Noticia Go integrates directly into your Nuix Discover workflow, giving you the power to tailor its AI analysis for each specific matter. By creating and refining your own AI prompts, you can achieve reliable, repeatable outputs perfectly suited to your case needs.
This guide provides a starting point with sample prompts for common eDiscovery tasks like relevance, privilege, PII detection, and event extraction. Think of these as templates to be adapted, not final commands.
The Power of Customization
Effective AI review hinges on providing the model with clear, specific, and context-rich instructions. The more detailed your prompt, the more accurate and useful the AI's response will be.
Best Practices for Custom Prompts:
- Be Specific: Instead of asking if a document is "important," define what "important" means for your case. Include key issues, date ranges, and names of key players.
- Provide Context: Give the AI the background it needs. If you're looking for privileged documents, provide the names of known attorneys and law firms involved in the matter.
- Iterate and Test: The first prompt you write may not be the best. Test your prompts on a sample set of documents. Analyze the AI's output and refine the prompt to improve accuracy. Small changes in wording can lead to significant improvements in the results.
- Structure the Output: Use the prompt to tell the AI exactly how to format its response. This makes the output easier to review and allows you to push results directly into Nuix Discover fields for efficient coding and analysis.
Below are starter prompts for you to customize. Remember to replace the bracketed text [User will input...]
with the specific details of your case.
Sample Prompt: Assessing for Relevance
This prompt guides the AI to assess a document's responsiveness based on case-specific criteria you provide. It is designed to work with Noticia Go's Relevance Scoring capabilities, which can help find responsive documents up to 85% faster.
Prompt Template:
Based on the document's content, assess its responsiveness to the following case criteria: [User will input specific relevance definition, key issues, date ranges, entities, topics, or questions central to the case].
Please provide your findings structured as follows:
Relevance Determination: [Clearly state RESPONSIVE or NOT RESPONSIVE]
Confidence Level: [Specify HIGH, MEDIUM, or LOW]
Detailed Justification: [Provide a comprehensive explanation for your determination. Reference specific information, themes, or language within the document and explicitly connect it to the provided case criteria. Explain how the document content supports or refutes responsiveness.]
Key Supporting Excerpt(s): [Extract one or two concise, impactful excerpts from the document that directly and strongly support your determination. If the document is lengthy and contains many relevant sections, choose the most illustrative ones. If no single excerpt is definitive but the overall content supports the determination, state "Overall document content supports the determination" and refer to your detailed justification. If NOT RESPONSIVE and no specific excerpt illustrates this well, state "N/A".]
Potentially Related Themes/Keywords Not Directly Responsive (if any): [If the document touches on themes or keywords related to the case criteria but isn't directly responsive, briefly note them here. Otherwise, state "N/A".]
Sample Prompt: Detecting Privilege
Use this prompt to leverage the Privilege Detection feature for identifying materials that are attorney-client privileged, work product, or subject to common interest privilege. Providing case-specific context is crucial for accuracy.
Prompt Template:
Analyze this document to determine if it contains privileged information. Consider the following case-specific context which may help identify privileged communications or work product (e.g., names of known attorneys, law firms, client personnel involved in legal matters, relevant litigation, anticipated litigation, or subject matter typically covered by privilege for this case): [User will input case-specific context for privilege].
Please provide your findings structured as follows:
Privilege Call: [Clearly state PRIVILEGED or NOT PRIVILEGED]
Privilege Type(s): [If privileged, specify all applicable types, for example: SOLICITOR-CLIENT COMMUNICATION, LITIGATION, COMMON-INTEREST PRIVILEGE, etc. If NOT PRIVILEGED, state "N/A".]
Involved Parties in Privileged Content (if applicable and identifiable): [Identify key individuals or entities and their roles. If not applicable, state "N/A".]
Detailed Justification for Privilege Call: [Provide a comprehensive explanation. Cite specific language, relationships, or subject matter.]
Key Supporting Excerpt(s) for Privilege: [Extract one or two concise excerpts from the document that best illustrate the basis for the privilege call. If NOT PRIVILEGED, state "N/A".]
Sample Prompt: Finding Personally Identifiable Information (PII)
This prompt activates the PII Finder feature, designed to classify personal data and assist with privacy regulations. You can customize it to look for specific types of PII relevant to your matter.
Prompt Template:
Thoroughly scan this document to identify all instances of Personally Identifiable Information (PII). Consider a broad range of PII types including, but not limited to: full names, dates of birth, SSNs, passport numbers, driver's license numbers, physical addresses, email addresses, phone numbers, financial account numbers, health information, biometric data, IP addresses, geolocation data.
Please provide your findings structured as follows:
PII Presence: [Clearly state PII PRESENT or NO PII DETECTED]
PII Types Found: [List all distinct types of PII found. If none, state "N/A".]
Detailed Summary and Examples of PII: [Summarize the count and provide redacted examples for each category. If none, state "N/A".]
Sample Prompt: Extracting Key Events
Leverage the Event Extraction functionality to quickly build timelines by pulling key facts from documents. Specify the event or topic you are investigating for a focused analysis.
Prompt Template:
Analyze this document to determine if it references or discusses the following specific event(s) or topic(s) of interest: [User will input a detailed description].
Please provide your findings structured as follows:
Event/Topic Presence: [EVENT/TOPIC MENTIONED or NOT MENTIONED]
Number of Distinct References: [Number or "0"]
Key Details Discussed: [Summary of facts, actions, or decisions]
Associated Dates/Timeframes: [All explicit dates or "N/A"]
Key Persons or Entities Involved: [List or "N/A"]
Supporting Excerpt(s): [1–3 excerpts or "N/A"]
Sample Prompt: Extracting "To" Email Addresses
Use this prompt to extract only the recipient email addresses from the first message in an email thread. It ignores Cc, Bcc, and From and returns clean results for easy processing or contact mapping.
Prompt Template:
Extract all "To:" email addresses from this thread. Only return the actual email addresses from inside angle brackets, but do not include the brackets (e.g. example@email.com). Do not include names, Cc, Bcc, or From addresses. Return one list of unique email addresses, separated by semicolons. Only read the first email at the top of the thread - Do not return any addresses from subsequent "To:" lines.
Was this article helpful?
That’s Great!
Thank you for your feedback
Sorry! We couldn't be helpful
Thank you for your feedback
Feedback sent
We appreciate your effort and will try to fix the article