or task.
How it works in law: Instead of an abstract task, clearly define the role, style, and context.
- Bad: "Write a claim to the supplier."
- Good: "Act as a senior lawyer in the field of arbitration law of the Russian Federation. Draft a pre-trial claim due to violation of delivery deadlines under Supply Contract No. 123. Style: formal business, strict, with references to the Civil Code of the Russian Federation."
Tip for Botman.one: Use system instructions in the bot settings to fix the AI's role (e.g., "Intellectual Property Lawyer Assistant"). This will set the vector for all service responses.
2. Specify Format
Legal services often require structured data. Chaotic text is difficult to integrate into a bot interface or automatically insert into a document template.
How it works in law: Clearly describe what the answer should look like.
- For documents: "The answer must contain sections: Header, Facts, Legal Basis, Demands."
- For integration: "Return the answer in JSON format, where keys are: clause_title, clause_text, risk_level."
Request Example:
text
1
2
3
4
5
6
This will allow your bot on Botman.one to beautifully display risks to the user or automatically save them to a database.
3. Provide Examples
The method known as "few-shot learning" significantly improves accuracy. The neural network understands the task better when it sees samples of correct solutions.
How it works in law: Instead of explaining in words what a "good formulation" is, show it.
- In the prompt:
text
1
2
3
4
5
6
7
8
9
10
11
This is especially useful for contract review services (Due Diligence), where it is important to maintain a single analysis standard.
4. Evaluate Quality
In legal practice, the price of an error is high. You cannot rely on the neural network's first answer. The quality evaluation principle requires implementing verification processes.
How it works in law:
- A/B Testing: Create two versions of a prompt on Botman.one for one task (e.g., one with examples, one without) and compare the results.
- Human Review: Implement a stage in the service where the AI's answer is marked by a lawyer ("Correct/Incorrect").
- Fighting Hallucinations: AI can invent law articles. Be sure to add this instruction to the prompt: "If you don't know the exact article, do not invent it, but write 'Requires Clarification'."
Toolset: Use the platform's logic to save the history of requests and responses. This will allow you to conduct an audit of your AI service's performance later.
5. Decompose Tasks
Complex legal tasks cannot be solved with a single request. Trying to make the AI immediately "analyze the case and make a decision" will lead to a superficial answer.
How it works in law: Break the process into a chain of steps (Chain of Thought):
- Step 1: Extracting facts from the client's text.
- Step 2: Finding applicable legal norms.
- Step 3: Applying norms to facts.
- Step 4: Forming the final conclusion.
Implementation on Botman.one: You can configure the bot scenario to sequentially call different AI modules. First, the bot asks for details (Step 1), then sends them for analysis (Step 2), and only then forms the conclusion. This increases accuracy and allows the user to see the logic of reasoning.
Security and Ethics
Working with legal data, remember confidentiality. Do not transfer clients' personal data or trade secrets to public models without appropriate processing. The Botman.one platform provides tools for secure integration, but the responsibility for the content of prompts lies with the developer.
Artificial intelligence will not replace a lawyer, but a lawyer using AI will replace one who does not. Following the five prompting principles will allow you to create reliable, accurate, and useful legal services on Botman.one.
Start small: take one routine task (for example, generating a standard contract) and apply these principles to it. Evaluate the result, improve the prompt, and scale the solution.