Skip to main content

Prompting Techniques

This guide will help you improve your prompts to get optimal results in Legora.

Klara Folke avatar
Written by Klara Folke
Updated over a week ago

Core Principles

Focus your prompts on the specific task, relevant context (the "why"), and desired output format.

The Foundation: Clear, Explicit Instructions

Legora excels when getting precise prompts. Make sure to include:

  1. Objective: Be explicit and specific about your desired outcome.

  2. Context: Provide the context and explain why certain aspects matter.

  3. Output: Define format with examples and structural details.

Universal Tips

  • Request step-by-step reasoning: To break down complex tasks.

  • Match your prompt writing style to the output: Formal prompts typically yield formal outputs.


Assistant

Best Practices

DO:

  • Be explicit: Clear, specific instructions yield better results.

  • Explain the "why": Providing context helps the model understand your goals.

  • Request comprehensive output: Add "Include all relevant details" or "Be thorough".

  • Use step markers: Number your steps (1, 2) for multi-part tasks.

  • Attach supporting materials: Include relevant documents (e.g., emails, PDFs) needed to complete the task effectively.

DON'T:

  • Write “I am a lawyer at a law firm”. Legora is designed with lawyers in mind and automatically recognises legal context. Instead, focus on clearly defining the roles of each party involved and their perspectives, and explicitly state which party you represent.

  • Mix unrelated or too-big tasks in one prompt. Instead, try to break it into multiple, smaller, explicit steps.

  • Use vague instructions like "review this document" or have a too broad scope for the task.

Example Prompt

Objective:

Generate a structured, actionable checklist based on the attached SPA to help clients monitor their obligations and responsibilities throughout the signing-to-closing phase. The checklist should align with the order of the provisions in the SPA and highlight the key tasks and milestones in a client-friendly format.

Extract and summarize all obligations between signing and closing from the SPA. For each item, extract:

• A clear description of the obligation

• The required action(s)

• The relevant deadline or timeframe

• The responsible party or parties

• A reference to the specific clause in the SPA

Context:

Our M&A team frequently advises on Share Purchase Agreements (SPAs), which involve detailed legal documentation setting out the key conditions and obligations to be fulfilled between signing and closing. While legal practitioners are familiar with these requirements, clients often find them complex and difficult to manage. Tracking pre-completion covenants, regulatory approvals, and closing conditions is essential to ensuring a timely and successful transaction.

Output:

Please ensure the closing checklist is organized into three tables.

1. Pre-Closing Obligations

2. Checklist – Preparation of Seller's Closing Deliverables

3. Checklist – Preparation of Purchaser's Closing Deliverables

Present the information in three structured tables with the following columns:

- SPA Reference: The clause number or heading in the SPA where the obligation appears (e.g., Clause 5.2(b) or Section 7: Pre-Closing Obligations)

- Obligation / Action: The set of actions that need to be performed between the signing and the closing

- Deadline: The deadline of the obligation or action, in accordance with the SPA - Responsible Party: Who is responsible for performing the action (e.g., Seller, Buyer, Both Parties, Third-Party Advisor)


Assistant - Database Search

Best practice

DO:

  • Front-load key terms: Start with the most specific legal terms.

  • Specify extraction needs: “What is the calculation methodology for earn-out provisions in SPAs?”.

  • Chain searches: "First search for UK loan agreements, then find negative pledge clauses for that agreement".

  • To make sure you find all the documents you’re looking for: Write, e.g, “Find all cobranding contracts in the project. Keep searching repeatedly until you can't find any cobranding contracts anymore before you read the documents you found.”

DON'T:

  • Use broad terms that will return too many results.

  • Use words like “all”, “best”, “highest”, “most”. Try to be more explicit to get better results.

    • If you want to do an exhaustive search that requires reviewing and comparing all documents in a database, use Tabular Review for the task.

  • Forget to specify the context, e.g, jurisdiction or industry, when relevant.

Example Prompt

Objective:

I want to understand how Legora protects user data.

Context:

I’m reviewing Legora’s data protection practices and need a clear explanation of the security or privacy measures mentioned across all relevant policies.

Output:

Provide a concise summary (bullet points) of the data protection practices outlined in Legora policies, including but not limited to privacy, security, and compliance policies. Focus on what specific actions or safeguards Legora takes to protect user data.


Assistant - Legal Research

Best practice

DO:

  • Be specific about jurisdiction: "UK Court of Appeal decisions on..." or "EU regulations regarding...".

