7.1. Exam Strategy

The AI-901 is roughly 40-60 questions in about 45 minutes, with a passing score of 700 out of 1000. That's a little under a minute per question on average, so pace matters but isn't punishing.

Time management. Make one pass answering everything you know quickly. Don't stall on a hard scenario — flag it and move on. A blank question and a wrong question score the same, so never leave anything unanswered; eliminate what you can and make your best choice before time runs out.

Flag and return. Use the review/flag feature for anything you're unsure of. Questions later in the exam sometimes jog your memory for earlier ones. Budget the last 5-7 minutes for flagged items.

Read the direction, not just the keywords. Many distractors share the scenario's vocabulary. The deciding detail is usually the direction of the task (audio-to-text vs. text-to-audio), the workload (understanding vs. generation), or the principle at stake (transparency vs. accountability). Phase 3 and Phase 5 traps live here.

Watch for "best" and "most appropriate." When several answers are technically possible, the exam wants the one that fits the constraints — usually the smallest/cheapest model that meets the need, the prebuilt analyzer over a custom one, or a plain model call instead of an agent when no actions are required.

Code questions. For Python snippets, identify the three moves: project client (setup) → OpenAI-compatible client (work) → send input/read output. A connection-string client signals the older Classic pattern; an endpoint + DefaultAzureCredential is current.