You will build a small, reusable set of examples that makes a repeated task consistent.
Some tasks come round every week and look slightly different each time: replies to the same five customer questions, notes from the Monday team meeting, a short update to your manager every Friday. You know what a good one looks like. You have written dozens. Yet each time you ask an assistant for help, you start from a blank prompt and get a slightly different result.
In this exercise you will build a small set of examples for one of those tasks, put it into a prompt you can reuse, and test it until the output is consistent. You will use what lesson 2.1, One good example beats a paragraph of description, taught about choosing examples, and what lesson 2.2, Style samples and examples of what to avoid, taught about voice.
Choose something you do at least weekly where consistency matters. Good candidates are replies to common queries, meeting notes in a set format, short status updates, product descriptions and job ad drafts. A task you do once a quarter is a poor choice, because you will not reuse the prompt often enough to make the effort pay.
The task should have a clear input and a clear output. For Daniel at the furniture shop, the input is a customer message asking where an order is, and the output is his reply. Those make up about a third of his inbox, and his part-timers answer them in three different styles.
Find three real examples of output you were happy with, along with the input each one answered. Real is the important word. Invented examples tend to be cleaner and more uniform than real work, and the model will copy that tidiness.
Pick three that differ from each other in useful ways. Daniel chose a reply to a polite customer whose order was on schedule, one to an annoyed customer whose sofa was a week late, and one to a customer who had given the wrong unit number. Different moods and situations, the same voice and structure.
Then strip out everything confidential. Replace names with [customer name], order numbers with [order number], addresses with [address]. Check for anything else that identifies a person, such as a phone number in a signature. The examples need to show the pattern. They do not need anyone's real details.
A reusable example prompt has four parts, in this order. First, a short paragraph of context and the task, as in module 1. Second, the examples, each clearly labelled, with the input and output visibly separate. Third, any rules about what to avoid. Fourth, a clear marker that the examples have ended and the new input starts.
Daniel's prompt opened with two sentences about the shop and the job: "I run customer service for an online furniture shop in Singapore. Draft a reply to the customer message at the end, following the pattern of the three examples." Then came "Example 1. Customer message:" with the first message, and "Reply:" with his reply. Examples 2 and 3 followed in the same layout. After them he added his habits to avoid from lesson 2.2, and finally: "New customer message:" followed by a blank for the new input.
The labels do real work. Without them, a model can mistake an example for part of the new message, or treat the new message as another example and answer all four.
Run the prompt on three new inputs you have not used as examples, each in a fresh chat. Read the outputs as a set and ask two questions. Do they follow the pattern? And is anything copied that should not be?
Over-copying is the most common problem. Daniel's first test runs all mentioned "a one-week delay", even for an order that was on time, because his late-sofa example was the longest and most detailed of the three. One example was dominating the others. He replaced it with a shorter late-delivery reply, and on the next round the outputs matched each situation.
Other signs of over-copying are the same opening line every time, the same product name turning up in unrelated replies, and outputs that all come out the length of your longest example. When you see one, replace or trim the example responsible rather than adding an instruction to fight it. The example is usually stronger than the instruction.
Record what you changed after each round and why. Two rounds are usually enough. If a third round still shows problems, look again at whether your three examples genuinely share the pattern you want.
At the end you should have one saved prompt with context, three labelled examples, any rules about what to avoid, and a marked place for new input. Alongside it, keep a short record: the three test inputs, a line on how each output did, and what you changed. Daniel's record was six lines long. His part-timers now paste the same prompt, and the replies sound like the same shop wrote them.
Keep this prompt somewhere you can find it, because module 9 turns prompts like it into templates. For now, decide which weekly task you will use and go and find your three examples.
Save a prompt containing three labelled examples, test it on three new inputs, and record what you changed after testing.
Junxiong-WFG Organisation is an authorised representative of AIA Financial Advisers Private Limited (Reg. No. 201715016G).