--- title: "Use examples to get AI output that actually fits" description: "Teach a format, style, or classification with examples, including how to avoid copying their accidental details." author: Graham date: 2026-09-06 url: https://powerofai.ca/learn/prompt-with-examples --- # Use examples to get AI output that actually fits Teach a format, style, or classification with examples, including how to avoid copying their accidental details. One good example can explain “what I mean” better than ten style adjectives. ## Show the shape of a good answer Words such as professional, concise, or engaging mean different things to different people. An example can make your preference visible. If you need a weekly update, supply a short fictional update with the headings, level of detail, and kind of next action you want. Then identify which aspects should carry over. The assistant should copy the structure, not the fictional facts. ## Include a boundary case Suppose you are sorting customer messages into delivery, returns, and other. An ordinary example is easy. A message saying “The parcel arrived, but the item is broken” reveals whether your categories are clear. Decide how you would label it before testing the assistant. If reasonable people would disagree, improve the categories or allow a needs-review label. Ambiguous instructions create ambiguous automation. ## Separate the demonstration from the new input Label sections clearly: example input, example output, and new input. Explain what the example demonstrates, such as tone or formatting. If the demonstration includes a specific person, date, or number, say that those details belong only to the example. Use invented or permitted material when real examples contain information you should not share. ## Test on something you did not use to teach A request that reproduces the sample perfectly may still fail on a new case. Try a shorter input, one with missing information, and one that needs a different category. Check whether the same rule holds. Keep the examples that explain real distinctions; remove redundant examples that make the brief longer without making it clearer. ## A worked example Original teaching example, not a recorded model run. Vague starting point: Write a professional project update. ### A more useful brief Use this shape: “Done: bookings confirmed. Blocked: awaiting room access. Next: Sam checks access by Tuesday.” Apply the three headings to the notes below. Use only names and dates in my notes. Write “owner needed” when no owner is given. Why this helps: The example supplies structure, while the instruction prevents invented owners and imported sample details. ## Your turn Write one example of a useful output for a recurring task, then try the request on three different inputs. You are looking for this: The result follows your structure and leaves genuinely missing information visible. ## A quick check What should an assistant copy from a fictional example? 1. Every name and number 2. The features you identified, such as structure and tone 3. Whatever makes the answer sound convincing Answer: The features you identified, such as structure and tone Specify which parts demonstrate your preference and which are merely sample data. ## Sources and how to read this guide Examples and exercises are original teaching material, not recorded model runs, measured time savings, or certifications. Official sources were consulted September 6, 2026. Product behavior can change. - [OpenAI · Prompt engineering](https://developers.openai.com/api/docs/guides/prompt-engineering): Official background on instructions, context, examples, and evaluating changes. ## Take the idea further - [What to give AI: build a useful context pack](/learn/context-that-actually-helps) - [Test an AI workflow before you depend on it](/learn/test-an-ai-workflow) [All learning paths](/learn) | [Practice challenges](/practice).