Microsoft AI credentials prep: AI-900 and Copilot lessons, quizzes and a certificate
Microsoft's AI credentials range from AI-900 Azure AI Fundamentals — the entry-level exam that proves you understand AI concepts and Azure AI services — to the Copilot skills that matter in everyday Microsoft 365 work. This free prep course covers both tracks in plain English, with practice questions written in the style of the real exam.
Entry exam
AI-900: Microsoft Azure AI Fundamentals
Format
Roughly 40-60 multiple-choice questions, about 45 minutes
Cost
Around USD 99 exam fee (varies by country)
Prerequisites
None; basic cloud familiarity helps
Also covered
Microsoft 365 Copilot skills for Word, Excel, Outlook and Teams
- Business and IT staff who need a recognised, entry-level AI certification
- Teams rolling out Microsoft 365 Copilot who need practical skills fast
- Career changers who want a structured, examinable AI credential
- Anyone preparing for AI-900 who wants free practice questions
The five AI workloads
AI-900 organises everything around workload types: machine learning (predicting values or categories from data), computer vision (images and video), natural language processing (text and speech), document intelligence and knowledge mining (extracting structure from forms and documents), and generative AI (creating new content). Exam questions usually ask which workload fits a described scenario.
Machine learning types
Regression predicts a number (tomorrow's sales). Classification predicts a category (spam or not). Clustering groups similar items with no labels. Regression and classification are supervised — they need labelled examples; clustering is unsupervised.
Features, labels and training
Features are the inputs, the label is the answer you want predicted. You train on one portion of the data and validate on data the model has not seen, so you can tell whether it learned the pattern or memorised the examples.
Practice quiz
1. Predicting next month's revenue as a number is which type of machine learning?
2. Grouping customers into similar segments without labels is:
3. Reading key fields from scanned invoices is best described as:
Vision, language, speech
Azure AI Vision handles image analysis, OCR and object detection; Custom Vision trains on your own labelled images. Azure AI Language covers sentiment, key phrases, entity recognition, summarisation and question answering. Azure AI Speech does speech-to-text, text-to-speech and translation.
Document Intelligence and Search
Azure AI Document Intelligence extracts fields from invoices, receipts, IDs and custom forms. Azure AI Search powers knowledge mining — indexing your content so an app, or a generative model, can retrieve the right passage before answering.
Azure OpenAI and Machine Learning studio
Azure OpenAI Service provides generative models with enterprise controls. Azure Machine Learning is the workspace for training and deploying your own models, including automated machine learning for people who do not write code.
Practice quiz
1. You need to extract totals and dates from thousands of supplier invoices. Which service?
2. Which service would you use to detect the sentiment of customer reviews?
3. To train an image classifier on your own labelled product photos you would use:
The six principles
Fairness, reliability and safety, privacy and security, inclusiveness, transparency, and accountability. Exam questions describe a scenario and ask which principle it illustrates, so learn the distinctions rather than the list order.
Telling them apart
Fairness is about equal treatment across groups. Reliability and safety is about consistent, safe behaviour including under unexpected input. Privacy and security is about protecting data. Inclusiveness is about people with different abilities and backgrounds being able to benefit. Transparency is about people understanding how and why a system decided. Accountability is about humans remaining answerable for outcomes.
Practice quiz
1. A loan model approves one demographic far less often. Which principle is at risk?
2. Explaining to a user which factors drove a decision supports which principle?
3. Designing captions and screen-reader support primarily supports:
Copilot works on your own content
Copilot is grounded in the files, mail, chats and meetings you already have access to through Microsoft Graph. It never surfaces content you had no permission to see — which also means messy, mislabelled or badly stored files produce weak answers.
Where it saves real time
Summarising a long Teams meeting into decisions and owners, drafting a document from existing files, finding the thread where something was agreed, turning a spreadsheet into trends and charts, and clearing an Outlook backlog with drafted replies.
Prompting Copilot well
Reference specific files, people and dates: 'Summarise the decisions in the Q3 planning meeting from last Tuesday and list owners' beats 'summarise my meetings'. Then verify names, figures and commitments before acting on them.
Practice quiz
1. What does Microsoft 365 Copilot use to ground its answers?
2. Which prompt gets the most useful Copilot result?
How AI-900 questions behave
Most questions are scenario-to-service or scenario-to-principle mappings, plus some drag-and-drop and case-style items. There is no penalty for guessing, so answer everything. Eliminate the two obviously wrong services first — that usually leaves a clear winner.
Study plan that works
Two weeks is enough for most people: read the Microsoft Learn path for each workload, do practice questions after each module, and note every question you get wrong with the reason. Review only that error list in the final two days.
After the exam
Pick one process in your own team — invoice handling, support triage, meeting follow-ups — and rebuild it with the service or Copilot feature you just learned. The credential proves knowledge; the rebuilt process is what gets noticed.
Practice quiz
1. You are unsure of an AI-900 answer. What should you do?
2. Most effective final revision step?
Six mixed questions across every module. Score 70% or higher to unlock your certificate.
1. Detecting whether an email is spam is which machine learning type?
2. Which Azure service converts spoken audio into text?
3. A system behaves unpredictably with unusual input. Which responsible AI principle applies?
4. Copilot gives a weak answer about a project. The most likely cause is:
5. Which workload fits 'index our policy documents so staff can ask questions'?
6. In supervised learning, the label is:
Pass every module quiz and the final exam to unlock your verified certificate — with a unique ID and a public page anyone can use to confirm it is genuine.