Azure AI-901 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 Certified: Azure AI Fundamentals (AI-901) topics such as AI workload recognition, responsible AI, Microsoft Foundry basics, language, speech, vision, document extraction, and simple cloud resource choices. The prompts stay fundamentals-oriented, but they still ask you to choose the best fit for a realistic scenario.

Where these questions fit in the AI-901 guide

The sample set below is part of the Microsoft AI-901 guide path:

• Microsoft certification hub
• AI-901 exam guide
• AI-901 cheat sheet
• AI-901 study plan
• AI-901 FAQ
• AI-901 resources
• AI-901 glossary

AI-901 fundamentals sample questions

Work through each prompt before opening the explanation. AI-901 is a fundamentals exam, but the strongest answers still connect the workload, responsible AI concern, and Azure capability.

---

Question 1

Topic: Responsible AI in a loan triage app

A bank wants to use an AI model to help triage loan applications. The model will not make the final decision, but it will rank applications for employee review. The team is most concerned that some applicant groups may be ranked unfairly because of patterns in historical data. What should the team prioritize?

• A. Increase the model size so it can learn more subtle patterns.
• B. Evaluate the system for fairness, monitor outcomes across relevant groups, and keep human review in the process.
• C. Remove all logging so applicants cannot be associated with AI-generated rankings.
• D. Use only a chatbot interface so the model does not expose a numeric score.

Best answer: B

Explanation: The scenario is about fairness and accountability. A fundamentals-level answer should identify that AI behavior must be evaluated across groups, monitored after deployment, and kept under human oversight for a sensitive workflow.

Why the other choices are weaker:

• A may improve capability, but it does not directly address unfair ranking risk.
• C weakens transparency and auditability instead of improving responsible use.
• D changes the interface but does not evaluate or reduce biased outcomes.

What this tests: Recognizing responsible AI concerns when an AI system influences a high-impact human decision.

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

---

Question 2

Topic: Choosing a language capability

A support manager wants a dashboard that groups recent customer comments by sentiment and extracts common topics. The team does not need long-form generated responses. Which type of AI capability is the best fit?

• A. Language analysis for sentiment and key phrase extraction.
• B. Image classification because comments can be grouped into categories.
• C. Speech transcription because customer comments are text.
• D. Generative chat because every text problem requires a conversational model.

Best answer: A

Explanation: The requirement is to analyze existing text, not generate new text. Sentiment and key phrase extraction match the task directly and keep the solution simpler than a broad chat workflow.

Why the other choices are weaker:

• B applies to images, not text comments.
• C turns audio into text, but the comments are already text.
• D overgeneralizes generative AI and misses the simpler analysis capability.

What this tests: Mapping a workload to the correct AI category instead of choosing a generative model by default.

Related topics: Text analysis; Sentiment; Key phrases; Service fit

---

Question 3

Topic: Extracting fields from invoices

An accounting team receives invoice PDFs from many suppliers. They need to extract invoice number, vendor, date, total amount, and line items into a table for review. Which approach best matches the requirement?

• A. Use document extraction so structured fields can be pulled from the files and reviewed.
• B. Use speech recognition because invoices contain human language.
• C. Use anomaly detection because every invoice total is a prediction.
• D. Use translation because supplier names may come from different regions.

Best answer: A

Explanation: The clue is structured extraction from documents. A document extraction flow is built for pulling fields and table-like information from forms, receipts, invoices, and similar files.

Why the other choices are weaker:

• B handles audio-to-text, not PDF field extraction.
• C may detect unusual values later, but it does not extract invoice fields.
• D may be useful for language conversion, but it does not solve the primary extraction task.

What this tests: Distinguishing document extraction from text, speech, vision, and anomaly workloads.

Related topics: Document extraction; Structured fields; Review workflow; Workload category

---

Question 4

Topic: Using a deployed model from an app

A student builds a simple web app that sends prompts to a model deployment in Microsoft Foundry and displays the response. The app must authenticate to the model endpoint and avoid placing secrets directly in source code. Which pattern is strongest?

• A. Hard-code the endpoint key into the JavaScript file so the browser can call the model directly.
• B. Call the model from a backend service that uses managed identity or a protected secret store, then return the result to the app.
• C. Put the key in a public README file and ask users not to share it.
• D. Disable authentication because the app is only a prototype.

Best answer: B

Explanation: Even at the fundamentals level, the exam can test basic cloud security habits. A backend can protect credentials and call the model endpoint without exposing secrets in client-side code.

Why the other choices are weaker:

• A exposes the key to anyone who can inspect the page source or network traffic.
• C publishes the secret and relies on policy instead of a technical control.
• D ignores the access-control requirement.

What this tests: Understanding the basic implementation boundary between app code, model endpoints, and protected credentials.

Related topics: Microsoft Foundry; Model endpoint; Authentication; Secure app design

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.