# Azure AI Search Skillset

> Search skillsets run AI enrichment during indexing, covering OCR, entity extraction, and embedding, billed per transaction through the attached AI Services account rather than the search service itself. ZopNight discovers every skillset and links it to the account that bills for it, because a full re-index replays enrichment charges across the whole corpus.

Source: https://zop.dev/integrations/azure/ai-search-skillset
Updated: 2026-08-19

---

Skillsets apply AI enrichment (OCR, entity extraction, embedding) to documents during indexing, billed through attached AI Services accounts per transaction. Every indexer run through a skillset spends money.

## Enrichment bills per transaction, on another resource

A skillset moves the cost somewhere most bills reviews never look. The search service's search-unit meter is unaffected by enrichment; instead, each skill executed against each document generates transactions on the AI Services account the skillset is attached to. OCR over a scanned archive, entity extraction across a document store, embedding generation for a vector field: all of it meters per document, per skill, on an account that may sit in a different resource group from the search service that triggered it.

## Skillset-to-account linkage in inventory

Discovered via the search enricher and linked to the AI Services accounts that bill for enrichment. Skillsets are data-plane definitions, enumerated with the service admin key alongside their sibling indexers and indexes, and the link ZopNight preserves is the one that matters in a cost review: this skillset, invoked by that indexer, bills that account. Without the link, an AI Services account showing unexplained transaction volume is a mystery; with it, the driver is one lookup away.

## When re-indexing re-bills the whole corpus

Skillset waste is event-shaped rather than continuous. A schema change that forces a full re-index replays every skill over every document. The entire historical enrichment bill is paid again in an afternoon. High-frequency indexer schedules turn small incremental charges into a steady burn. And over-built skill chains from prototyping, OCR plus key phrases plus entities plus embeddings, keep billing every skill even when the application only ever queries the vectors.

## Reviewing skillsets alongside their indexers

Azure portal → AI Search → select the service → Search management → Skillsets lists each definition and the skills it chains; each skillset's attached AI Services resource is declared in its definition. Reading this view next to the Indexers list, which shows which schedules fire through which skillsets, is the fastest route to the enrichment charges worth cutting.