Create and extract attribute values from documents - Walkthrough

Extract structured values from documents using attributes, then review and export those results to understand patterns across your content.

Using attributes, you can extract values from documents, review results across a document set, and export structured data for analysis. This walkthrough shows how this works in practice.

Benefits

Extracting attribute values from documents helps you:

  • Identify key clauses and concepts across large document sets.
  • Normalize how important information is detected and reviewed.
  • Reduce manual document review effort.
  • Export structured data for reporting and analysis

What you'll do

In this walkthrough, you'll:

  • Create custom attributes using plain-language definitions.
  • Run extraction across a set of documents.
  • Review extracted values.
  • Filter and export attribute data.

Before you begin

For this exercise:

  • A set of 40 sample documents is provided. These are the sample collections found on the Home page.
  • The documents include a mix of agreement types to demonstrate cross-document extraction (20 sales and 20 vendor agreements).
  • Several example attribute definitions are included so you can see results immediately (pre-loaded on the Attributes page).

Custom attributes are data points defined in plain English. When you create an attribute, the system analyzes documents to locate text that matches the definition you provide.

Instead of manually searching documents, you describe what you want to find, and extraction runs automatically across all selected documents.

Follow the process to create a custom attribute using the values in the three examples below.

These examples demonstrate common, high-value concepts that appear across many documents.

Example attribute: Audit rights

What it finds
Clauses that give either party the right to audit books, records, or physical locations to ensure compliance.

Attribute values:

  • Attribute name: Audit rights
  • Attribute type: Text
    • Attributes can return different value types depending on your use case. Selecting the appropriate value type improves consistency and downstream usability.
      • Text - Values, such as Yes, No, or whole paragraphs of alphanumeric strings.
      • Numbers - Only numeric strings.
      • Dates - Anything in a standard date format. e.g., Feb 12, 2026 or 2/12/2026
  • Attribute mode: Strict
  • Document type: Legal Policy and Compliance Documents
  • Attribute group: Legal review
    • You can make a new group by clicking into the field and typing the name.
  • Detailed attribute description: Clauses indicating one party is entitled to share revenue or profit with the counterparty that are contingent on company value or financial performance milestones. Include royalty arrangements and bonuses tied to revenue or valuation targets. Exclude fixed payments or grants with no performance condition.

Example attribute: Revenue or profit sharing

What it finds
Clauses where one party is entitled to share revenue or profit with another party based on company value or financial performance milestones, including royalties or bonuses.

Attribute values:

  • Attribute name: Revenue/profit sharing
  • Attribute type: Text
  • Attribute mode: Strict
  • Document type: Legal Policy and Compliance Documents
  • Attribute group: Legal review
  • Attribute definition: Clauses indicating one party is entitled to share revenue or profit with the counterparty that are contingent on company value or financial performance milestones. Include royalty arrangements and bonuses tied to revenue or valuation targets. Exclude fixed payments or grants with no performance condition.

Example attribute: Non-compete restrictions

What it finds
Clauses that restrict a party from competing in business, promoting or distributing competing products, or working for competitors.

Attribute values:

  • Attribute name: Non-compete restrictions
  • Attribute type: Text
  • Attribute mode: Strict
  • Document type: Legal Policy and Compliance Documents
  • Attribute group: Legal review
  • Attribute definition: Restrictions preventing a party from competing in business, promoting/marketing/distributing competing products, or working for competitors. Include any geographic or time limitations and carve-outs. Do not include exclusivity arrangements, endorsement restrictions, or clauses stating no restriction exists.

Once extraction finishes:

  1. Go to the Collections page.

  2. Select one of the sample collections.

  3. Select the Adjust attribute view icon.

  4. Find and select your new attributes.

    A maximum of seven attributes can be added.

  5. Select Apply to populate the page with the selected attributes.

    The collection view with the "Add attribute" selector exposed.

    The collections page refreshes to display the selected attributes as column headers and shows extracted values inline with the document they were extracted from.

    The updated collections page with attribute columns added and populated.
    Hover over an attribute to view the full extracted text in a tooltip.

  6. Optional: Select a document to review it individually.

    The right-hand panel shows all enabled attributes for that document, including extracted text and citations that link to the exact location in the document.

    An individual document selected, exposing the list of attributes.
    Selecting the numbered citation in the attribute opens a view of the document, highlighting exactly where the extraction came from.

After reviewing the attribute values that have been extracted, you may find that some refinement of the attribute is required.

It’s common for initial results to be too broad or too narrow. Refining an attribute definition and re-running extraction helps improve accuracy over time.

  1. Select the Attributes tab to open the list of attributes.

  2. Find and select the attribute you want to refine.

    The attribute definition is opened automatically.

  3. Select the Evaluate attribute button under the definition. 

    The attribute page with the definition exposed and the "Evaluate attribute" button highlighted.

