Generative AI earns its place where a business needs new content produced quickly and at scale, or needs to turn a large amount of information into something a person can use. This lesson walks through the common use cases the exam expects you to recognize, and how to tell a genuine generative AI fit from a task that only needs a prediction.
Creating media
Generative models can produce images, video and audio from a text description. A marketing team can generate product shots in different settings, a game studio can draft concept art, and a training department can create narrated videos without a recording studio. Diffusion models drive most image generation, and Amazon Nova includes models that generate images and video.
Working with text
Text is the most common use. Summarization condenses long documents, meeting transcripts or support tickets into the key points. Translation converts text between languages while keeping tone and meaning. Content creation drafts emails, product descriptions, reports and social posts for a person to review. Code generation writes, explains or tests code from a plain-language request, speeding up developers rather than replacing them.
Assistants and customer service agents
AI assistants answer questions in natural conversation, often grounded in a company's own documents so the answers are specific to the business. Customer service agents go further: as well as answering, they can take actions such as checking an order's status, starting a return or booking an appointment. That makes them agentic, which a later lesson covers.
Search and recommendations
Generative AI improves search by understanding what a user means rather than matching keywords, and by writing a direct answer instead of returning a list of links. It also improves recommendation engines: embeddings capture what products, articles or songs are about, so the system can suggest items similar in meaning to what a user liked, and explain why in plain language.
Fit or not
| Need | Generative AI fit? |
|---|---|
| Draft 500 product descriptions in a consistent brand voice | Yes: new content at scale |
| Summarize 40-page reports for executives | Yes: condensing information |
| Predict next month's sales figure | No: a forecasting model predicts a number |
| Flag fraudulent card transactions | No: a classification model is the usual fit |
The test is whether the output is new content, such as text, images, audio, code or a written answer. If the output is a numeric prediction, such as a forecast, traditional ML is usually the better tool. (Foundation models can sort text into categories, but a traditional model trained for one narrow prediction is often cheaper.)
On the exam
Expect a business description followed by "which use case is this?" or "is generative AI appropriate?". Words like draft, write, summarize, translate, generate, converse and create point to generative AI. Words like predict, score, classify, detect and forecast usually point to traditional ML, even when the scenario mentions AI in general.