One service across four content types
Azure Content Understanding (via Microsoft Foundry — portal, Studio, or API) analyzes four content modalities with one consistent development process.
| Modality | Example use cases |
|---|---|
| Documents & forms | Extract field values — e.g. invoice data to automate payments. |
| Images | Chart analysis, defect detection, object/person detection. |
| Audio | Call summaries, sentiment analysis, key data from voice messages. |
| Video | Key points from recordings, presentation summaries, activity detection in footage. |
Building a solution
A solution centers on an analyzer — a trained tool that extracts information for a content type based on a schema. Thanks to generative AI, it needs minimal training data, though explicit field labeling improves performance.
Key concept · The four-step process
(1) Create a Foundry resource; (2) define a schema — the fields to extract, optionally from a template and sample content; (3) build an analyzer from the schema; (4) deploy and use it against new content from client apps. In Content Understanding Studio you create a project (which provisions storage and a key vault), upload a sample, apply a schema template, define fields, test at any point (reviewing extracted values and JSON), then build — making the analyzer available at the Foundry endpoint, with versioning for continuous refinement.
Submit, then poll
Client apps call the Content Understanding endpoint over HTTP, authorizing with a key in the request header (or Foundry + Entra ID).
Key concept · The asynchronous two-request pattern
(1) POST content to …/analyzers/{analyzer}:analyze (or analyzeBinary for raw file data) with an input URL. The response is asynchronous — it returns an Operation-Id / Operation-Location and a status of NotStarted. (2) GET the analyzerResults/{id} endpoint repeatedly until it completes, then read the JSON result — whose shape depends on the content type and schema. API version 2025-11-01.
Chapter 27 quiz
Thirteen questions on modalities, the build process, and the API. Answer first, then expand Show answer.
Part A · Multiple choice
Which four modalities does Content Understanding support?
- A Documents, images, audio, and video
- B Text and images only
- C SQL and NoSQL
- D Slides and spreadsheets
Show answer
A. One service, one process, four content types.
What is an analyzer based on?
- A A schema defining the fields to extract
- B A fixed prompt template
- C A vector index
- D A container image
Show answer
A. A schema, tied to a content type.
What's the first step of building a Content Understanding solution?
- A Create a Foundry resource
- B Publish to Teams
- C Train a model from scratch
- D Buy a GPU
Show answer
A. Then define a schema, build an analyzer, deploy and use.
What does creating a project in Studio provision?
- A Storage and a key vault, associated with a Foundry resource
- B A virtual machine
- C A domain name
- D Nothing
Show answer
A. The resources the workflow needs.
The analyze API is best described as:
- A Asynchronous — POST to submit, GET to poll for results
- B A single synchronous call
- C A streaming socket
- D A GraphQL query
Show answer
A. Returns an Operation-Id, then poll analyzerResults.
Which operation submits raw file data instead of a URL?
- A analyzeBinary
- B analyzeText
- C generate
- D remix
Show answer
A — analyzeBinary.
Part B · True or false
Generative AI lets an analyzer work with minimal training data.
Show answer
True. Explicit field labeling still improves performance.
You can test a schema at any point during development in Studio.
Show answer
True. Reviewing extracted values and JSON output.
Building an analyzer makes it available at the Foundry resource endpoint.
Show answer
True. With versioning for continuous refinement.
The analyze API returns final results in the first POST response.
Show answer
False. It returns an operation ID; you poll for the result.
Part C · Short answer
Name the four steps of building a Content Understanding solution.
Show answer
Create a Foundry resource, define a schema, build an analyzer, deploy and use it.
Give an example use case for the audio modality.
Show answer
Any: conference-call summaries, sentiment analysis of conversations, key data from voice messages.
What two response headers does the submit request return?
Show answer
Operation-Id and Operation-Location (with an initial NotStarted status).