10 Common AI-901 Exam Mistakes UK Candidates Make
Preparing for the Microsoft AI-901 Azure AI Fundamentals exam? Learn the ten most common mistakes UK candidates make and get practical tips to pass on your first attempt.
Examinotion Team

Last updated: 9 May 2026
The Microsoft AI-901 exam catches many UK candidates off guard. Although it earns the same Microsoft Certified: Azure AI Fundamentals badge as the retiring AI-900, the underlying exam is structurally different. Domain 2 alone accounts for 55-60% of the marks and is built around hands-on work in Microsoft Foundry, not the passive recognition that AI-900 rewarded.
After reviewing the official skills outline, the Microsoft Tech Community launch announcement, and feedback from candidates sitting AI-901 during its current beta window, we have identified the ten most common mistakes that lead to failure, and how you can avoid them.
1. Treating AI-901 Like a Refreshed AI-900
The single most expensive mistake is assuming that AI-901 is just AI-900 with new branding. It is not.
The reality: AI-900 was a description-led exam. You could pass it by reading documentation and watching short videos because most questions asked you to recognise what an Azure AI service does. AI-901 has been redesigned around implementation. Microsoft's official AI-901 skills outline, published on 15 April 2026, splits the syllabus into two domains:
- Domain 1: Identify AI concepts and responsibilities (40-45%)
- Domain 2: Implement AI solutions by using Microsoft Foundry (55-60%)
Every Domain 2 sub-skill begins with verbs like "deploy", "create", "build", or "implement". Reading-only preparation will leave you exposed across the larger half of the exam.
How to avoid this mistake: Treat AI-901 as a hands-on fundamentals exam from day one. Plan your study so that roughly 60% of your time is spent inside the Foundry portal and SDK, and only 40% on responsible AI and workload theory.
2. Skipping Hands-On Microsoft Foundry Practice
You cannot pass AI-901 without being comfortable inside Microsoft Foundry at ai.azure.com. Reading the docs is not enough.
What Domain 2 actually asks you to do:
- Deploy a model and interact with it in the Foundry portal
- Create effective system and user prompts for generative AI models
- Create and test a single-agent solution in the Foundry portal
- Build lightweight applications that include text analysis, vision, speech, and information extraction
Each of those bullets is a verb-led task that expects portal time, not portal screenshots.
How to avoid this mistake: Create a free Azure account in week one of your study plan. Deploy at least one model, send a prompt in the chat playground, and create one trivial single-agent solution before you reach the halfway point of your preparation. If you have never opened the Foundry portal, you are not ready to sit AI-901.
3. Underestimating the Python Literacy Requirement
Microsoft's official exam page states that AI-901 candidates should arrive with "basic Python syntax and programming technique". This is a step up from AI-900, which had no programming expectation at all.
What "basic Python" actually means in the AI-901 context:
- Reading and running a short script someone else wrote
- Understanding what a function call looks like and how arguments are passed
- Comfort with
pip installand a virtual environment - Recognising what an SDK client object is doing in five to ten lines of code
You do not need to be a production Python engineer, but if pip install azure-ai-projects and a five-line chat completion call look intimidating to you, that is a gap to close before exam day.
How to avoid this mistake: If you are new to Python, spend the first weekend of your study plan on the free Microsoft Learn Python beginners path. Aim for fluency in reading short scripts, not at writing your own from scratch.
4. Treating Responsible AI as Definitions Rather Than Applied Scenarios
AI-900 used to test responsible AI as a single domain bullet with definition-style questions. AI-901 breaks responsible AI into the six Microsoft principles and asks you to apply them to scenarios.
The six principles you must know:
| Principle | Core question the exam tests |
|---|---|
| Fairness | Does the AI system treat all groups of users equitably? |
| Reliability and safety | Does the system perform consistently and fail safely under stress? |
| Privacy and security | Are user data and the model itself protected? |
| Inclusiveness | Does the system work for people with diverse abilities and backgrounds? |
| Transparency | Can users and operators understand how the system reaches its decisions? |
| Accountability | Who is responsible for the system's behaviour and outcomes? |
What the exam actually asks: Questions are framed as scenarios, for example "A hiring tool consistently scores female applicants lower than male applicants with similar CVs. Which responsible AI principle is most directly violated?" Memorising the names is not enough; you need to apply them.
How to avoid this mistake: For each of the six principles, write your own one-line example of a violation, one mitigation, and one design artefact (a datasheet, a transparency note, or a fairness assessment). Practise answering scenario questions from at least one timed practice exam before booking your real sitting.
5. Studying With Outdated Azure AI Studio Vocabulary
A surprising number of UK candidates revise from older AI-900 study notes or third-party courses that pre-date the Microsoft Foundry rebrand. The vocabulary on AI-901 has moved on, and so have the underlying APIs.
