This question appears in AI certification exams, training courses, and job interviews. It usually arrives as multiple choice, with one correct option hiding among three that sound plausible.
Here is the answer, followed by a test you can apply to any task in about five seconds.
The Short Answer
A generative AI task is any task where the system produces new content rather than analysing existing content.
In a typical multiple-choice version of this question, the options look something like this:
- Identifying the main topic of a news article
- Calculating the total value of a sales spreadsheet
- Writing a poem based on a given theme
- Sorting customer emails into complaint and enquiry folders
The generative task is writing the poem. It creates something original that did not exist before.
The other three analyse, compute, or categorise information that already exists. Those are classification and computational tasks, not generative ones.
Why That Answer Is Correct
Generative AI models produce data autonomously, learning patterns from training data and then creating novel outputs that resemble that data without duplicating it.
Traditional AI approaches work differently. They classify, predict, or extract. They tell you what something is, or what will happen next, based on inputs.
A frequently used comparison makes this concrete. Traditional AI can identify whether a piece of music was written by Mozart. Generative AI can write a new composition that sounds like Mozart.
Both are legitimate applications of machine learning. Only the second one generates.
A Simple Test You Can Apply to Any Task
Ask one question: does the output exist anywhere before the model runs?
If the answer is no, the task is generative. If the answer is yes, and the model is just finding, labelling, or ranking it, the task is not.
Two supporting checks help with edge cases:
- Could the output be different each time and still be correct? Generative tasks usually have many valid outputs. Classification tasks have one.
- Is the model choosing from a fixed set of options? If yes, it is almost certainly classifying rather than generating.
Run those three checks and most exam questions resolve immediately.
Common Generative AI Tasks
Generative AI covers a wide range of output types, not just text.
| Task Type | Description | Example Tools |
|---|---|---|
| Text generation | Produces written responses from text input | ChatGPT, Claude, Gemini |
| Image generation | Creates images from text descriptions | DALL·E, Stable Diffusion, Midjourney |
| Code generation | Writes source code from natural language | GitHub Copilot, Code Llama |
| Audio and speech | Converts text to spoken output or music | Text-to-speech and music models |
| Video generation | Produces moving images from prompts | Text-to-video models |
| Scientific content | Generates molecular structures and designs | Research and pharma models |
Text generation is the category most people encounter first, usually through conversational chatbots. It remains the largest use case by volume.
Note that generative AI is not limited to creative work. Writing code, drafting a policy summary, producing synthetic training data, and designing candidate molecules are all generative tasks, and none of them are art.
Tasks That Look Generative but Are Not
Exam questions often include distractors that involve AI without involving generation. These are the common ones.
Classification. Identifying the topic of an article, tagging an image, or sorting emails into folders. The model picks from existing categories.
Prediction. Forecasting next quarter’s sales or estimating whether a customer will churn. The output is a number or probability, not new content.
Extraction. Pulling names, dates, or invoice totals out of a document. The information already sits in the source.
Recommendation. Suggesting products or films. The system ranks existing items rather than creating new ones.
Detection. Flagging fraudulent transactions or spam. This is binary classification wearing a different name.
Calculation. Any arithmetic or aggregation task. That is computation, not AI generation at all.
Generative vs Predictive vs Classification
The three categories are easiest to hold apart side by side.
| Dimension | Generative AI | Predictive AI | Classification AI |
|---|---|---|---|
| Core job | Creates new content | Forecasts a future value | Assigns a label |
| Output form | Text, image, audio, code | Number or probability | Category from a fixed set |
| Valid answers | Many | One, with error margin | One |
| Example | Draft a product description | Forecast next month’s demand | Tag a review as positive |
Most real business systems combine all three. A support platform might classify an incoming ticket, predict its urgency, and generate a draft reply. Only that last step is generative.
Where the Line Genuinely Blurs
Honest answer: some tasks sit awkwardly between categories, and exam questions rarely acknowledge it.
Summarisation. The source text already exists, but the summary is newly written. Most practitioners classify this as generative, because the model composes sentences rather than extracting them. Extractive summarisation, which pulls existing sentences verbatim, is not generative.
Translation. The meaning exists in the source. The target-language text does not. Modern neural translation is generative in mechanism, though it was long treated as its own category.
Question answering. Extractive question answering finds a span of text in a document, which is not generative. Generative question answering synthesises an answer from multiple sources, which is.
Retrieval-augmented generation. The retrieval step is search. The generation step is generative. The combined system is usually described as generative because that is where the output comes from.
If an exam question involves one of these, look at how the option is worded. “Summarise this report in three sentences” is generative. “Find the sentence that states the conclusion” is not.
Why This Distinction Matters Beyond the Exam
Getting this right changes how organisations budget, govern, and evaluate AI work.
Different risks. Generative systems can hallucinate, producing confident and wrong output. Classifiers fail differently, with measurable precision and recall. Governance for the two should not be identical.
Different evaluation. You measure a classifier with accuracy against a known answer. You cannot do that with a poem, a product description, or a design. Generative output needs human review or task-specific benchmarks.
Different cost profiles. Generation is typically far more compute-intensive per request than classification, which affects unit economics at scale.
Different tool choices. Reaching for a large language model to solve a classification problem is a common and expensive mistake. Sometimes a simple classifier is the right answer, and it will be faster, cheaper, and more reliable.
Quick Reference for Exam Questions
If you are working through a test, scan the options for these verbs.
Usually generative: write, create, draft, compose, design, generate, produce, summarise, translate, illustrate.
Usually not generative: identify, classify, detect, sort, rank, predict, forecast, calculate, extract, find, label.
The verb is a strong signal because it describes the relationship between the model and the output. Creating and producing point one way. Finding and labelling point the other.
The One-Line Version
If the AI makes something that did not exist before, it is a generative AI task. If it labels, ranks, finds, or forecasts something that already exists, it is not.
That single sentence answers almost every version of this question you are likely to meet.
FAQs
Which task is a generative AI task?
Any task producing new content, such as writing a poem, drafting an email, generating an image, or writing code from a description.
Is image classification a generative AI task?
No. Classification assigns a label to an existing image. Generating a new image from a text prompt is the generative equivalent.
Is summarising a document a generative AI task?
Abstractive summarisation is generative because the model writes new sentences. Extractive summarisation, which copies existing sentences, is not.
What is the difference between generative and predictive AI?
Generative AI creates new content; predictive AI forecasts a future value or probability from historical data.
Is a chatbot answering questions a generative AI task?
Yes, when the chatbot composes its own response. A rule-based bot returning pre-written replies is not generative.
