AWS AI practice questions with answers
Fifteen free AWS AI services practice questions with explained answers, written as short business scenarios. AWS AI services are pre-trained, API-based services for images, documents, text, speech, conversation and recommendations that need no model training. They are tested in the AWS Certified AI Practitioner (AIF-C01) and the AWS Certified Machine Learning Engineer Associate (MLA-C02) exams. All questions are original and written against the official outlines.
How to use these questions
Attempt each question before opening the answer. Name the input data type and the wanted output first, then match the service. Note the constraint word in the stem (“least development effort”, “no ML expertise”, “live”, “already recorded”), because it usually separates the right answer from a plausible one. The AWS AI services topic guide has the input-to-output cheat sheet and the list of services closed to new customers. Difficulty rises across the three sections: the first five questions match one input to one output, the middle five separate services that look alike, and the last five add constraints such as custom categories, availability and fewest services.
Warm-up: one input, one output
Question 1. A finance team receives supplier invoices as scanned PDFs and wants vendor name, invoice date and line items returned as structured fields, with no model training. Which service fits best?
- A. Amazon Textract
- B. Amazon Rekognition
- C. Amazon Comprehend
- D. Amazon Polly
Show answer
Answer: A
Amazon Textract extracts text, forms, tables and invoice fields (AnalyzeExpense) from scanned documents. Rekognition detects objects, faces and scene text but does not return document structure. Comprehend analyses text that is already digital. Polly turns text into speech.
Question 2. A webinar platform must show captions for a live speaker with a delay of only a few seconds. Which approach meets the requirement?
- A. Amazon Polly
- B. Amazon Transcribe with streaming transcription
- C. Amazon Translate
- D. Amazon Comprehend
Show answer
Answer: B
Amazon Transcribe supports streaming transcription, which returns text while the audio is still being spoken. Polly works in the opposite direction (text to speech). Translate converts text between languages and does not recognise speech. Comprehend analyses existing text and cannot process audio.
Question 3. A news app wants to read written articles aloud to users who prefer listening. Which service converts the article text into natural-sounding audio?
- A. Amazon Transcribe
- B. Amazon Lex
- C. Amazon Polly
- D. Amazon Rekognition
Show answer
Answer: C
Amazon Polly is the text-to-speech service. Transcribe does the reverse, converting audio to text. Lex builds conversational bots and is not needed to read an article aloud. Rekognition analyses images and video.
Question 4. An e-commerce company wants its English product descriptions available in German, French and Japanese, with the least development effort. What should it use?
- A. Amazon Polly
- B. Amazon Comprehend
- C. Amazon Textract
- D. Amazon Translate
Show answer
Answer: D
Amazon Translate is the managed machine translation service for text. Polly outputs speech, not translated text. Comprehend detects language, sentiment and entities but does not translate. Textract extracts text from scanned documents, and the descriptions are already text.
Question 5. A photo-sharing site must automatically flag uploaded images that contain explicit or violent content before they are published. Which service provides this without training a model?
- A. Amazon Rekognition
- B. Amazon Comprehend
- C. Amazon Textract
- D. Amazon Personalize
Show answer
Answer: A
Amazon Rekognition image moderation (the DetectModerationLabels operation) returns moderation labels for unsafe content. Comprehend moderates text, not images. Textract extracts document text and fields. Personalize recommends items from interaction data and does not inspect images.
Scenarios: choosing between close neighbours
Question 6. A streaming service has years of data on which users watched which titles. It wants each user to see a personalised list of recommended titles on the home page. Which service is the best match?
- A. Amazon Lex
- B. Amazon Personalize
- C. Amazon Comprehend
- D. Amazon Rekognition
Show answer
Answer: B
Amazon Personalize builds recommendations from user-item interaction data. Lex runs a conversation and does not rank items. Comprehend analyses text, and the data here is interaction events. Rekognition analyses images and video content, not viewing behaviour.
Question 7. A clinic wants a chatbot, usable by voice or text, that recognises when a patient wants to book an appointment, collects the date and doctor, and then calls a function that writes the booking. Which design fits?
- A. Amazon Comprehend with custom entity recognition
- B. Amazon Polly with an AWS Lambda function
- C. Amazon Lex with an AWS Lambda fulfilment function
- D. Amazon Transcribe with an AWS Lambda function
Show answer
Answer: C
Amazon Lex recognises the intent, fills the slots (date, doctor) and invokes an AWS Lambda function for fulfilment. Comprehend can analyse text but does not run a dialogue or collect slots. Polly only speaks text aloud. Transcribe converts audio to text but has no intent or slot logic.
Question 8. A company stores customer support emails as text in Amazon S3 and must detect names, addresses and phone numbers in them so the messages can be flagged for review. Which service detects this personally identifiable information?
