Purpose

Use this lesson to supply a model when the desired style or classification is hard to describe. The aim is to produce work that a colleague can review and act on, not merely text that sounds convincing.

Lesson Explanation

Scenario: Classify customer feedback as billing, product, or support.

Better instruction: Classify each comment as Billing, Product, or Support. Example: “My invoice is wrong” → Billing. Return comment ID and category only.

One or two clean examples show a pattern more reliably than adjectives such as “concise” or “on brand.” Examples must match the real task and must not contradict the labels, format, or exceptions you request. Each result uses one of the three categories and contains only the ID and category.

Work in a short cycle: provide the relevant input, state the result and limits, inspect the draft against the acceptance criteria, then revise the instruction when a requirement is missing. Keep the final human decision with the person responsible for the work.

Practice Questions

1. A prompt asks an AI to categorize support tickets as “urgent” or “not urgent” but gives no example of either category. What does this lesson recommend adding?

   View Answer

   One or two clean examples showing the pattern, since examples demonstrate a pattern more reliably than adjectives alone.

2. Why does this lesson say examples work better than adjectives like “concise” or “on brand”?

   View Answer

   Adjectives are open to interpretation, while a concrete example shows exactly what the desired pattern looks like.

3. A prompt includes an example that contradicts the format instructions given elsewhere in the same prompt. What does this lesson warn about this situation?

   View Answer

   Examples must not contradict the labels, format, or exceptions requested – a contradictory example can confuse rather than clarify.

4. Write a one-example prompt for classifying emails as “Spam” or “Not Spam,” following this lesson’s guidance.

   View Answer

   Something like: “Classify each email as Spam or Not Spam. Example: ‘You won a free prize, click now!’ → Spam. Return only the email ID and category.”

5. A prompt gives three examples, but two of them are inconsistent with each other in how they apply the categorization rule. What problem does this create, per this lesson?

   View Answer

   Inconsistent examples undermine the pattern being demonstrated, since the examples themselves don’t agree on how the rule should be applied.

6. What does this lesson say happens when examples don’t match the real task?

   View Answer

   They can mislead the AI’s output, since the examples must match the real task and not contradict the labels, format, or exceptions requested.

7. A prompt for extracting key dates from contracts includes one clear worked example showing input text and the correctly extracted date. What principle does this reflect?

   View Answer

   Showing a pattern through a concrete example, which this lesson says demonstrates the desired behavior more reliably than a description alone.

8. How many examples does this lesson suggest are often sufficient to show a pattern?

   View Answer

   One or two clean examples.

9. A prompt asks for output “in our usual style” without providing any example of that style. What is the risk, according to this lesson?

   View Answer

   Without a concrete example, “usual style” is ambiguous and the AI is likely to guess at what that means, possibly incorrectly.

10. A prompt provides an example that is more complex and unusual than the typical case the AI will actually need to handle. What issue might this create?

    View Answer

    If the example doesn’t genuinely represent the real, common task, it may mislead the AI toward handling unusual cases rather than the typical ones.

11. A team wants an AI to reformat customer names as “Last, First” and provides the example “Smith, John” from “John Smith.” What is this an example of, per this lesson?

    View Answer

    Using an example to show the pattern, a more reliable method than describing the desired transformation abstractly.

12. Why might providing zero examples for a nuanced classification task lead to inconsistent AI output, according to this lesson’s reasoning?

    View Answer

    Without a concrete example to anchor the pattern, the AI has to infer the exact boundaries of each category from description alone, which is less precise than seeing the pattern demonstrated.