  • Specify wanted document type: E.g, court case, supreme court case, law.

  • Include relevant legal concepts: Use proper legal terminology and doctrines.

  • Specify time periods: "Post-2020 cases dealing with..." or "Current legislation as of 2025".

  • Pick the correct source of jurisdiction: E.g, the UK.

DON'T:

  • Expect comprehensive legal advice—Legal Sources provides research, not opinions.

  • Assume all jurisdictions are equally covered.

  • Assume Legora can find “all” of something, to get the best results, try to be more explicit, e.g, “Court cases referencing law… “.

  • Assume Legora knows the importance of cases. Instead, specify what type of cases are authoritative and important for this task.

  • Request specific output formats. Legal research results are presented according to a legal hierarchy, so requesting specific output formats such as tables or custom subheadings is generally not recommended.

Example Prompt

Objective

Identify and analyse risk factors disclosed in prospectuses issued by companies in the Swedish real estate sector, to support a structured understanding of company-specific risks.

Context

Review prospectuses from companies in the real estate sector that are available in the Swedish Financial Supervisory Authority’s (Finansinspektionen) prospectus register. Identify and analyze 5 risk factors.

The scope of your analysis should cover Company-Specific Risks – e.g., risks related to the business model, financial health, leadership, or strategic decisions.

Output

Present the risks found.


Assistant - Web Search

Best Practices

DO:

  • Layer your searches: Start broad → narrow → extract specific insights

  • Request synthesis: "Compare and contrast the approaches between …"

  • Specify recency: "Focus on developments from Q1 2025 onwards"

  • Specify region of interest: E.g., "inheritance law in Sweden" instead of "inheritance law"

  • Write the URL to search within a specific website: While the assistant can search the entire web by default, adding a URL helps the assistant narrow the focus to that content. Make sure the link is publicly accessible (not paywalled or login-protected). E.g., “Search within this page: [paste URL]”

DON'T:

  • Rely on single searches for complex topics.

  • Forget to verify that information is up to date and reflect recent changes in everything from laws to events, and more.

  • Include confidential client information in the web search.

  • Ask for authoritative sources: "Prioritise regulatory bodies, major law firms, and recent court decisions".

  • Request specific output formats. Web search results are presented according to a legal hierarchy, so requesting specific output formats such as tables or custom subheadings is generally not recommended.

Example Prompt

Objective
Conduct a risk-focused web search on Ericsson, identifying potential legal, financial, operational, reputational, or strategic risks based on publicly available information. The goal is to surface events or developments from 1 February 2024 to 1 February 2025 that may require further assessment in the context of legal due diligence.


Context
You are assisting with a legal due diligence review of the Swedish telecommunications company Ericsson. Your task is to search external public sources to identify any potential risk indicators or red flags that could affect Ericsson's legal, financial, operational, reputational, or commercial position. Your research should focus on reading the following, but is not limited to:
• Press releases from Ericsson
• Financial reports and performance updates
• Operational incidents or disruptions

• Legal or regulatory investigations, sanctions, or lawsuits

• Media coverage indicating reputational risks

• Any other developments relevant to risk assessment

Prioritise information from credible, authoritative sources, including:

• Regulatory authorities

• Major legal and financial news outlets

• Official company communications (e.g. investor relations pages, press statements)

• Public filings

Output

Present identified risks.


Tabular Review - Label Prompts

Best Practices

DO:

  • Be precise: Clearly define each label and the exact prompt needed to generate it.

  • Use consistent terms: Include a brief glossary in your prompt to define key terms, and stick to those definitions.

  • Classify with tags: Use clear rules like “If it looks like X, tag it as Y.” Allow for multiple tags when applicable.

  • Reference columns with “@”: Use the “@” symbol to refer to the other labels in your prompt when needed. E.g. if number in @Size is less than 100, assign the tag small.

  • Use correct formats for labels: If a label should be a e.g number, date, use the exact format.

  • Keep it cell-focused: Make sure the prompt works for filling in one cell, from one document at a time.

  • If you aim to extract exact text into a cell, incorporate the following into your prompt, e.g

    “Extract and include the exact text that you found in the agreement at the end of your answer within double citation marks (and format bold) as well as the reference to the exact article or section in the document (e.g., Section 4.5)”

  • Be explicit when format matters: If only a specific format is allowed (e.g., just names with no other text in the cell), state it clearly in the prompt — e.g., “Only write the parties’ names.”

