Azure AI-900 Practice Exams

How the AI-900 exam is structured

Azure AI Fundamentals (AI-900) is a Microsoft certification with approximately 40 questions in 60 minutes. The table below shows the exam weighting published for each domain, so you can decide where to spend your study time instead of working through the syllabus in order.

Domain	Exam weight	What it covers
Natural Language Processing	25%	Describe features of NLP workloads
AI Workloads and Considerations	20%	Describe AI workloads and considerations
Machine Learning Principles	20%	Describe fundamental principles of machine learning
Generative AI	20%	Describe features of generative AI workloads
Computer Vision Workloads	15%	Describe features of computer vision workloads

A passing study plan usually trails the weighting: spend the most sessions on Natural Language Processing (25%), then check the next domain down before moving on. If your accuracy on a high-weight domain stays under 80%, more practice there is worth more than another pass at a low-weight topic.

Worked sample AI-900 questions

These questions come from the Azure AI Fundamentals (AI-900) practice bank. Read the rationale even when you answer correctly — the explanation is usually where the exam technique lives.

Sample question 1 — Natural Language Processing

A consumer research firm receives thousands of open-ended survey responses every month. Analysts want a ranked list of the main topics mentioned in each response so they can build a word cloud and spot emerging complaints, without reading every comment. They have no labeled examples and do not need to assign fixed categories. Which Azure AI Language feature should the analysts use?

A. Run key phrase extraction over each survey response.
B. Score each response with the prebuilt sentiment feature.
C. Train a custom text classification model on the comments.
D. Detect the language of each response before analysis.

Answer: A. Option A is the right fit because key phrase extraction returns the main topics in each response, giving analysts a ranked set of terms for the word cloud without any labeled examples. The sentiment feature in option B reports tone, not the topics being discussed. A custom text classification model in option C needs labeled examples for fixed categories, which the firm does not have and does not want. Language detection in option D identifies the language and says nothing about the subject matter.

Sample question 2 — Computer Vision Workloads

A property portal attaches an estimated sale price to every listing photo it publishes. The data science group has three years of completed sales, each with a numeric amount, and wants Azure Machine Learning to output a dollar figure for each new listing rather than a category such as budget, mid-range, or luxury. Which model type should the group train?

A. A classification model that assigns each listing to a price band such as low, medium, or high.
B. A regression model that predicts a continuous numeric price from listing features.
C. A clustering model that groups similar neighborhoods without using the sale amount.
D. A forecasting model that projects the number of listings arriving next month.

Answer: B. Option B is correct because regression predicts a continuous numeric price from listing features, which matches the portal's need for a dollar figure. A regression model learns from the completed sale amounts and returns a value that can be displayed directly. Option A produces a price band rather than a number, so it cannot supply the exact estimate the portal wants. Option C groups properties by similarity and never uses the sale amount as a target. Option D projects future listing counts, not the price of a specific property.

Sample question 3 — Natural Language Processing

An IT service desk already maintains a public FAQ page with several hundred question-and-answer pairs and wants a bot that replies with the closest matching answer, including a short follow-up when confidence is low. Staff will keep editing the FAQ source, and the team would rather import that existing content than label intents and entities from scratch. Which service best fits this requirement?

A. Conversational language understanding, trained with intents and labeled entities for each FAQ topic.
B. Custom named entity recognition, trained to tag products and error codes inside the FAQ text.
C. Custom question answering, which indexes the FAQ pairs and returns the best matched answer.
D. Azure AI Translator, which translates the FAQ entries into every supported language.

Answer: C. Option C is correct because custom question answering indexes the existing FAQ pairs and returns the closest matched answer, with the ability to ask a follow-up when the match is weak. Conversational language understanding, option A, needs intents and labeled entities and does not serve an answer from a knowledge base on its own. Custom named entity recognition, option B, only tags spans and supplies no reply. Azure AI Translator, option D, changes language rather than answering the question.

How to use this AI-900 practice set

Answer the 10 free questions without notes. Guessing is more useful than looking the answer up — the wrong answers tell you which domain to revisit.

Read every explanation, including the ones you got right. Note the reasoning for each distractor, not just the correct option.

Drill by domain, starting with Natural Language Processing (25%), then take a mixed set so you practise switching between topics.

Finish with a timed practice exam. Aim to be consistently above the passing score before you book the real AI-900 exam.

AI-900 practice questions FAQ

How many free AI-900 practice questions can I try?

You can answer 10 AI-900 practice questions free without creating an account. Each one is graded instantly with a written explanation. A free account keeps your progress, and the full bank (300+ questions) unlocks with a subscription.

What topics does the AI-900 exam cover?

The AI-900 exam is organised into 5 domains: Natural Language Processing (25%), AI Workloads and Considerations (20%), Machine Learning Principles (20%), Generative AI (20%), Computer Vision Workloads (15%). The weighting above shows how much of the exam each domain represents, so budget study time in the same proportion.

How long is the AI-900 exam and how is it scored?

The AI-900 exam has approximately 40 questions in 60 minutes, with a passing score of 700. Practice sets here are scored the same way as the real exam so you can see whether you are above or below the line before test day.

How should I start studying for Azure AI Fundamentals (AI-900)?

Start with Natural Language Processing, the highest-weighted domain, then work down the list. Answer a mixed set after each domain pass — mixing domains forces you to identify what a question is asking before recalling the content, which is what the real exam does.

Are these the same questions that appear on the real AI-900 exam?

No. These are original practice questions written against the published AI-900 exam objectives. They mirror the style, difficulty, and domain weighting of the real exam, but the actual exam content is confidential and is not reproduced here.

Microsoft AI-900 Practice Exams

Available Practice Exams

Your Progress

Exam Domains: