A reliable way to answer scenario questions is to read the desired output before you think about services or buzzwords. Ask: is the system supposed to create new text, classify something into labels, or pull existing facts from a document? That step helps separate the required output from the implementation. Foundation models can perform extraction too; compare available document-extraction capabilities, validation needs and operational effort rather than assuming only one model family can return fields.
Start with the output shape
Study example: a company has two requests.
Request 1: “Draft a product description from a few bullet points about features and target audience.” The output is new marketing text.
Request 2: “Extract invoice fields such as vendor name, invoice number, and total due from uploaded PDFs.” The output is structured data already present in the document.
These outputs point to different approaches. Request 1 fits text generation with a foundation model because the goal is to produce original language. Request 2 fits document data extraction because the goal is to identify and return fields, not invent prose.
When you practice this distinction, focus on the noun that describes the answer: paragraph, summary, label, number, field, or ranking. If the answer is “field values from a document,” choose extraction. If the answer is “fresh wording,” choose generation. For more scenario-based drills, the public bank of AWS AI Practitioner practice questions is useful when you specifically work through relevant questions and compare why one output type fits better than another.
Worked multiple-choice question
A retail company wants to speed up two workflows. First, marketing wants a first draft of product descriptions based on short internal notes. Second, accounting wants key fields captured from supplier invoices and stored in a database. Which approach best matches the two workflows?
- A: Choose one foundation model for both workflows solely because both involve text, without evaluating extraction or validation requirements
- B: Use document field extraction for both tasks because both start from written material
- C: Use a foundation model for product descriptions and document field extraction for invoices
- D: Use classification for product descriptions and regression for invoices
Correct answer: C.
Why C is right: product descriptions require generating new text, so a foundation model is appropriate. Invoice handling requires pulling known fields from existing documents, so extraction is the better fit.
Why the others are wrong: A uses “both involve text” as its entire justification and skips evaluating the extraction requirements. A foundation-model extraction system could be suitable when supported by testing and validation; the question does not establish that case. B ignores that marketing needs new content, not field capture. D uses prediction task names that do not match either requested output.
Which constraint changes the answer?
Now change one constraint: accounting no longer wants fixed fields only. Instead, they want a polite email summarizing unusual invoice terms for a human reviewer. The answer changes because the desired output changed from structured extraction to generated narrative. Extraction may still help find source facts, but generation is now part of the requirement.
A different constraint would not change the core answer. If invoices arrive as image PDFs instead of text PDFs, the task is still extraction because the desired output remains field values. The input format changed, not the output shape.
This is why a short review sheet matters: it keeps you sorting tasks by output, not by fashionable terminology. If you want a broader refresher before doing more questions, the AWS AI Practitioner study guide is a practical companion for revisiting generation, extraction, classification, and responsible AI tradeoffs in plain exam-focused language.
Official exam reference: AWS AI Practitioner exam guide. These study examples are independently written.