Define the Fields First for the Copilot Studio Document Extract Node

How do you create the fields?

The official documentation describes three methods: a sample document, a template, or adding fields manually. Either way, define values that appear once, such as a vendor name or date, separately from values that repeat, such as line-item rows. Field names are also used when you connect data in later steps, so use names a reader can understand right away. Sample-suggested fields are only a starting point, so you can add, remove, and rename them.

How do you check the extraction results?

Upload one representative document and compare the values and table structure side by side with the original. Also test documents with empty fields, documents in a different layout, and documents with long tables. A value having been produced is not the same as it matching the original. Before attaching actions such as saving or sending notifications, decide how you will handle it when a value is wrong.

Can't you just use a general-purpose agent node?

Extract is built for getting structured values out of documents. Tasks that need to call tools and make decisions over several steps have to be designed separately. If the documents contain personal information, also check where it is stored, within your existing work permission boundaries.

Source: Microsoft Copilot Studio Extract node documentation. Date checked: 2026-10-10.