- A. Amazon Textract
- B. Amazon Polly
- C. Amazon Translate
- D. Amazon Comprehend
Show answer
Answer: D
Amazon Comprehend can detect PII entities in text. Textract reads text out of scanned documents but the emails are already text. Polly generates audio. Translate changes the language of text and does not identify sensitive entities.
Question 9. A city app lets users photograph street signs and shop fronts. The app must read the text visible in each photo, which is not a formatted document. Which service is the most direct choice?
- A. Amazon Rekognition text detection
- B. Amazon Comprehend
- C. Amazon Translate
- D. Amazon Lex
Show answer
Answer: A
Amazon Rekognition can detect text in images such as signs and storefronts. Textract is built for documents with structure (forms, tables, invoices), which a street photo is not. Comprehend needs text as input and cannot read an image. Translate converts text that has already been extracted.
Question 10. A support centre has Spanish call recordings already stored in Amazon S3. It needs English text and the sentiment of each call. Which sequence of services produces this?
- A. Amazon Polly, then Amazon Translate, then Amazon Comprehend
- B. Amazon Transcribe, then Amazon Translate, then Amazon Comprehend
- C. Amazon Comprehend, then Amazon Transcribe, then Amazon Translate
- D. Amazon Textract, then Amazon Translate, then Amazon Polly
Show answer
Answer: B
Transcribe converts the Spanish audio to Spanish text (a batch job, because the files already exist), Translate converts that text to English, and Comprehend returns sentiment. Polly produces audio and is not needed. Starting with Comprehend or Translate fails because neither accepts audio as input. Textract reads documents, not recordings.
Edge cases and traps: constraints and custom needs
Question 11. A legal team has thousands of contracts as text and must sort each into its own categories such as NDA, lease or supply agreement. These categories are specific to the firm, so no built-in feature knows them. The team has labelled examples but little ML expertise. What is the best approach?
- A. Comprehend built-in sentiment analysis
- B. Amazon Polly with a lexicon
- C. Amazon Comprehend custom classification
- D. A custom model built from scratch on Amazon SageMaker AI
Show answer
Answer: C
Comprehend custom classification trains a text classifier on labelled examples through a managed workflow, which suits proprietary categories and limited ML expertise. The built-in sentiment feature returns only positive, negative, neutral or mixed. Polly is text to speech. Building and hosting a custom algorithm in SageMaker AI is possible but needs far more effort than the requirement allows.
Question 12. A bank onboarding system receives photos of US passports and driver licences and needs the holder name, document number and expiry date as fields. Which Textract capability is designed for this?
- A. Textract AnalyzeExpense
- B. Textract plain text detection only
- C. Rekognition text detection
- D. Textract AnalyzeID
Show answer
Answer: D
Textract AnalyzeID extracts fields from identity documents such as US passports and driver licences. AnalyzeExpense targets invoices and receipts. Plain text detection returns lines of text without labelling which value is the document number. Rekognition can compare faces but is not the service that extracts identity document fields.
Question 13. As of October 2026, a team with a brand-new AWS account wants to forecast product demand from historical sales. The AWS documentation says Amazon Forecast no longer accepts new customers. Which option is available to this team?
- A. Amazon SageMaker Canvas
- B. Amazon Forecast
- C. Amazon Personalize
- D. Amazon Textract
Show answer
Answer: A
AWS points new customers who need forecasting to Amazon SageMaker Canvas, which can build forecasting models without code. Amazon Forecast is closed to new customers. Personalize recommends items and does not forecast quantities. Textract extracts document text and has no forecasting function.
Question 14. A research group needs a model that uses a proprietary loss function and custom features on sensor readings. No managed AI service covers the task. Which choice is appropriate?
- A. Amazon Comprehend custom classification
- B. Amazon SageMaker AI with a custom training script
- C. Amazon Rekognition
- D. Amazon Personalize
Show answer
Answer: B
When no managed AI service fits and a custom algorithm is required, a custom model is built, trained and deployed with Amazon SageMaker AI. Comprehend, Rekognition and Personalize are pre-trained services for text, images and recommendations, so none lets the team supply a custom loss function for sensor data.
Question 15. A health provider records patient calls and needs a transcript in which names and account numbers are already masked when it is delivered, using the fewest services. Which approach does this?
- A. Amazon Rekognition on the audio files
- B. Amazon Polly with a pronunciation lexicon
- C. Amazon Transcribe with PII content redaction
- D. Amazon Translate to the same language
Show answer
Answer: C
Amazon Transcribe offers content redaction, so the transcript can be produced with PII masked in a single step. Rekognition analyses images and video. Polly produces audio. Translate does not identify sensitive data. Running Comprehend afterwards can find PII but would add a service, and the unredacted transcript would already exist.
Where to go next
Read the AWS AI services topic guide for the cheat sheet, then try the related RAG and LLM practice questions and the zero trust and IAM practice questions. Take the free 20-question tests for AIF-C01 and MLA-C02, and read multiple-choice exam tips for handling constraint words.