Azure AI Service Names — the Exam's Bread and Butter

A large share of fundamentals questions reduce to one pattern: "You need to do X. Which service should you use?" You know the concepts; the gap is Microsoft's current product names — and "current" matters, because the exam tests today's names and several old ones were retired or renamed.

The renames to internalize first

• "Cognitive Services" is a retired umbrella name → the family is now Azure AI services, and the platform hub is Azure AI Foundry / Microsoft Foundry.
• Form Recognizer → Azure AI Document Intelligence (forms, invoices, OCR of structured documents — and yes, transcripts).
• Individual services now carry the "Azure AI" prefix: Azure AI Vision, Azure AI Language, Azure AI Speech, Azure AI Translator, Azure AI Search (formerly "Cognitive Search"), Azure AI Content Safety.

If a practice question offers "Cognitive Services" as an answer against an "Azure AI …" option, prefer the current name.

Which service for what — the drill table

• Azure OpenAI (in Foundry) — LLMs: generation, chat, summarization, embeddings; the engine behind custom copilots and RAG answers.
• Azure AI Search — the retrieval half of RAG: indexes your documents, hybrid keyword + vector search, feeds grounding context to the model.
• Azure AI Document Intelligence — extract structured data from documents: forms, invoices, receipts, IDs; prebuilt and custom extraction models.
• Azure AI Vision — image analysis: image classification, object detection, OCR of images, face detection.
• Azure AI Language — text analytics: sentiment, key phrases, entity recognition (NER), PII detection, summarization, conversational language understanding (CLU), question answering.
• Azure AI Speech — speech-to-text, text-to-speech, speaker recognition, real-time speech translation.
• Azure AI Translator — text translation across languages (document translation too).
• Azure AI Content Safety — moderation: detect harmful, unsafe, or inappropriate content in text and images (including AI outputs).
• Azure Machine Learning — the classical ML workbench: train, register, deploy, and monitor your own models; automated ML; the "ML studio" the duty statement names. Not for calling prebuilt AI — for building models.
• Microsoft Foundry (Azure AI Foundry) — the unifying platform where models, agents, and AI apps are built and deployed (next lesson).

The disambiguation questions the exam loves

• Extract totals from scanned invoices? → Document Intelligence (not Vision — Vision does general OCR/objects; Document Intelligence understands document structure).
• Detect sentiment in support tickets? → Language (not OpenAI — prebuilt beats generative for a classic text-analytics task, and that's the fundamentals-level "right" answer).
• Search your own documents to ground an LLM's answers? → Azure AI Search + Azure OpenAI.
• Train a custom demand-forecasting model on tabular data? → Azure Machine Learning (classical ML, not an AI service API).
• Real-time subtitle translation of a meeting? → Speech (speech translation), not Translator (text) alone.
• Filter harmful outputs from a chatbot? → Content Safety.

The pattern behind the pattern: prebuilt Azure AI service for standard perception/language tasks; Azure OpenAI for generative and reasoning tasks; Azure Machine Learning when you train your own model on your own data. Answer from that triage and most service questions fall.

Say-it-once mapping to your world

Your ops platform's AI failure analysis = an Azure OpenAI-shaped workload; your 106-field credential parsing = Document Intelligence territory; ainews search = Azure AI Search's hybrid pattern; the anomaly pilot = the Azure ML tier (when it graduates from statistics). Anchoring each service to something you've built makes the names stick without flashcards.