Old terminology to retire from your notes:
| Old term | Current term used on AI-901 |
|---|---|
| Azure AI Studio | Microsoft Foundry portal |
| Azure AI Foundry (the brand) | Microsoft Foundry |
| Azure AI Services (the brand) | Foundry Tools |
| Assistants API (Agents v0.5 and v1) | Responses API (Agents v2) |
| Hub plus Azure OpenAI plus Azure AI Services | Single Foundry resource with projects |
Why this matters: AI-901 questions assume current Foundry vocabulary. If a question mentions a "Foundry project endpoint" and you only know about the old hub-plus-services pattern, you will struggle to recognise the right answer. The Microsoft Tech Community announcement Evolving the Microsoft Certified: Azure AI Fundamentals Certification makes the rebrand explicit.
How to avoid this mistake: Discard pre-2026 study materials unless they have been updated for AI-901. Stick to Microsoft Learn paths refreshed after April 2026, the official AI-901 study guide, and the What is Microsoft Foundry? overview page.
6. Never Running a Foundry SDK Chat Client
The AI-901 skills outline explicitly asks candidates to "create a lightweight chat client application by using the Foundry SDK". This is not an optional stretch goal, it is a syllabus requirement.
What candidates often do instead:
- Read the SDK quickstart without ever running it
- Watch a recorded walkthrough at 1.5x speed and tick the box mentally
- Assume the Foundry portal chat playground covers the SDK questions
None of those count. The exam asks you to recognise what a pip install, an authentication call, and a chat completion request look like in code. If you have never run them yourself, the multiple-choice answers will all look plausible.
How to avoid this mistake: Install the Foundry SDK in a virtual environment, paste the official quickstart into a Python file, and run it once. Watch the request go out, the response come back, and the cost appear in the Azure portal. The whole exercise takes under thirty minutes and pays for itself in the exam room.
7. Skipping Single-Agent Creation in the Portal
AI-901 only requires single-agent understanding (multi-agent is a higher-tier exam topic), but you still need to have created and tested one agent in the Foundry portal at least once.
What the exam tests:
- Recognising the role of an agent compared with a raw chat completion
- Knowing what tools and memory an agent uses
- Understanding the Responses API (Agents v2) shape and how it differs from the older Assistants API
- Building a lightweight client application that calls an agent
How to avoid this mistake: During week two of your study plan, create one trivial single-agent solution in the Foundry portal. Even an agent that does nothing useful is enough to lock in the workflow. You can delete it afterwards.
8. Neglecting Azure Content Understanding for Information Extraction
The smallest of Domain 2's four clusters is also the most overlooked. Microsoft's skills outline lists four sub-skills for information extraction:
- Extract information from documents and forms by using Azure Content Understanding in Foundry Tools
- Extract information from images by using Content Understanding
- Extract information from audio and video by using Content Understanding
- Build a lightweight application with information extraction capabilities by using Content Understanding
What candidates often miss: Many candidates focus heavily on generative AI in cluster 1, run a couple of vision and speech demos in clusters 2 and 3, then arrive at the exam having never opened Content Understanding. Information-extraction questions are scored exactly the same as everything else in Domain 2.
How to avoid this mistake: Run Content Understanding against three different inputs before exam day: one PDF or form, one image, and one short audio or video clip. The Foundry Tools section of the portal walks you through each. Knowing what the output JSON looks like is worth several scenario-recognition marks.
9. Ignoring Multimodal Model Exercises
Two AI-901 sub-skills require multimodal model use, not just multimodal awareness:
- Respond to spoken prompts by using a deployed multimodal model
- Interpret visual input in prompts by using a deployed multimodal model
Why this matters: A multimodal question on AI-901 will typically describe a real input, a spoken request to summarise a meeting, or an image attached to a chat prompt, then ask you which model class or deployment configuration is correct. Candidates who have only used text-in-text-out chat completions will hesitate.
How to avoid this mistake: During your speech and vision study weeks, deploy at least one multimodal model in the Foundry portal and send it both a voice clip and an image. The end-to-end exercise takes about twenty minutes and the muscle memory is hard to fake on exam day.
10. Relying Only on Reading Without a Timed Practice Exam
The official Microsoft study guide is essential, but it is not sufficient on its own. Microsoft does not yet publish a free practice assessment for AI-901 (those typically appear around eight weeks after a beta exam reaches general availability), so candidates often skip timed practice entirely.
Why this is a mistake:
- The exam interface and review screen are unfamiliar
- You do not know how long Microsoft questions actually take you to read carefully
- You have no realistic measure of where your weak domains sit
- You discover gaps in the exam room rather than in your preparation
How to avoid this mistake: Take at least one full-length, timed practice exam before booking your real sitting. Use it to identify weak sub-skills and revisit the corresponding Microsoft Learn modules. Examinotion's AI-901 practice exam gives you 40-question, exam-format tests with detailed explanations for every answer, at £12.99 with unlimited attempts until you pass.
