Azure Content Understanding: analyzers, modes and structured output

Content Understanding processes documents, images, audio and video through analyzers that combine classic extraction with generative field extraction. Configure the analyzer first, then let the agent reason over its output.

Analyzers

Azure Content Understanding in Foundry Tools is the recommended service for multimodal document processing. Analyzers combine OCR, layout and table extraction (including multipage tables) with generative field extraction. They return structure-preserving Markdown plus JSON fields.

Analyzer	Does
`prebuilt-read`	Text and barcodes, no layout, no model needed
`prebuilt-layout`	Words, paragraphs, figures, tables, sections, barcodes (QRCode, MicroQRCode); no LLM deployment
`prebuilt-invoice` and other prebuilt domain analyzers	Fields across varied layouts, no training
`prebuilt-documentSearch`	RAG analyzer using generative models
`prebuilt-documentFieldSchema`	Proposes a field schema
Custom analyzer (based on `prebuilt-document` or a copied template)	Your schema, fields and validation (for example, checking against contract terms)

Confidence, grounding and segmentation

Confidence and grounding are opt-in. Set estimateFieldSourceAndConfidence: true on the analyzer, or estimateSourceAndConfidence per field.
Each field then gets a 0–1 confidence and a source location, and the app routes values below a threshold (for example 0.80) to human review.
enableSegment splits documents for classification.
Groundedness guardrails and search.score measure other things, not extraction confidence.

Processing modes

Use standard mode (the default, with the lowest cost and latency, one file) for high-volume single invoices.
Use pro mode (reasoning, multiple files and a reference knowledge base) for cross-document validation.
Historically, pro mode existed only in the 2025-05-01-preview API, which is now retired; today it is being replaced by agentic mode (2026-06-01-preview, initially one file per request). In exam scenarios, pro mode is still the expected choice for cross-document validation.

REST operations

Create or replace an analyzer with REST PUT {endpoint}/contentunderstanding/analyzers/{analyzerId}?api-version=..., which returns 201 plus Operation-Location.
POST ...:analyze runs an analyzer, GET reads one and PATCH updates one.
An analyze call returns 202 Accepted, which means only “queued”. Poll Operation-Location until the status is Succeeded. The GA API (2025-11-01) keeps this 202 plus Operation-Location pattern.
Video results list key-frame times.

Common trap: using POST to create an analyzer - creation is PUT to the analyzer’s URL; POST with :analyze runs it.

Supported formats

PDF, TIFF, .docx/.xlsx/.pptx, JPEG, PNG, BMP and HEIF/HEIC.
WebP isn’t supported.

Get the whole book

This note is one section of Ultra Transcenders AI-103: Developing AI Apps and Agents on Azure, an independent study guide that explains every topic the exam covers by technology, with comparison tables, diagrams and the common traps, plus a glossary linked to Microsoft Learn.

Amazon.co.ukKindle: coming soonPaperback: coming soon

Amazon.comKindle: coming soonPaperback: coming soon

Publishing soon on Amazon in Kindle and paperback editions.

About the book · Free AI-103 glossary · All AI-103 study notes

More AI-103 study notes

Guardrails in Microsoft Foundry: intervention points, Prompt Shields and actions
Where guardrails check an agent run, what Prompt Shields catch, and when to block or annotate.
Choosing agent tools in Foundry Agent Service
File search, Azure AI Search, Bing grounding, function, OpenAPI, MCP and code interpreter tools compared.
Agent oversight: approvals, constraints and tool-access controls
How to keep humans in the loop and limit what an agent's tools can do.
Prompt engineering techniques for Azure AI apps
System messages, few-shot examples, chain of thought and grounding, and which fix suits which prompt problem.
Azure AI Search indexers, skillsets and the knowledge store
The indexer pipeline stages, built-in and custom skills, and knowledge store projections.
Vector, hybrid and semantic search in Azure AI Search
Vector fields and profiles, HNSW vs exhaustive KNN, hybrid queries with RRF, and the semantic ranker.
Video generation with Sora 2 in Microsoft Foundry
What Sora 2 can generate, its parameters and limits, and how the asynchronous jobs work.