Azure AI Search knowledge stores support three projection types, each mapping to a specific Azure Storage destination and data shape. Table projections write to Azure Table Storage and require a schematized shape of rows and columns, which is best suited to structured, tabular enrichment output. Object projections write JSON documents to Azure Blob Storage and are used when you need the full JSON representation of an enrichment tree or whole document — this makes Object projection the correct choice for unstructured JSON data, since it preserves the JSON structure without forcing it into a rigid table schema. File projections also write to Azure Blob Storage but are specifically used to persist the binary content extracted/normalized from source documents (such as normalized images and the text/content derived from scanned PDF documents processed through OCR), making File projection the correct choice for the extracted text data originating from the scanned PDFs. This distinction — Table for tabular data, Object for whole-document JSON, File for extracted/normalized binary content from source documents — is documented in the Azure AI Search knowledge store projections concepts.
Community Discussion