Add a ground truth to each dataset record

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

    • Add a ground truth to each record in a dataset. A ground truth is an expected correct output for a given record. During evaluation in Now Assist Skill Kit, your custom skill's actual output is compared against the ground truth to measure accuracy and identify areas for improvement.

    About this task

    Ground truth content should match your custom skill's output type. For a summarization skill, enter an ideal paragraph-length summary as the ground truth. For a content generation skill, enter the expected output text. The closer your ground truth examples are to real, high-quality outputs, the more meaningful your evaluation results will be.

    Before you begin

    Role required: sn_data_kit.admin, sn_data_kit.analyst

    The Create ground truth guidelines button is controlled by a system property that is not enabled by default. Before you begin, verify that the sn_data_kit.enable_ground_truth system property exists on your instance and is set to true. If the property does not exist, create it:

    1. Navigate to All > System Properties > System Properties.
    2. Select New.
    3. Fill in the following fields and select Submit:
      • Name: sn_data_kit.enable_ground_truth
      • Type: String
      • Value: true
    Note:
    Create this property in the Global application scope. If you create it in the wrong scope, the button doesn't appear.

    Procedure

    1. Navigate to All > Now Assist Data Kit > Home, and open an existing data set in the Datasets tab.
    2. Select the Create ground truth guidelines button.
    3. On the form, fill in the fields.
      Table 1. Add ground truth info form
      Field Description
      Type
      • Manual
      Ground truth type
      • Summarization
      • Content generation
      • Other
      Instructions Guidelines for labelers and linguists who manually add the ground truth.
      Column type If you chose AI generated ground truth, select the column type.
      • String
      • Number
      • JSON
      Column label The name of the column that is added for the ground truth in the dataset.
    4. Select Confirm.
      A new Record detail page opens.
    5. On the form, fill in the fields.
      Table 2. Record detail info form
      Field Description
      Form label Incident record number.
      State Time of the record creation.
      Summarization Manual entry of the ground truth.
      You can manually enter the ground truth and rate the ground truth. These options appear in a separate column in the dataset record.
    6. Select Save and next
      The ground truth column is added in a separate column in the dataset. Use the Add to data collection pop-up window to combine similar records from different datasets.
    7. Add the dataset to an existing collection by selecting Add to data collection.
      1. Enter the data collection name, description, data collection category, and relevant tags.
      2. To create a data collection, slide the toggle bar.
    8. Select Next.
      A new data collection page populates.
    9. Select the Choose columns form and then select the columns that you want to add to the data collection.
    10. Select the Choose records form and select how you want to sample the dataset again (manually or sampling method).
      For a random sample, select Run to preview the record. Choose a sampling method, and select Run to preview the record.
      The selected records are added to the data collection.
    11. Select the data collection to preview the records.
    12. SelectPublish to make the data available for validation.
      When you publish a collection, the data freezes curation and makes the dataset available for use through ServiceNow SDK.
    13. Select Confirm to make your collection available.
      The data collection is published and available in ServiceNow SDK for evaluation.