AI-901 study guide: Azure AI Fundamentals exam domains, plan and common mistakes

Written and reviewed by the DevOpsCode Lab Team. Last updated 19 November 2026.

AI-901 is a Microsoft fundamentals exam on artificial intelligence in Azure. It checks two things: that you understand common AI concepts and what each kind of AI workload can do, and that you can build simple AI solutions with Microsoft Foundry. It is a newer exam, so read the official Microsoft exam page for the current skills outline before you start.

What the exam tests

AI-901 mixes concept questions with practical ones. Concept questions describe a business need (“summarize support tickets”, “read text from scanned forms”, “flag harmful output”) and ask which kind of AI fits. Practical questions ask how you would do it in Microsoft Foundry: which model to pick, how to deploy and test it, how to ground it on your own data and how to keep its output safe. The second domain carries more weight, so hands-on time in the Foundry portal pays off.

Domains and how much each counts

Domain Weight What to be able to do Questions in our bank
Identify AI concepts and capabilities 43% Recognize generative AI, natural language processing, speech, computer vision and information extraction workloads; explain machine learning basics such as training, inference and evaluation; describe responsible AI principles (fairness, reliability and safety, privacy and security, inclusiveness, transparency, accountability) 93
Implement AI solutions by using Microsoft Foundry 57% Create a Foundry project; choose and deploy models from the model catalog; test prompts in the playground; write system prompts; ground responses on your data; build simple agents; apply content filters and evaluate output 122

Implementation is more than half of the exam. Concepts still matter, because a wrong idea of what a workload does leads to the wrong Foundry choice.

Exam format

Our practice tests use 45 questions in 45 minutes with a 70% pass mark. Microsoft reports results for its exams on a 1–1000 scale with 700 to pass. Because AI-901 is new, its question styles and length may change; check the official Microsoft exam page for the current format, price and objectives before you book.

Who should take it

  • People new to AI who want a structured, vendor-specific starting point.
  • Developers, analysts and administrators who will build or support AI features in Azure.
  • Business and technical staff who need to judge which AI workloads fit which problems.

No data science background is required. Basic comfort with the Azure portal and with reading short code samples helps in the Foundry domain.

A 2-week study plan

Week Focus Practice
1 AI concepts: workload types, machine learning basics, how large language models work at a high level, tokens and prompts, responsible AI principles Practice mode on the concepts domain; take one full test at the end of the week as a baseline
2 Microsoft Foundry: projects, model catalog, deployments, playground, system prompts, grounding on your data, agents, content filters, evaluation Domain quiz for Foundry; then two full timed tests and the mistakes quiz; book the exam when you score 80%+ twice

Create a Foundry project in a free or sandbox Azure subscription, deploy one chat model and test the same prompt with and without a system message. That one lab covers a large share of the practical questions.

Topics people often get wrong

  • Training versus inference. Training builds or adjusts a model from data; inference is using a trained model to produce output. Calling a deployed model from an app is inference.
  • Grounding versus fine-tuning. Grounding (retrieval-augmented generation) adds your data to the prompt at run time and keeps it current; fine-tuning changes the model weights with extra training. Most scenarios that say “answer from our documents” want grounding.
  • System prompt versus user prompt. The system prompt sets the model’s role, rules and tone for the whole conversation; the user prompt is the individual request.
  • Classification versus extraction. Classification assigns a label to a whole item; extraction pulls specific fields such as names, dates or totals out of it.
  • Content filters versus responsible AI principles. Principles describe goals; content filters are a concrete control that blocks or flags harmful input and output. Questions about stopping harmful responses want the control, not the principle.
  • Model versus agent. A model generates a response to a prompt; an agent uses a model plus instructions and tools to carry out multi-step tasks.

How to use the practice tests

Start with the free AI-901 questions to see the style. Then take one full timed test in exam mode for a baseline and look at the score by domain. Spend most of your time in the weaker domain with domain quizzes, and finish with the mistakes quiz before each new full test. The readiness number tells you when your recent scores hold above the pass mark across both domains.

Frequently asked questions

How hard is AI-901? It is a fundamentals exam. People who already use Azure often need about a week; newcomers to AI usually need two.

Do I need to write code? You should be able to read short code samples and understand what a call to a deployed model does. The exam does not ask you to build an application from scratch.

Does the certification expire? No. Microsoft fundamentals certifications do not expire, so there is no renewal assessment.

Are these the real exam questions? No. Every question is written by us for the published skills outline. We never use leaked or recalled exam content.

DevOpsCode Lab is not affiliated with or endorsed by Microsoft. Exam names are trademarks of their owners. Check the official Microsoft exam page for the current price, format and objectives before you book.