Content extraction methods

  • Release version: Australia
  • Updated August 7, 2026
  • 2 minutes to read

    • The Extract information from documents skill (Information Extraction skill) supports three extraction methods—field, table, and Q&A—each suited for different types of document data. Understanding which method to use helps you capture and store the right information in ServiceNow.

    The Information Extraction skill uses AI-assisted parsing to extract data from documents (invoices, purchase orders, and contracts) and then populates records in ServiceNow. Select a method or combination of methods that matches the type of information you want to capture and store in ServiceNow tables.

    Field extraction

    Field extraction captures specific, isolated data points from a document—values that appear exactly once—and maps each to a single field on a ServiceNow record.

    Use field extraction for data such as:

    • Invoice numbers, purchase order numbers, or document identifiers
    • Vendor or customer names
    • Dates, such as an invoice date, due date, or contract start date
    • Monetary totals or tax amounts
    • Addresses or other contact details

    You can bundle related single fields into a logical group. For example, a Billing Address group can capture Street, City, State, and ZIP from the same region of a document.

    Table extraction

    Table extraction captures dynamic, multi-row data from documents where the number of entries varies between documents, such as invoice line items or expense report details. A table acts as a two-dimensional grid of rows and columns that captures a list of related records.

    Use table extraction for data such as:

    • Invoice or purchase order line items (quantity, description, unit price, total)
    • Expense report breakdowns
    • Parts lists or bills of materials
    • Multi-page flight itineraries or timesheet entries

    Q&A extraction

    Q&A extraction uses generative AI to answer open-ended, natural language questions about the contents of a document. It is suited for unstructured text that does not fit neatly into a predefined field or table schema. The extraction processes meaning rather than matching a fixed pattern.

    Warning:
    Q&A extraction uses generative AI and may produce inaccurate or incomplete results. Review all AI-generated output before using it to populate records.

    Use Q&A extraction for questions such as:

    • The reason for termination
    • Special shipping or delivery terms
    • The designated project manager
    • Narrative clauses or conditional language from contracts

    You can also use Q&A extraction to categorize documents.

    Selecting an extraction method

    Use the following guidance to choose the appropriate extraction method.

    Table 1. Extraction method selection guidance
    Extraction method When to use Example
    Field extraction Use when the data point appears once and has a well-defined type. A number, date, or name.
    Table extraction Use when the document contains a list or grid of repeating items where the row count is unknown or varies across documents. Line items in an invoice or purchase order.
    Q&A extraction Use when the information is embedded in narrative text and requires interpretation rather than pattern matching. An answer based on contract terms or policy text.

    You can combine multiple methods within a single use case to handle all of the data types present in a document.