  4. Select up to 10 documents to evaluate. 

    Select documents that you expect to contain the attribute you are evaluating.

    • All files are selected by default. Use the collections dropdown to filter your document list to a single collection.
    • Use the Search by File Name field to filter the list of documents to only display documents with the searched-for name. string.
  5. Select the Evaluate files button.

    The "Evaluate attribute" list with filters applied.

  6. The page refreshes to show a table of documents with the extracted attribute to the right of the file name and the attribute description at the bottom of the page.

    Review the Detailed attribute description to provide context for your evaluation.

  7. Hover over a document row, and select the Review button.

    The "Evaluate attribute" page with the "Review" button highlighted.

    The page refreshes to display the location in the document where the value was extracted, and a copy of the content that was extracted to the right of that.

    The document citation with the in-document location highlighted and the text of the extraction beside it.

    Use the Back button to return to the list view.

  8. Review each document in turn to see how the attribute description evaluated and extracted the data.

    Over several documents you will likely see a trend that either affirms that the attribute is properly being extracted, or isn't.

  9. When the attribute isn't extracting the correct values, use the Show Explanation button to insert a new column that includes the reasoning of the AI for why that content was selected for this attribute.

    This provides insight into how you might adjust the attribute description and refine it for your purpose.

  10. Once you have a grasp of why the attribute extraction is happening like it is, edit the Detailed attribute description directly in the evaluation page and Save it.

  11. After the description is updated, the Re-analyze documents button is exposed.

    Select Re-analyze documents to see how your updated description impacts the extraction process.

    The attribute evaluation page with the "Re-analyze the document" button highlighted.

  12. It's worth spending some time repeating steps 8-11 until your results are as expected.

  13. If you continue to have unacceptable results, use the Advanced options in the attribute definition to provide positive and negative examples.

    1. Return to the Attributes page
    2. Select your attribute.
    3. Expand the Advanced options 
    4. Enter 1-5 examples to help the AI understand what you are or are not looking for.
    The attribute definition highlighting the Advanced options

You can further work with extracted values by filtering documents based on the attribute's presence and content.

Multiple filters can be applied across multiple attributes at the same time to narrow the collection to only the documents that match your criteria.

To filter a list of documents in a collection:

  1. Open the collection that contains extracted attribute values.

  2. (Optional but recommended) Have the attribute you want to filter by displayed. This will provide some visual feedback regarding how your document set is reacting to the filtering.

    Filters can be applied without having the attribute displayed.

  3. Select the Filter data button to open the filter builder in the right hand context panel.

  4. Select + Add filter to open the first filter definition.

  5. The filter definition contains two elements:

    • Filter by - This dropdown list contains all of the enabled attributes available to the user in the account.
    • Condition - This dropdown contains a dynamic list of options that can be applied to the selected Filter by attribute. It can be as simple as two elements (e.g., Before and After) or a larger list for more subjective values.
      • Some conditions will require additional input values to provide meaningful content, and those fields will be dynamically added as the filter is built.

    Define the filter based on the attributes you want to explore within the confines that matter.

    The filters being applied are displayed as tiles across the top of the document list. They can be removed by selecting the X on the right side of the tile.

  6. Repeat steps 4-5 to layer in the boundaries to extract the data that you need.

    The collection view with the "Filter by" panel exposed

Once you have reliably extracted data, you may need to export it for reporting, validation, or sharing with other teams.

To extract attribute data:

  1. Go to the Collections page and select the collection from which you want to extract the attribute data.

  2. Select the documents from which you want to extract the data.

  3. In the Document Attributes panel, select the ellipsis and pick the Export extracted data menu option.

    The collections page with the "Document Attributes" panel exposed and the "Export extracted data" menu option highlighted.

  4. The page refreshes to display the Export extracted data panel.

    All attribute groups are listed, with the individual attributes nested within them.

    Select all of the attributes you would like to export.

  5. Select Export when all of the germane attributes have been selected.

    The "Export extracted data" panel with several attributes selected.

  6. A message is displayed indicating that the export is underway, and in short order, a CSV file with the attributes and their values for each document selected will be downloaded to your local system.

    An example CSV file with the exported attributes.

Frequently asked questions

  • How long does extraction take?
    Extraction typically completes in a few minutes for small document sets. Larger sets scale proportionally.
  • Can an attribute definition be changed after creation?
    Yes. Attributes can be edited and re-evaluated at any time.
  • Can attributes extract dates, amounts, or names?
    Yes. Select the appropriate value type when creating the attribute to improve the extraction results.

Related topics

  • ATTRIBUTES OVERVIEW
  • CREATE A CUSTOM ATTRIBUTE
  • TEST AND VALIDATE A CUSTOM VALUE
  • EXPORT ATTRIBUTE VALUES

Adobe, Inc.

Get help faster and easier

New user?