Deep Dive Ask Guide

This guide is written to provide advice on how to write questions for Deep Dive that result in useful, relevant responses.

Effective question types

Deep Dive is intended to help you precisely identify key information or explore insights relevant to concepts, issues, or events in your matter. The types of questions the tool is best suited to answer are: 

  • Open-ended investigative-style questions
  • Questions pinpointing specific quotes
  • Questions related to a specific theme or topic

Ineffective question types

As the tool is designed to surface only the information most relevant to your question, its results are not intended to be comprehensive. In other words, Deep Dive is not intended to replace traditional search or filters. Avoid questions that:

  • Ask for a comprehensive list of documents that meet a certain criteria
  • Require a judgement, such as whether text is "true" or whether a certain document is the "best" example. Deep Dive is good at identifying relevant quotes from your documents, but it is not able to assess external qualities of the content.
  • Involve complex logic or numerical analysis
  • Specify the date range of interest

    Tip

    To ask a question pertaining to a specific date range, create a search on the Search page to identify your subset of documents. Then select Ask a question from the results table.

  • Rely on context outside the documents and case description

Best practices for asking questions

Here are some best practices and tips to help you get the most out of Deep Dive:

  • Ask one question at a time: This will help Deep Dive be more focused and provide a more nuanced response to your single question at hand
  • Ask each question in isolation of previous questions: Deep Dive does not take into account previously asked questions or responses when evaluating your current question
  • Start with question words like who, what, where, when, why, or how: It may be helpful to use a classic question word to get started
  • Be specific: Use proper nouns and precise terms to help the tool identify the document text that supports generating the most useful answer
  • Iterate on name variations: If an entity may be referred to using multiple names and it’s not obvious that the names refer to the same entity (e.g. “Mark Smith” is the same as “Dr. Smith”), it may be helpful to ask multiple questions using alternative entity names
  • Avoid complex logic: Rather than ask a question that requires linking several clauses together, separate each clause into its own question, or re-evaluate the question to define a narrower, more precise scope
  • Don't ask for a specific format: Deep Dive's gives its best responses when it formats the response without guidance 
  • Narrow your document set before asking: Open the results table for a document set you know will be relevant and select Ask a question. This ensures that the tool is using a set of documents that you know to be relevant before generating a response.

Examples of effective and ineffective questions 

Effective questions leverage Deep Dive's ability to analyze text across documents to answer specific questions. Ineffective questions require either precise numerical answers that are better returned by a search query, or are too broad and complex to generate a useful answer. 

Effective ✅Ineffective ❌
  • Who did the board of directors of ABC Corp consult about the 2024 Q1 cost cuts?
     
  • How did John Smith express his discontent towards the revised budget proposal?

     

  • What did Company X and Company Y discuss regarding the price of wool in 2025?

     

  • Who at Company X was aware that the proposed wool price was below market rate?



 

  • Show me all documents containing the term “cost cuts” between 1/1/24 and 3/31/24.

     

  • How many emails did John send to the Finance team last year?

     

  • Did Company X and Company Y collude to form an oligopoly in the sweater industry, and what evidence is there around Company Z’s efforts to report suspicions of this behavior to the FTC?