About these questions
These five questions are original. We wrote them to match the style of the AI-200 skills measured, with a little extra weight on the data management domain because it is the largest. They are not real exam questions, and the real exam will word things differently. They follow the skills list from Microsoft’s study guide dated October 7, 2026.
How to use these
Try each question without looking at your notes, and pick an answer before you open the explanation. Then read the explanation even when you were right, because the reason the closest wrong option fails is where the learning is. Note any domain where you slipped and revisit it on the domains explained page.
Question 1 (Develop containerized solutions on Azure)
A team stores its inventory API image in an Azure container registry. The Dockerfile starts from a company base image kept in the same registry under a stable tag, and the security team patches that base image every month. The API image must be rebuilt after each patch, even when nobody has pushed new API code. Which Azure Container Registry Tasks trigger should you use?
- A. A source code commit trigger
- B. A timer trigger set to the first day of each month
- C. A base image update trigger
- D. A pull request trigger
Show the answer and explanation
Correct answer: C. A base image update trigger watches the base image named in the Dockerfile and rebuilds the dependent image when that base image changes in the registry. That matches the need exactly: the rebuild follows the patch, not a code change. The timer trigger is the closest wrong option. It rebuilds on a calendar, so it could run before the patch lands or run when nothing changed. Commit and pull request triggers react to Git activity, which the scenario says will not happen.
Question 2 (Develop AI solutions by using Azure data management services)
A support team is adding a container in Azure Cosmos DB for NoSQL that will hold about eight million article embeddings of 1,536 dimensions. Searches must stay fast and use few request units. Missing an occasional near-match is acceptable. Which vector index type should you configure?
- A. flat
- B. diskANN
- C. quantizedFlat
- D. HNSW
Show the answer and explanation
Correct answer: B. The diskANN index performs approximate nearest-neighbor search and is designed for large vector sets, giving low latency and low request unit cost in exchange for a small loss of recall. Microsoft’s guidance puts it ahead of the other types when a search covers more than about 50,000 vectors. The closest wrong option is quantizedFlat. It compresses vectors but still scans them, so it suits smaller sets. A flat index is an exact brute-force search, and HNSW is not one of the vector index types offered by Cosmos DB for NoSQL.
Question 3 (Connect to and consume Azure services)
A billing service reads invoice messages from an Azure Service Bus queue that uses default settings and the peek-lock receive mode. One malformed message makes the consumer abandon it on every attempt. What eventually happens to that message?
- A. It moves to the dead-letter subqueue after 10 delivery attempts.
- B. It is redelivered until its time to live expires.
- C. It moves to the dead-letter subqueue after the first abandon.
- D. It is deleted after the lock expires five times in a row.
Show the answer and explanation
Correct answer: A. Each abandon, or lock expiry, increases the message’s delivery count. When the count exceeds the maximum delivery count, which is 10 by default, Service Bus moves the message to the dead-letter subqueue with the reason MaxDeliveryCountExceeded, where you can inspect it. Option B is the closest wrong answer: time to live is a separate limit and does not keep a failing message cycling forever. Service Bus does not dead-letter on the first failure, and it does not silently delete a message because locks expired.
Question 4 (Secure, monitor, and troubleshoot Azure solutions)
You query a workspace-based Application Insights table named AppRequests. A dashboard must show how many different users called the API on each day of the last 14 days. Some users call the API hundreds of times a day. The query already starts with `AppRequests | where TimeGenerated > ago(14d)`. Which line should come next?
- A. `| summarize count() by UserId, bin(TimeGenerated, 1d)`
- B. `| distinct UserId | count`
- C. `| summarize count() by bin(TimeGenerated, 1d)`
- D. `| summarize dcount(UserId) by bin(TimeGenerated, 1d)`
Show the answer and explanation
Correct answer: D. The dcount function counts distinct values, and bin(TimeGenerated, 1d) groups rows into one-day buckets, so you get one number of unique users per day. Option C is the closest wrong answer: it counts requests per day, so a heavy user would be counted hundreds of times. Option A returns a row per user per day rather than one total per day. Option B counts unique users once across all 14 days, with no daily breakdown.
Question 5 (Develop AI solutions by using Azure data management services)
A team creates an Azure Managed Redis instance to store embeddings in a RediSearch vector index. The RediSearch module is enabled when the instance is created, and the index data must never be evicted to free memory. Which two settings should the team choose? Choose two.
- A. The OSS clustering policy
- B. The Enterprise clustering policy
- C. The allkeys-lru eviction policy
- D. The NoEviction eviction policy
- E. The volatile-lru eviction policy
Show the answer and explanation
Correct answers: B and D. Microsoft’s documentation says the RediSearch module requires the Enterprise clustering policy and the NoEviction eviction policy. NoEviction also fits the requirement that index data is never removed to make room. The OSS clustering policy is the closest wrong option for the first choice, because it is the other clustering mode but does not work with RediSearch. The allkeys-lru and volatile-lru policies both remove data under memory pressure, which the scenario forbids.
More guides for AI-200
- AI-200 Study Plan: 4-Week and 8-Week
- Is the AI-200 Exam Hard?
- AI-200 Prerequisites: What to Know First
- AI-200 Renewal and Retakes Explained
- Is AI-200 Worth It? An Honest Look
- AI-200 Cheat Sheet: Key Facts to Review
- AI-200 vs AI-300: Which Exam First?
- AI-103 vs AI-200: Which Exam Should You Take?
Next step
If these felt comfortable, read the exam format page and think about booking. If a domain felt shaky, try more questions in the AI-200 practice tests.