1.Why do answers change depending on the question?
In Lesson 4 we saw that a language model looks at the text so far and picks the most plausible thing to come next. That makes the prompt the beginning of that "text so far." If the input is vague, the range of plausible answers is wide, and the model picks the most ordinary, safest answer. If the input is specific, the range of plausible answers narrows and gets closer to the answer you want.
It is exactly like handing work to a person. If you just tell a new coworker, "Write a report," they will not know what to write. You have to tell them who will read it, what decision it is meant to support, how long it should be and what shape it should take before you get a usable result. An AI does not even have the context to read between the lines, so you need to spell it out all the more.
2.The four elements of a good prompt
Not every prompt needs all four, but they make a good checklist for finding what is missing when you don't like an answer.
| Element | What it asks | Example |
|---|---|---|
| Goal | What are you trying to do, and where will you use the result? | I want to make the case for a bigger budget in a meeting |
| Context | Who will read it, and what is the current situation and background? | The reader is the head of finance, who is very comfortable with numbers |
| Constraints | What must be followed and what must be avoided | Under 300 words, no exaggeration, base it only on the attached material |
| Format | The shape of the result | A one-line title + three key lines + one table |
- Wrong: Make me a workout plan.Right: I'm an office worker in my 40s and want to build stamina. My knees are weak, so I'd like to avoid running. Make me a 4-week workout plan I can do at home without equipment, 20 minutes a day, 4 days a week, as a table organized by day of the week.It includes the goal (building stamina), the context (age, knees), the constraints (no equipment, 20 minutes, 4 days a week) and the format (a table by day of the week).
- Wrong: Fix this text.Right: The text below is an apology notice to customers. Don't change the content; just make it more polite and shorter. Below the revised text, explain in three lines what you changed and why.If you don't say what to change and what to keep, the AI may change the content as well.
3.Tips for being specific
Being specific does not mean writing more. It means replacing words that can be interpreted in different ways with numbers and examples. Change "short" to "in three sentences," "simple" to "so a middle school student can understand," and "a good title" to "five titles of no more than 20 characters, each containing a number," and the results change.
If you have material, paste it in. The AI has no way of knowing your company's situation or the article you read. If you provide the material to base the answer on and say "based on this material," there is less room for the AI to make things up. It is even better to mark the beginning and end of the material so it doesn't get mixed up with your instructions.
Korean Language Basics, Lesson 10, "Clear Sentences for Prompts" (a Korean-language course), practices writing sentences with a clear subject and object and avoiding vague references. In the end, much of the skill of writing good prompts is simply the skill of writing clearly in your own language.
- Replace vague adjectives (short, simple, good) with numbers or criteria
- State who will read it and where it will be used (for whom, for what)
- Also write down what not to do (no jargon, no guessing)
- Decide the shape of the result you want (table, list, number of paragraphs, length)
- Paste in supporting material, and keep the material separate from the instructions
4.Missing information is the problem, not missing magic words
You often hear that there are "secret phrases" for writing good prompts — that adding certain words or changing your tone makes the answers dramatically better. Wording can affect the result a little, but when you don't like an answer, the cause is usually not the wording but missing information that the AI has no way of knowing.
So when an answer is not what you expected, go back to the four-element checklist before changing the wording. Find which of the goal, context, constraints and format is missing and fill it in, and the result will change even if you ask in the same tone. Below is an example that walks through this process step by step.
- Look at the result and note what you don't like: it is too packed, there is nothing for a child to enjoy, and the travel times are long.
- Find the missing context: a seven-year-old and a grandmother are coming along, and walking for long is hard.
- Find the missing constraints: no more than two places a day, within 30 minutes' drive of the lodging, and a one-hour break in the afternoon.
- Find the missing format: a table by date, with columns "Time, Place, Travel time, Notes."
- State the goal too: a trip where the whole family doesn't get worn out, rather than seeing as much as possible.
5.Don't try to finish in one go — refine through conversation
It is normal for the first answer not to be perfect. Seeing the first answer often makes it clearer what you actually wanted. Rather than asking again from scratch, pinpoint what to fix — "more detail on the second item," "make the tone more formal," "change the order of the table's columns" — and it will improve quickly.
You can also go the other way and have the AI ask questions first. If you say, "Before doing this, first ask me what else you need to know," the AI will look for the missing information and ask about it. This is especially useful when you yourself don't know what you left out.
- Goal: persuade the team members to use the new schedule-management method.
- Context: the audience is 8 team members who are used to the current method and may find switching a hassle.
- Constraints: a 10-minute presentation, honestly covering the effort it takes to switch, not just the benefits.
- Format: about 6 slides, each with a one-line title and three key lines; the last slide covers expected questions and answers.
6.What not to put in a prompt
With an online AI service, whatever you put in a prompt is sent to the service provider's servers. Depending on the service and your settings, it may be stored or used to improve quality. So the rule is not to include national ID numbers (such as Social Security numbers), bank account or card numbers, passwords, other people's personal information or your company's confidential material.
If you really must work with real material, change names to "Person A" and company names to "Company B," and keep only as many numbers as you need. When using AI at work, first check which services and rules your company allows. We go into more detail in Lesson 8.
📌 Key points
- A prompt is a work order — if it is vague, the AI picks the safest, most ordinary answer
- Use the four elements — goal, context, constraints, format — to check what is missing
- Replace vague adjectives with numbers, audiences and criteria, and paste in supporting material
- The first answer is a draft — pinpoint what to fix, or have the AI ask questions first
- Never put personal information, passwords or confidential material in a prompt
🤖 Try asking AI like this
Copy a prompt and replace the [ ] parts with your own situation. Don't take the answer on trust — check it against this lesson.
When you can't even work out what to ask
I'm planning to [the task you want to do]. Don't answer right away. First, ask me five questions about the information you'd need to produce a good result. Once I answer, do the task based on my answers.
When you want feedback on a prompt you wrote
Below is a prompt I'm about to send to an AI. Check it against four criteria — goal, context, constraints and format — point out what's missing, and suggest one improved version of the prompt. [my prompt]
When you want writing in a specific shape
Based on the material below, write a [type of writing] to send to [the reader]. Length: [number of words or paragraphs]. Tone: [tone]. Must include: [content]. Must not include: [content]. [material]
- General principles of prompt writing commonly found in guides to using generative AI
Reached every goal above? Mark the lesson complete.
Storage is unavailable in this browser, so this lasts only for this page.🤖 AI Basics
- 1What Is AI? — Rules and Learning
- 2Machine Learning Basics — Data, Model, Training, Evaluation
- 3Neural Networks, Intuitively — Small Calculations Add Up to Judgment
- 4How Do Large Language Models Produce Text?
- 5Prompt Basics — Saying Exactly What You Want
- 6Advanced Prompting — Roles, Examples, Steps, Format
- 7Hallucination and Verification — How to Doubt What Sounds Plausible
- 8Privacy, Copyright and Ethics — Using AI Responsibly
- 9Working with AI — Delegate, Check, Connect
- 10Learning Strategies for the AI Era — The Basics Shape Your Questions