Azure AI-900 Sample Questions with Explanations

These original sample questions are designed to help you check how the exam topics appear in decision-style prompts. They are not taken from the live exam.

Use these sample questions as a guided self-assessment for Microsoft Azure AI Fundamentals (AI-900) topics such as AI workload recognition, responsible AI, machine learning basics, vision, language, document intelligence, and generative AI service fit. AI-900 is a fundamentals exam, so the best answer usually classifies the workload before choosing a product name.

Where these questions fit in the AI-900 guide

The sample set below is part of the Microsoft AI-900 guide path:

• Microsoft certification hub
• AI-900 exam guide
• AI-900 cheat sheet
• AI-900 study plan
• AI-900 FAQ
• AI-900 resources
• AI-900 glossary

AI-900 fundamentals sample questions

Work through each prompt before opening the explanation. The goal is to practice the exam habit: identify the workload, then choose the simplest Azure AI concept or service family that matches it.

---

Question 1

Topic: Classifying a maintenance workload

A manufacturing team collects vibration and temperature readings from equipment. They want to predict whether a machine is likely to fail soon so maintenance can be scheduled before downtime occurs. Which AI workload category best fits the requirement?

• A. Translation, because readings must be converted into another language.
• B. Predictive machine learning, because historical measurements can be used to estimate future failure risk.
• C. Computer vision, because the system is monitoring physical machines.
• D. Generative AI, because the system should create maintenance instructions from scratch.

Best answer: B

Explanation: The requirement is prediction from historical and current measurements. At AI-900 depth, that points to a machine learning scenario such as predictive maintenance rather than a language, vision, or content-generation workload.

Why the other choices are weaker:

• A is about language conversion, not failure prediction.
• C would fit image or video analysis, but the prompt describes sensor readings.
• D may help write instructions later, but it is not the core workload requested.

What this tests: Recognizing predictive machine learning scenarios before choosing an Azure service.

Related topics: Machine learning; Predictive maintenance; Workload classification; AI fundamentals

---

Question 2

Topic: Responsible AI for a hiring assistant

A company plans to use an AI assistant to summarize resumes and highlight candidates for recruiter review. Leaders are concerned the system could repeat unfair patterns from past hiring data. Which practice is most directly aligned with responsible AI?

• A. Remove all human review so every candidate is processed consistently.
• B. Increase the maximum prompt length so the model can read more resumes at once.
• C. Use only generated summaries and delete the original resumes.
• D. Evaluate outputs for fairness, monitor recommendations across relevant groups, and keep human decision-makers accountable.

Best answer: D

Explanation: The risk signal is unfair treatment in a sensitive human decision workflow. A responsible design evaluates and monitors outcomes, keeps humans accountable, and treats the AI system as decision support rather than an unquestioned authority.

Why the other choices are weaker:

• A removes human accountability instead of improving it.
• B is a capacity setting and does not address fairness risk.
• C weakens transparency and reviewability.

What this tests: Applying responsible AI principles to a high-impact business process.

Related topics: Responsible AI; Fairness; Accountability; Human review

---

Question 3

Topic: Extracting data from receipts

An expense app needs to read uploaded receipt images and extract merchant name, transaction date, tax, total amount, and line items into structured fields for review. Which capability is the best match?

• A. Document intelligence or form extraction for structured fields from receipts.
• B. Speech recognition because receipts contain words.
• C. Translation because merchant names may be unfamiliar.
• D. Anomaly detection because every receipt has a total amount.

Best answer: A

Explanation: The clue is structured extraction from semi-structured documents. AI-900 expects you to separate document extraction from generic language, speech, or anomaly scenarios.

Why the other choices are weaker:

• B handles audio-to-text, not receipt images.
• C may translate text, but it does not extract the required fields and line items.
• D may flag unusual totals later, but it does not read the receipt.

What this tests: Matching document-processing requirements to the right AI workload.

Related topics: Document intelligence; Receipts; Structured extraction; Service fit

---

Question 4

Topic: Grounding a generative AI assistant

A support team wants a generative AI assistant to answer questions using the company’s product manuals instead of relying only on the model’s general training. Which design idea best addresses that goal?

• A. Use image classification to label product screenshots.
• B. Disable prompts that contain product names.
• C. Ground the assistant with retrieved content from the product manuals before generating an answer.
• D. Translate every question into several languages before sending it to the model.

Best answer: C

Explanation: The prompt asks for answers based on company documents. Grounding through retrieval gives the model relevant source context and reduces the chance that it answers from general knowledge alone.

Why the other choices are weaker:

• A is a vision task, not a document-grounded response pattern.
• B blocks useful questions instead of improving answer grounding.
• D changes language handling but does not connect answers to the manuals.

What this tests: Understanding basic generative AI grounding and retrieval concepts.

Related topics: Generative AI; Grounding; Retrieval; Microsoft Foundry

Independent study note

Tech Exam Lexicon and IT Mastery are independent study tools. They are not affiliated with, endorsed by, or sponsored by Microsoft or any certification body.