Which tasks are generative AI tasks? Examples

A generative AI task is one where the AI makes something new, such as writing an email, drawing a picture, or summarising a report. Tasks where the AI sorts, predicts, or spots something, such as flagging spam, forecasting sales, or recognising a face, are not generative.

In this guide, you will see plenty of everyday examples of both kinds, learn a quick test you can apply to any task, and look at the cases that tend to trip people up. If you have not read it yet, our guide to the types of AI you’ll meet introduces the two families first.

Everyday examples of each kind

Here are some common tasks side by side. The left column makes something; the right column judges, sorts, or forecasts.

Generative AI tasks (makes something new)	Not generative (sorts, predicts, or spots)
Writing a reply to an email	Flagging an email as spam
Drawing a picture from a written description	Recognising your face to unlock your phone
Summarising a long report	Sorting customer reviews into positive and negative
Translating a letter into Spanish	Detecting which language a letter is written in
Writing a short story or poem	Forecasting next month’s sales
Composing a piece of music	Spotting a suspicious payment on your bank card
Writing computer code from a description	Estimating what a house will sell for
Creating a product description for a shop	Predicting which customers might cancel a subscription
Turning meeting notes into an action list	Reading a postcode on an envelope
Making a realistic voice read out your text	Turning your speech into a typed transcript

Notice that the right-hand column is still clever AI. It learned from huge numbers of examples, which is called machine learning, the same as the generative tools. The difference is in what comes out at the end.

A quick test for any task

When you are unsure, ask yourself one question: did the output exist before the AI ran?

• If the AI hands back something that was not there before, like a new paragraph, image, or tune, it is generative.
• If the AI picks from options that already existed, like “spam” or “not spam”, a name from a list, or a number on a scale, it is predictive or sorting.

A second way to check is to look at the shape of the answer. A label, a yes or no, a score, or a forecast usually means sorting or predicting. A sentence, picture, or piece of code built from scratch usually means generating.

The tricky cases

Some tasks sit close to the line, and this is where most of the confusion comes from. It is fine if these take a moment to settle.

Summarising is generative. It can feel like the AI is “only shortening” your text, but it writes fresh sentences in its own words. That new writing is what makes it generative.

Translating is generative. The AI writes a new version of your text in another language, choosing the words as it goes. Detecting which language a text is in, on the other hand, is a sorting task.

A recommendation list can be predictive. When a streaming service suggests films, it usually picks from titles that already exist, guessing which you will like. That is prediction. If a chat tool writes you a paragraph explaining why you might enjoy each film, the explanation is generative.

Chat tools can do both. Tools like ChatGPT are built on a large language model, which produces text one word at a time. You can ask one to sort reviews into positive and negative, and it will, but it does so by writing out its answer. The task is sorting; the tool is generative. If you are curious how that writing happens, our guide on how an LLM works walks through it.

Try it yourself

The fastest way to feel the difference is to give one tool both kinds of job and then ask it to explain. Copy this into any AI chat tool:

I'm learning the difference between generative AI tasks and
predictive or sorting tasks. Here are six tasks:

1. Writing a birthday message for my sister
2. Deciding whether a photo shows a cat or a dog
3. Summarising a news article in three sentences
4. Forecasting tomorrow's temperature
5. Translating a recipe into French
6. Flagging a bank payment as unusual

For each one, tell me whether it is generative or predictive,
and explain in one plain sentence how you can tell.

Then try adding a few tasks of your own, from your work or your home life, and see whether you agree with its answers. If you disagree, ask it to explain its reasoning. Talking it through is a great way to make the idea stick.

A common first instinct

A common first instinct is to decide based on how impressive the AI seems, and to assume anything clever must be generative. Face recognition and fraud detection are remarkably clever, yet neither makes anything new. You will get the right answer more often if you ignore how smart the tool feels and look at what it hands back.

Next steps

If you would like a refresher on what AI is in the first place, our guide to what AI is starts from the very beginning. Otherwise, carry on to the next guide in this topic, where we look at whether AI should live on the internet or on your own device.