Browse Certification Practice Tests by Exam Family

Microsoft AI-900 Cheat Sheet: AI Fundamentals

Review Microsoft Azure AI Fundamentals (AI-900) workloads, Azure service families, AI concepts, and common service-selection traps before practicing in IT Mastery.

AI-900 rewards clear workload recognition. Use this cheat sheet to separate machine learning, computer vision, natural language processing, generative AI, and responsible AI concepts before you answer timed practice questions.

Use this with practice. Review the workload distinctions, then try the free AI-900 diagnostic or open the full IT Mastery practice bank.

Try AI-900 on Web Free AI-900 diagnostic

Exam snapshot

FieldDetail
IssuerMicrosoft
Exam nameMicrosoft Azure AI Fundamentals
Exam codeAI-900
IT Mastery statusLive AI-900 practice available
Transition noteCheck Microsoft Learn before exam day because AI-900 has a published transition path toward AI-901.

Domain map

DomainPractice weightWhat to knowCommon trap
AI workloads and considerations19%Responsible AI principles, prediction, classification, anomaly detection, and common workload categoriesMemorizing service names without recognizing the workload
Machine learning on Azure19%Training, evaluation, regression, classification, clustering, and Azure Machine Learning basicsConfusing custom model training with prebuilt AI services
Computer vision on Azure19%Image analysis, object detection, OCR, facial recognition concepts, and vision inputsChoosing text analytics for image or video problems
Natural language processing on Azure19%Sentiment, key phrases, entities, translation, speech, and language understandingMissing whether the input is speech, text, image, or document layout
Generative AI workloads24%Prompts, completions, summarization, content generation, grounding, and responsible useTreating generative AI as the answer to every AI scenario

Must-know distinctions

DistinctionHow to decide
Classification vs regressionClassification predicts a category; regression predicts a numeric value.
Clustering vs classificationClustering groups unlabeled data; classification uses labeled examples.
Computer vision vs OCRVision interprets images; OCR extracts text from images or documents.
Speech-to-text vs text translationSpeech-to-text transcribes audio; translation changes text from one language to another.
Sentiment vs key phrasesSentiment detects opinion; key phrases extract important terms.
Generative AI vs extractionGenerative AI creates new content; extraction identifies existing facts or fields.
Responsible AI fairness vs privacyFairness addresses bias and equitable treatment; privacy protects personal or sensitive data.
Azure Machine Learning vs Azure AI servicesAzure Machine Learning supports custom ML workflows; Azure AI services provide prebuilt capabilities.

High-yield checklist

  • Identify the input type first: structured data, image, audio, text, document, or prompt.
  • Identify the desired output: prediction, category, transcript, translation, summary, generated content, or extracted field.
  • Match basic ML terms to the scenario: features, labels, training, evaluation, inference, model, and dataset.
  • Use prebuilt AI services when the task is common and does not require custom model training.
  • Use Azure Machine Learning when the scenario asks for custom training, experiments, model evaluation, or lifecycle management.
  • Keep responsible AI principles attached to concrete risks: bias, privacy, safety, transparency, reliability, and accountability.
  • Remember that generative AI may need grounding when it must answer from specific enterprise content.
  • Do not assume every AI solution needs code-heavy custom training.

Common traps

  • Selecting generative AI for a deterministic recognition or extraction task.
  • Confusing sentiment analysis with language translation.
  • Treating OCR as the same thing as understanding a full document workflow.
  • Choosing regression for a category answer or classification for a numeric prediction.
  • Ignoring the role of labeled data in supervised learning.
  • Missing responsible AI constraints because the scenario sounds purely technical.

Practice strategy

Start with the free AI-900 diagnostic and group misses by workload type. If most misses are service-selection errors, drill the topic pages until the input/output pattern is clear. If most misses are AI-concept errors, review the ML and responsible AI distinctions before returning to mixed practice.

Because AI-900 is a fundamentals route, avoid overtraining on memorized stems. The target is fast recognition of the workload, the Azure service family, and the reason the distractors do not fit.

Official source

Revised on Monday, May 25, 2026