The Azure Content Understanding skill (Microsoft.Skills.Util.ContentUnderstandingSkill) uses Azure Content Understanding document analyzers in Foundry Tools to extract both text and images while returning locationMetadata (page numbers and bounding-polygon/source position data) for each text chunk and each normalized image, which is exactly what's needed to build page-level citations with bounding regions for both content types. Critically, Microsoft's documentation states that for tables spanning multiple pages, the Azure Content Understanding skill can recognize and extract cross-page tables as a single unit — a capability the alternative Document Layout skill lacks (it outputs tables as plain text on a per-page basis and does not merge them across pages without additional custom logic). The Document Extraction skill does not provide text location metadata (no polygon/page-number data for text), and the GenAI Prompt skill is a generative/LLM skill for producing new content (e.g., summaries or image captions), not a content-extraction skill, so neither meets the stated requirements.
Community Discussion