Amazon Rekognition vs Amazon Comprehend — Choosing the Right Service
The AWS AI Practitioner (AIF-C01) exam loves scenario questions that map a business problem to the right managed AI service. This page walks through a representative question, then gives you the decision rule that covers the Rekognition-vs-Comprehend split.
Short answer
The correct answer is B. Amazon Comprehend.
Amazon Comprehend is a managed natural language processing (NLP) service that extracts insights from text. Out of the box, it offers entity recognition, sentiment analysis, key phrase extraction, language detection, topic modeling, and — directly relevant here — PII detection and redaction. You send raw text to the DetectPiiEntities or ContainsPiiEntities API, and Comprehend returns the entity type (NAME, PHONE, CREDIT_DEBIT_NUMBER, SSN, ADDRESS, EMAIL, and more) along with character offsets you can use to redact. No training data is needed; the models are pre-trained by AWS. The scenario is clearly text-based (chat transcripts and email threads) and explicitly asks for a fully managed, no-training solution — both are hallmark Comprehend clues. Comprehend also supports batch and asynchronous analysis jobs, which fit the 'millions of messages' scale. The decisive clues in the question stem are 'text', 'PII', and 'no model training required'. Anytime you see those together on AIF-C01, Comprehend is almost always the answer. If the input were images or video instead of text, you would switch to Rekognition — but here the content is written language, which is Comprehend's lane.
The Question
A customer support team wants to automatically scan millions of chat transcripts and email threads to flag any messages that contain personally identifiable information (PII) such as names, phone numbers, and credit card numbers, so analysts can redact them before archiving. The team wants a fully managed AWS service with no model training required. Which service best fits this requirement?
Why B is correct
Amazon Comprehend is a managed natural language processing (NLP) service that extracts insights from text. Out of the box, it offers entity recognition, sentiment analysis, key phrase extraction, language detection, topic modeling, and — directly relevant here — PII detection and redaction. You send raw text to the DetectPiiEntities or ContainsPiiEntities API, and Comprehend returns the entity type (NAME, PHONE, CREDIT_DEBIT_NUMBER, SSN, ADDRESS, EMAIL, and more) along with character offsets you can use to redact. No training data is needed; the models are pre-trained by AWS. The scenario is clearly text-based (chat transcripts and email threads) and explicitly asks for a fully managed, no-training solution — both are hallmark Comprehend clues. Comprehend also supports batch and asynchronous analysis jobs, which fit the 'millions of messages' scale. The decisive clues in the question stem are 'text', 'PII', and 'no model training required'. Anytime you see those together on AIF-C01, Comprehend is almost always the answer. If the input were images or video instead of text, you would switch to Rekognition — but here the content is written language, which is Comprehend's lane.
Why the other options are wrong
Amazon Rekognition is a computer vision service. It analyzes images and video for faces, objects, text in images, unsafe content, and PPE — not free-form text documents. It cannot scan email or chat transcripts for PII because those are text inputs, not image inputs.
Amazon Transcribe converts speech (audio or video) to text. It can redact PII while transcribing audio calls, but the scenario already starts with written chat and email text, so there is no audio to transcribe. Using Transcribe here adds a step that does not apply.
Amazon Polly does the opposite of what is needed — it converts text to lifelike speech for applications like IVRs and read-aloud features. It does not analyze text or detect PII, so it does not fit a redaction workflow at all.
Rekognition vs Comprehend: vision vs language
Amazon Rekognition and Amazon Comprehend are both fully managed AI services that expose pre-trained models through simple APIs, so you do not have to train, tune, or host anything. The difference is the type of data they understand. Rekognition works on pixels. Comprehend works on words. Rekognition handles computer vision tasks on images and video. Common Rekognition capabilities include object and scene detection, face detection and comparison, face search against a custom collection, celebrity recognition, text in images (OCR), unsafe content moderation, and PPE detection for workplace safety. Typical use cases are moderating user-uploaded images on a social app, verifying identity by comparing a selfie to an ID photo, tagging a media library, flagging violent or explicit frames in streaming video, and checking whether workers in a construction site feed are wearing helmets and vests. Comprehend handles natural language processing on text. Common Comprehend capabilities include entity recognition, sentiment and targeted sentiment, key phrase extraction, language detection, syntax analysis, topic modeling over large document sets, custom classification, custom entity recognition, and PII detection and redaction. Typical use cases are routing support tickets by sentiment, extracting people, organizations, and locations from news articles, detecting and redacting PII before archiving transcripts, and clustering large document sets into topics. A useful exam heuristic: if the input is an image or a video, think Rekognition. If the input is text — a document, a transcript, an email, a review, a chat — think Comprehend. A related trap is Amazon Textract, which extracts structured text from scanned documents and forms. Textract turns a PDF or image of a form into text and key-value pairs. Comprehend then analyzes that extracted text. On the exam, watch for 'scanned document' or 'form fields' as Textract clues, and 'sentiment', 'entities', or 'PII in text' as Comprehend clues.
Ready to see how you'd score?
Take the free practice quiz and find out which AWS Certified AI Practitioner domains you need to focus on. No signup required.
Practice 5 similar questions
Same cert, same or adjacent domain. Use these after reviewing the explanation.