Skip to main content

How to Ask Effective Questions

The quality of an AI Chat answer depends on how clearly you define what you need. Use the guidance below to improve accuracy and make outputs easier to verify.

Be specific about scope​

Avoid broad prompts that do not define a topic, time window, or source set.

  • Avoid: “Tell me about this case.”
  • Use: “Summarize the private vehicle sales mentioned in these WhatsApp invoices and identify the main individuals involved.”

Add details that help the system retrieve the right evidence:

  • Names of people, companies, or organizations
  • Date ranges (or a start and end date)
  • Identifiers (phone numbers, IBANs, license plates, email addresses)
  • File or folder context (for example “this folder” or “these selected documents”)

Examples:

  • “In which documents does Valerio Simoni appear, and in what context?”
  • “Which Relationships connect this company to luxury vehicle sales in 2023?”
  • “List all phone numbers and IBANs mentioned in the selected files, grouped by source document.”

Define the output you want​

State what you want AI Chat to produce. If you do not specify an output format, you may get a narrative answer that is harder to review.

Common output types:

  • A summary (short or detailed)
  • A structured list (entities, identifiers, events)
  • A comparison (differences across documents or versions)
  • A list of Relationships (who is connected to whom and why)
  • A timeline (ordered by date)
  • An “open questions” list (gaps, contradictions, missing evidence)

Examples:

  • “Summarize all references to private vehicle sales in this folder. Output: bullet list with date, seller, buyer, amount, and source file.”
  • “How are these two Records connected? Output: the shortest connection path and the supporting sources.”
  • “What unanswered questions remain in this dataset? Output: top 10 questions and which document(s) might answer each one."

Use AI Chat for interpretation, not exact counting​

AI Chat is strongest when you need interpretation and synthesis across sources. Use it to:

  • Interpret context and meaning across documents
  • Connect Records, Concepts, and Relationships
  • Summarize long text or many documents
  • Extract themes, patterns, and inconsistencies
  • Propose follow-up lines of inquiry For exact counts or exhaustive totals across large datasets, use Search and filters where possible. Language models respond based on retrieved context and do not guarantee complete numerical aggregation unless the full relevant set is explicitly retrieved and reviewed.

Examples of safer phrasing:

  • “Based on the retrieved sources, list the transactions you can cite. Do not estimate totals.”
  • “If you cannot confirm a count from the sources, say so and list what you found.”

Ask for citations and verify For sensitive findings or anything that goes into a report, ask for traceability. Use prompts like:

  • “Provide citations for each statement.”
  • “For every name, amount, and date, cite the source document and the passage.” Then verify:
  • Review the references section in the response.
  • Open each cited file or Record and confirm the detail in the original content.
  • Treat summaries as pointers. Do not rely on them without checking the underlying evidence.

Iterate in small steps​

When a question is complex, break it into two passes:

  1. Extract facts with citations
  2. “List all vehicle sale transactions with date, parties, and amount, with citations.”
  3. Analyze patterns
  4. “Based on the cited transactions, identify the top three individuals involved and how they are connected.” This approach makes results easier to validate and reduces missed details.