AI-901 Exam Quick Facts
Before you book your exam, ensure you understand the format. Details below are confirmed as of 9 May 2026; always verify on the official AI-901 exam page when you book.
| Detail | Information |
|---|---|
| Exam code | AI-901 |
| Official title | Microsoft Azure AI Fundamentals |
| Certification earned | Microsoft Certified: Azure AI Fundamentals |
| Current status | Beta (opened 21 April 2026) |
| Passing score | 700 / 1,000 |
| Skill domains | 2 (concepts and responsibilities; Foundry implementation) |
| Format | Multiple choice, multi-select, scenario-based |
| Delivery | Pearson VUE test centre or OnVUE online proctored |
| Formal prerequisites | None |
| Recommended background | Basic cloud concepts, basic Python |
| Renewal | Fundamentals certifications do not expire |
Frequently Asked Questions
Is AI-901 harder than AI-900?
Yes. AI-900 could be passed largely from reading and watching videos. AI-901 expects you to have deployed a model in Microsoft Foundry, sent chat completions through the SDK, tested an agent, and run Content Understanding. It also assumes basic Python literacy. That said, AI-901 remains a fundamentals-tier exam with a standard 700 / 1,000 pass mark and no formal prerequisites. Most UK candidates with a modest technical background can prepare in three to four weeks of focused study.
Can I pass AI-901 without any Azure experience?
Technically yes, because there are no formal prerequisites, but it is much harder. Domain 2 is 55-60% of the exam and almost entirely hands-on work in Microsoft Foundry. Realistically, plan at least one week of Foundry portal and SDK practice before exam day. A free Azure account is enough; you do not need a paid subscription.
How much Python do I really need to pass AI-901?
Enough to read short scripts confidently, run a quickstart in a virtual environment, and recognise what an SDK client call looks like. You do not need to write production code. If you are completely new to Python, spend the first weekend of your study plan on a free beginner Python path before tackling the Foundry SDK.
What if I fail the AI-901 exam?
You can retake the exam 24 hours after your first attempt. Subsequent retake waiting periods may apply. Use the time between attempts to address the specific Domain 2 areas where you struggled, particularly any Foundry Tools cluster (text, speech, vision, or information extraction) where you scored low. Microsoft does not penalise you on transcript for a failed attempt.
Do I still need to take AI-901 if I already hold AI-900?
No. Microsoft confirms that fundamentals certifications do not expire, so your AI-900 credential remains valid indefinitely. Many candidates choose to re-earn Azure AI Fundamentals via AI-901 to signal currency with Microsoft Foundry, but it is not required. If you are weighing up whether to take AI-900 before it retires on 30 June 2026 or jump straight to AI-901, read our AI-900 vs AI-901 transition guide for a scenario-based recommendation.
How often does AI-901 content change?
Microsoft updates exam content regularly to reflect platform changes, and the Foundry stack is moving quickly. Check the official study guide for AI-901 for the current skills measured before you book. Anything not present in that document is unlikely to be on the exam.
Next Steps
Now that you know the ten common mistakes to avoid, build a structured study plan that addresses each one:
- Week 1: Read the official study guide end to end; complete the responsible AI principles with worked examples
- Week 2: Set up a free Azure account; deploy your first model in the Foundry portal; run the SDK quickstart
- Week 3: Work through every Domain 2 cluster (generative AI, text and speech, vision, information extraction)
- Week 4: Take a timed practice exam; revisit weak areas; book your sitting
For a fuller week-by-week structure, see our complete AI-901 exam guide. When you are ready to test yourself under exam conditions, Examinotion's AI-901 practice exam gives you 40-question, timed tests with detailed explanations at £12.99 with unlimited attempts until you pass.
Avoiding these ten mistakes will not guarantee a pass, but it will close the gaps that catch most UK candidates off guard.
Sources:
- Exam AI-901: Microsoft Azure AI Fundamentals (Microsoft Learn)
- Study guide for Exam AI-901 (Microsoft Learn, 15 April 2026)
- Microsoft Certified: Azure AI Fundamentals (Microsoft Learn)
- Evolving the Microsoft Certified: Azure AI Fundamentals Certification (Microsoft Tech Community, 15 April 2026)
- What is Microsoft Foundry? (Microsoft Learn)
- Get started with AI applications and agents on Azure (Microsoft Learn)
Related Articles

How to Prepare for AB-731 in 30 Days: A Structured Study Plan for AI Transformation Leaders
A structured, evidence-based 30-day study plan for the Microsoft AB-731 AI Transformation Leader exam, mapped to the official skills outline and built around free Microsoft Learn resources for UK candidates.

Microsoft Exam Retake Policy 2026: How to Pass After Failing First Time
Failed an AB-730, AB-731, AB-100, AB-900 or AI-901 exam? Microsoft lets you retake after 24 hours, then 14 days between attempts up to five tries in 12 months. Use your score report to fix the right two domains and book again on day 21 to 28.

AB-731 Case Study Questions: How to Tackle Microsoft AI Strategy Scenarios (2026)
A tactical guide to AB-731 case study and scenario questions. Learn the SCALE framework, work through a Microsoft AI strategy example, and walk into the exam with a pre-exam checklist for the AI Transformation Leader certification.