DON'T:

  • Use ambiguous terms without examples.

  • Forget to specify management of missing data.

Example Prompt & Label Definition Example


Playbooks

Check out this page to understand how to write optimal rules in Playbooks.


Word Add-in

Best Practices

DO:

  • Request iterative improvements: "Make this clause more buyer-friendly, explaining each change".

  • Request reasoning: "Analyse this limitation of liability clause step-by-step, then suggest improvements".

  • Specify tone precisely: "Formal contract language" vs "Plain English summary".

  • Provide business context: This helps generate more nuanced drafting.

  • Choose the correct mode dependent on desired output: ask can answer questions about the document, and edit can suggest changes to the document based on your prompt.

  • Use highlights intentionally: If you want to work on a specific part of the document, highlight it. Otherwise, the whole document will be looked at.

DON'T:

  • Work with tracked changes visible

Example Prompts

Ask example:

Objective:

Analyze the employment agreement to identify and summarize risks that may weaken the confidentiality protections for the employer.

Context:

The current confidentiality clauses may leave the employer exposed to risks

involving unauthorized disclosure, misuse, or insufficient definition of

proprietary and sensitive business information. The goal is to uncover both

explicit and implied risks—such as vague language, loopholes, unenforceable

terms, or gaps in scope, that could compromise the employer’s control over

confidential data, trade secrets, internal systems, or client information.

Output:

Provide a table listing each identified risk. Each row should include:

• A short description of the risk

• Whether the risk is explicit or implied

• The section or excerpt of the agreement where the risk appears (if available)

• A brief explanation of why it is a risk for the employer

Edit example:

Objective:

Revise this employment agreement to strengthen the confidentiality protection

for the employer.

Context:

The current language does not sufficiently safeguard the employer’s proprietary

data and sensitive business information. The revised clause should bolster the

employer’s rights and protections concerning confidential information, trade

secrets, internal processes, client data, and other forms of proprietary

material.

Output:

Provide a revised version using formal contract language.


Advanced Techniques

Have Multiple Reasoning Steps

When you need deep analysis, request the reasoning in steps:

...

Do this step-by-step within the context given above:

1. Review the attached loan agreements using Tabular review. Use the template that you have for Loan agreements.

2. Based on the review, draft a summary of potential red flags about a potential transaction where our client would acquire the Lender in these agreements.

...

Using Examples with Style Matching

The model pays attention to examples and output formatting:

...

Format EXACTLY like this example for the pic cases:

Warranty 5.1: "Full ownership of all Intellectual Property"

- Risk Level: HIGH

- Issue: No carve-out for third-party or open source components

- Commercial Impact: Potential breach if any OSS is included in product

Use Follow-up Prompts for Iterative Refinement

Build on responses for deeper analysis.

First prompt:

Summarize the payment terms in this services agreement

Second prompt:

Draft a payment structure that mitigates each of these risks

while remaining commercially acceptable. Explain the trade-offs of each.


Troubleshooting Common Issues

Response too general?

  • Add explicit instructions: "Be specific" → "Include clause numbers, quote key language, and explain practical implications".

  • Request comprehensive analysis: "Include all relevant details".

  • Provide context about why specificity matters.

  • Be mindful of context windows: Legora uses techniques to work with long documents, but prompts still work best when they ask for one or a few well-scoped tasks at a time.

Missing important details?

  • Use numbered lists to ensure nothing is skipped.

  • Ask for confirmation: "After reviewing the entire document, confirm you've analysed sections 1-15", or "Double-check you've addressed every point in my request".

  • Request "think step-by-step through each section" and provide what each step includes.

  • For better follow-through, add: "Complete all requested tasks before summarising".

  • Legora memory from older messages in the conversation is limited, meaning a higher accuracy can be achieved by using fewer prompt messages.

  • Legora works best with documents shorter than 150 pages.

Wrong format or style?

  • Match your prompt style to the desired output.

  • Use normal text as input, without markdown formatting, to get output as formatted as normal text.

  • Include a format example within your prompt.

  • Specify exact structure: E.g., "Use the following headers in this order...", "Don’t use bullet points".

Not finding relevant precedents?

  • Try alternative legal terminology.

  • Break complex searches into steps.

  • Search broadly first, then filter results.

  • Include related practice areas.

Additional Tips

  • If output seems truncated, add: "Provide comprehensive analysis without length constraints".


Need additional support? Contact [email protected].

Did this answer your question?