AzureJuly 7, 202612 min read

AI-102 Practice Questions (2026): 10 Free Azure AI Engineer Questions with Answers

Ten exam-style questions across all six AI-102 domains — each with the correct answer and a plain-English explanation on click. Score yourself at the end to see if you are exam-ready.

40–60Questions
100 minDuration
700/1000To pass
6Domains
$165USD fee

The AI-102: Designing and Implementing a Microsoft Azure AI Solution exam is the hands-on gate for the Azure AI Engineer Associate badge. It rewards people who have actually wired up Azure AI services — keys and endpoints, Vision, Language, Document Intelligence, Azure AI Search, and Azure OpenAI — not people who only memorised definitions.

Below are 10 free AI-102 practice questions written in the same style as the real exam, spread across all six domains of the 2026 blueprint. Read each one, pick your answer, then hit Show answer & explanation. Keep score — there is a readiness check at the bottom.

The 10 questions Domain coverage Score yourself FAQ
How to use this: answer honestly before revealing. The explanation matters more than the letter — if you got it right for the wrong reason, treat it as wrong and revisit that service in the docs.

10 AI-102 practice questions

Q1EasyPlan & manage

You are building an app that calls both the Vision and Language capabilities of Azure AI. You want one key and one endpoint, and a single line item on the bill. Which resource should you provision?

  1. A separate Computer Vision resource and a separate Language resource
  2. A single multi-service Azure AI services resource
  3. An Azure Machine Learning workspace
  4. An Azure AI Foundry hub with no connected services
Show answer & explanation
Correct: B

A multi-service Azure AI services resource exposes many capabilities (Vision, Language, Speech, and more) behind one key and endpoint, with consolidated billing. Separate single-service resources (A) each get their own key/endpoint and bill line — the opposite of what is asked. An Azure ML workspace (C) is for custom model training, not for calling prebuilt AI services.

Q2MediumPlan & manage

Security requires that an Azure AI services resource be reachable only from your virtual network, never over the public internet. What should you configure?

  1. Rotate the account keys every 24 hours
  2. Switch authentication to Microsoft Entra ID only
  3. Create a private endpoint and disable public network access
  4. Add the resource to an Azure Front Door profile
Show answer & explanation
Correct: C

A private endpoint gives the resource a private IP inside your VNet; combined with disabling public network access, traffic never traverses the public internet. Key rotation (A) and Entra ID auth (B) are good hygiene but do not restrict the network path. Front Door (D) is a public global entry point — the opposite of private.

Q3MediumContent moderation

A community app must automatically flag user text and images for hate, sexual, violent, and self-harm content, with a severity level for each category. Which service fits best?

  1. Azure AI Content Safety
  2. Azure AI Language sentiment analysis
  3. Azure AI Vision tags
  4. Azure AI Translator profanity filtering
Show answer & explanation
Correct: A

Azure AI Content Safety analyses text and images across four harm categories — Hate, Sexual, Violence, Self-harm — returning a severity level per category. Sentiment (B) measures positive/negative tone, not harm. Vision tags (C) label objects. Translator profanity (D) only masks swear words during translation.

Q4EasyComputer vision

You need to extract both printed and handwritten text from scanned images and photos. Which Azure AI Vision capability should you call?

  1. Image Analysis – tags and captions
  2. The Read OCR capability
  3. Spatial Analysis
  4. Background removal
Show answer & explanation
Correct: B

The Read OCR capability of Azure AI Vision extracts printed and handwritten text and returns bounding boxes for lines and words. Tags/captions (A) describe the scene, not the text. Spatial Analysis (C) tracks people in video. Background removal (D) is an image-editing operation.

Q5MediumComputer vision

In Azure AI Custom Vision you must not only identify that a product appears in a photo but also return where it is, as a bounding box. Which project type do you create?

  1. Image classification (multiclass)
  2. Image classification (multilabel)
  3. Object detection
  4. Semantic segmentation
Show answer & explanation
Correct: C

Object detection returns a label and bounding-box coordinates for each detected item. Both classification modes (A, B) only assign labels to the whole image with no location. Custom Vision does not offer semantic segmentation (D).

Q6HardNLP

You are modernising a chatbot that used LUIS to detect user intent and extract entities. LUIS is retired. Which Azure AI Language feature replaces it?

  1. Custom Question Answering
  2. Conversational Language Understanding (CLU)
  3. Key phrase extraction
  4. Named entity recognition (prebuilt)
Show answer & explanation
Correct: B

Conversational Language Understanding (CLU) is the successor to LUIS: you train intents and entities from example utterances. Custom Question Answering (A) replaces the old QnA Maker for FAQ-style bots. Key phrase extraction (C) and prebuilt NER (D) are unsupervised and cannot learn your app-specific intents.

Q7MediumNLP / Speech

A medical dictation app mis-transcribes drug names during real-time speech-to-text. You must improve accuracy for that specialist vocabulary without hurting general recognition. What is the best approach?

