AI-103 Practice Question 80: Skillset OCR Entity Enrichment

This is what AI enrichment is for: a skillset attached to the indexer performs "an enrichment, such as translating text or optical character recognition (OCR) on an image file," and Microsoft lists OCR and Entity Recognition among the built-in skills that turn unstructured content into "searchable and filterable fields in an index", automatically, on every indexer run, with no per-document code. Because the pages are scanned images, OCR has to run first to recognize the text; Microsoft's own guidance notes that when a skillset includes both, "the input to Entity Recognition should be document/merged_text," i.e., the OCR output merged back with the document content, so entity extraction sees the recognized text. Output field mappings then send the extracted organizations and locations to filterable fields in the index.

- B is wrong because default document cracking only extracts the *existing* digital text and structure from a file. It never runs OCR on scanned images and has no concept of named entities; it can't produce either capability the firm needs. - C is wrong because it requires labeling training samples up front, runs as a separate per-request pipeline invoked from application code, and does nothing to populate the search index's fields at indexing time the way a skillset does. - D is wrong because semantic ranking is a query-time reranker over results a search has already returned. It never extracts or writes any field into a document; it only reorders existing hits.

Reference: AI enrichment overview