Gains:
- Can explain why vague instructions produce generic and inaccurate results
- Can transform vague adjectives into measurable, numerical and well-defined instructions
- Understands the value of clearly stating what is not wanted and constraints
In the previous unit, we saw the six components of the prompt. We now focus on the feature of these components that determines the most value: clarity. The number one reason you get disappointing answers from AI is vague instructions. The model cannot read your mind; The more vague you say to him, the more general and inaccurate you'll get back. In this unit, we will learn with concrete examples how to translate uncertainty into measurable, numerical and bounded instructions.
Why Does Uncertainty Give Bad Results?
The model fills in any blank space with its "most likely" prediction. When you say "write a short text" it might be 3 short sentences for you but 2 paragraphs for the model. When you say "use a professional tone", the formality you have in mind may not match the coldness produced by the model. Ambiguous adjectives (nice, good, effective, short, professional) convey little information to the model because they mean different things to everyone. Clarity is replacing these adjectives with their measurable equivalents.
The logic of turning a vague adjective into a measurable instruction
Behind every vague adjective there is actually a numerical or concrete expectation. Your job is to uncover it:
- “Short” → “80 words or less” or “3 sentences”.
- “Professional” → “Formal, jargon-free, no second-person singular tone.”
- "Detailed" → "At least 2 examples and 1 numerical data under each heading".
- “Simple” → “Language free of technical jargon that a high school graduate can understand.”
- “A few” → “Between 3 and 5”.
Five Dimensions of Clarity
Size
ambiguous
net
length
"Keep it short"
"Maximum 100 words"
quantity
"A few suggestions"
"Exactly 5 suggestions"
target audience
"For everyone"
"For SME owners who do not understand accounting"
tone
"In appropriate language"
"Formal but warm; don't use emojis"
Scope
"Tell me about the topic"
"Just tell me about pricing and return policy"
Reviewing an instruction along these five dimensions turns clarity into an almost mechanical habit.
Say What You Don't Want
Clarity includes not only what you want, but also what you don't want. The model tends to fill wherever no boundaries are drawn. Negative constraints such as "Promising a price", "Mentioning a rival brand", "Adding information that is not in the text", "Don't ask questions, answer directly" significantly increase the output. Most people only write positive instructions; whereas a one-sentence "don't" list is one of the most powerful tools that maintains the direction of the output.
Four Copiable Templates
1) Conversion framework from vague to clear:
Task: [concrete verb + subject]Length: [number of words/sentences]Quantity: [how many/items]Target audience: [who, knowledge level]Tone: [concrete adjectives]Scope: only [X and Y]; Don't enter [Z].Don't: [negative constraints]
2) Social media post:
Write a LinkedIn post.- Topic: [product/event]- Length: 3 short paragraphs, ~120 words in total.- Let the first sentence attract attention, the last sentence be a clear call.- Target audience: industry professionals.- Don't: use the hashtag more than 3 times; Don't use exaggerated adjectives.
3) Product description:
Introduce this product.- Exactly 4 articles.- Each article no more than 15 words.- Address the non-technical customer.- Each article should focus on a benefit, not a feature.- Don't: make up numerical claims; just use the features I gave. Features: [list]
4) Ask for feedback (clarity check):
Evaluate the following instruction for clarity. List each point that remains unclear and suggest a measurable alternative for each. Instruction: [your own prompt]
Weak Prompt / Strong Prompt
Weak:
Write a nice apology message to the customer.
Strong:
Write an apology message to a customer whose shipment is 4 days late. - Maximum 90 words. - Sincere but professional tone; accept the blame, don't make excuses.- Offer a concrete compensation: 10% discount code on the next order.- Don't: explain the reason for the delay at length; Don't use emojis.
The second prompt makes "beautiful" measurable: length, tone, concrete compensation, and two negative constraints. The result is based on instruction, not guesswork.
Three Mini Cases
Case 1 — Call center. One support team was working with a "write a polite reply" pattern, and replies sometimes got too long and the customer got bored. When the restriction "Maximum 60 words, give the solution in the first sentence" was added, the average response length decreased and the satisfaction score measured by the team increased in the next period.
Case 2 — Financial reporting. When an analyst said "summarize the trends" the model was making up interpretations. The instruction "Just rely on the numbers in the table I gave; do not add a trend that does not exist; show the relevant number in parentheses in each sentence" reduced the fabrication rate to almost zero.
Case 3 — Content editor. An editor was receiving scattered texts with vague instructions like “Write SEO friendly.” When we concretized this as "Specify this keyword once in the title and the first paragraph; use 3 subheadings; let each title be a question sentence", the texts were ready for publication in a single attempt.
Tip: After writing your prompt, ask yourself: "If I give this instruction to 5 different people, will there be 5 different outputs?" If your answer is yes, there is still an ambiguous point. Clarity means narrowing the interpretation down to a single correct meaning.
Caution: Excessive clarity is also a trap. If you list 15 conflicting constraints, the model cannot meet them all at the same time and will ignore some of them. Write down the 3-4 most critical constraints first; add iteratively if necessary.
Common mistakes
- Relying on vague adjectives. Saying "good", "nice", "professional" and not writing the measurable equivalent.
- Not specifying length. The model always chooses a different length than you expected.
- Bypassing the negative constraint. Not saying what is not wanted.
- Not specifying the target audience. The same content is written completely differently for the specialist and for the child.
- Stack contradictory constraints. Wanting "too short" and "too comprehensive" at the same time.
In summary
- Ambiguous instruction causes the model to fill in the gaps with its own prediction, resulting in generic/inaccurate output.
- Clarity is replacing vague adjectives with their measurable equivalents (word count, piece, concrete tone).
- Five dimensions of clarity: length, quantity, audience, tone, and scope.
- Write down what you don't want as well as what you want; negative constraints aggregate the output.
- "If I give it to 5 people, will there be 5 different outputs?" testing reveals hidden ambiguity.
Application task
Take a prompt you recently wrote and mark each vague adjective in it (short, good, detailed, appropriate, etc.). Turn each into a measurable response and add a "don't" line. Run the same task with the old and new prompt and compare the two outputs; Note the difference in clarity in one sentence.
checklist
- [ ] I can recognize ambiguous adjectives in my prompt and make them measurable.
- [ ] I apply the five dimensions of clarity (length, amount, mass, tone, scope).
- [ ] I write a "don't" constraint on each important task.
- [ ] I make sure to specify the target audience.
- [ ] I avoid conflicting or excessive constraints.