  1. Increase the audio sample rate to 48 kHz
  2. Switch to text-to-speech instead
  3. Train a Custom Speech model (or add a phrase list) with the domain terms
  4. Enable profanity masking
Show answer & explanation
Correct: C

Custom Speech lets you adapt the base speech-to-text model with domain audio/text, and a phrase list can boost specific terms with no training. Sample rate (A) helps only if audio was under-sampled. Text-to-speech (B) is the reverse task. Profanity masking (D) is unrelated.

Q8EasyDocument Intelligence

You must pull the vendor, invoice number, line items, and totals out of thousands of supplier invoices with minimal custom work. Which Azure AI Document Intelligence option is fastest?

  1. Train a custom neural model from 500 labelled samples
  2. Use the prebuilt invoice model
  3. Use the Read model and parse fields with regex
  4. Use a general classification model
Show answer & explanation
Correct: B

The prebuilt invoice model already extracts vendor, invoice ID, dates, line items, and totals out of the box — no labelling or training. A custom model (A) is only worth it for non-standard layouts. Read + regex (C) is brittle. A classification model (D) sorts documents but does not extract fields.

Q9HardKnowledge mining

In Azure AI Search you want to OCR images, detect key phrases, and translate documents automatically as they are ingested from Blob Storage into the index. Which component orchestrates that enrichment?

  1. A scoring profile
  2. A synonym map
  3. A skillset attached to an indexer
  4. A semantic ranker
Show answer & explanation
Correct: C

A skillset is the AI-enrichment pipeline (OCR, key phrases, translation, custom skills) that an indexer runs over source data before writing to the index. Scoring profiles (A) tune relevance. Synonym maps (B) expand queries. The semantic ranker (D) re-ranks results at query time — none of them enrich content at ingestion.

Q10HardGenerative AI

A support bot built on Azure OpenAI must answer only from your internal knowledge-base documents and cite them, instead of inventing answers. What is the recommended pattern?

  1. Raise the temperature to 1.0 for more creative answers
  2. Fine-tune GPT on the whole internet
  3. Use Azure OpenAI On Your Data with Azure AI Search (retrieval-augmented generation)
  4. Increase max tokens and hope for the best
Show answer & explanation
Correct: C

Grounding the model on your own documents via Azure OpenAI On Your Data + Azure AI Search (RAG) retrieves relevant passages and forces answers — with citations — from that context, sharply cutting hallucination. Higher temperature (A) increases invention. Broad fine-tuning (B) does not add your private data. More tokens (D) does nothing for grounding.

What these questions cover

The 10 questions are weighted to mirror the real AI-102 blueprint, so your score here is a rough proxy for the live exam. These are the six domains and their approximate exam weights:

Natural language processing30–35%
Plan & manage an Azure AI solution20–25%
Computer vision15–20%
Knowledge mining & Document Intelligence10–15%
Generative AI solutions10–15%
Content moderation (Content Safety)10–15%

Score yourself

Count how many of the 10 you got right before revealing the answer. Then read the band you land in honestly — the goal is a real pass, not a good feeling.

0–5Not ready yet. You are guessing on core services. Rebuild fundamentals in Vision, Language, and Azure OpenAI before booking.
6–8Close. You know the services but slip on edge cases (private endpoints, RAG, skillsets). Drill a full timed bank and target weak domains.
9–10Exam-ready signal. Confirm with a couple of full-length timed mocks under exam conditions, then book with confidence.

Want the full AI-102 question bank?

These 10 are a taster. The ExamCert AI-102 app runs hundreds of exam-style questions with explanations, timed mocks, and weak-domain tracking — the real exam interface, on your phone.

App Store Google Play Free practice test AI-102 exam guide

AI-102 exam FAQ

How many questions are on the AI-102 exam?

Typically 40 to 60 questions in about 100 minutes, mixing multiple choice, multi-select, drag-and-drop, hotspot, and one or more case studies. Case studies lock once you leave them, so read the scenario fully before answering.

What score do I need to pass AI-102?

A scaled 700 out of 1000. Because scoring is scaled, 700 is not the same as 70% correct — harder and case-study items carry more weight.

Is AI-102 still current in 2026?

Yes. AI-102 is the active Azure AI Engineer Associate exam. The refreshed blueprint added a content-moderation domain (Azure AI Content Safety) and expanded generative AI with Azure OpenAI and retrieval-augmented generation, so make sure your prep covers both.

Are practice questions enough to pass?

They are essential but not sufficient. AI-102 is hands-on — provision the services in a free/sandbox subscription, call the APIs, and use practice questions to find and close weak domains.

ExamCert Team — we build exam-style practice banks and prep apps for 90+ IT certifications. Questions here are original, written to match the AI-102 objectives; they are not real exam items.

Related: AI-102 exam guide · AI-102 complete study guide · Free practice tests