Beginner

Being Clear and Direct

Vague prompts produce vague answers. Learn the specificity spectrum, golden rules for precision, and how to eliminate ambiguity from every prompt you write.

14 min read 10 examples Chapter 2 of 11

Why Vague Prompts Fail

When you send Claude an ambiguous prompt, you aren't just giving incomplete instructions — you're asking Claude to resolve multiple competing interpretations simultaneously. Every ambiguous word is a branch point. The word "good" can mean accurate, well-written, concise, creative, or safe depending on context. Claude must choose one interpretation, and it may not choose yours.

This is the ambiguity tax: every vague term in your prompt forces Claude to make a guess. Enough guesses, and the output drifts far from what you wanted. Clarity isn't about being rigid — it's about eliminating the interpretation gaps that lead to misaligned results.

💡
The Ambiguity Tax
Every vague word in your prompt is a branch point where Claude must guess your intent. Three ambiguous words means Claude could be on one of eight different paths — and only one of them is the one you wanted.

The Specificity Spectrum

Prompts exist on a spectrum from dangerously vague to over-constrained. Both extremes produce poor results — for different reasons.

🌫️
Too Vague
"Write something about our product." Claude has no idea what product, who the audience is, what format, what tone, or what length. Output: generic filler that satisfies no real need.
🎯
Just Right
"Write a 3-sentence product description for our B2B analytics dashboard, targeting CFOs, emphasizing time savings." Clear enough to get focused output, open enough to let Claude be good.
🔒
Over-Specified
"Use exactly 47 words. Start with 'Imagine'. Include the word 'seamless' in sentence 2." Micro-managing format produces stilted, unnatural output. You're fighting Claude's strengths.

Golden Rules for Clear Prompts

1
Use strong action verbs
Don't say "tell me about" — say "explain", "list", "compare", "summarize", "classify", "rewrite", "debug", "translate", "evaluate". Action verbs set Claude's processing mode immediately.
2
Be literal, not metaphorical
"Give me a deep dive" is a metaphor. "Write a 600-word technical analysis covering X, Y, and Z" is literal. Claude takes instructions literally — write them that way.
3
State constraints explicitly
If something must NOT appear in the output, say so. "Do not include pricing information." "Avoid technical jargon." "Do not recommend competitors." Implicit constraints are invisible constraints.
4
Name the audience
Who is reading this? "Write for a non-technical executive audience" produces completely different output than "write for a senior software engineer". Audience specification changes vocabulary, depth, and framing.
5
Specify the purpose
What will this output be used for? An internal Slack message, a formal report, a landing page, or a training document each demands a different register. Purpose shapes tone, length, and structure.

Common Vague Words and Their Replacements

These are the most common offenders — words that seem specific but leave enormous interpretive space. Replace them with precise alternatives:

Vague → Precise Word Substitutions 
VAGUE WORD        →  PRECISE REPLACEMENT(S)
──────────────────────────────────────────────────────
"good"            →  "accurate and factual" / "persuasive" / "grammatically correct"
"better"          →  "more concise (under 100 words)" / "more formal in register"
"improve"         →  "fix grammar errors" / "make the argument clearer" / "shorten by 30%"
"write about"     →  "summarize" / "argue for" / "explain the mechanism of" / "list"
"deep dive"       →  "write a 500-word analysis covering [specific points]"
"briefly"         →  "in 2-3 sentences" / "under 50 words"
"comprehensive"   →  "covering X, Y, and Z with examples for each"
"professional"    →  "formal business register, no contractions, third person"
"simple"          →  "use vocabulary appropriate for a 10th-grade reading level"
"modern"          →  "published or relevant after 2022" / "uses current best practices"
"soon"            →  "within the next paragraph" / "in the first sentence"
"relevant"        →  "directly related to [specific topic]"
"explain"         →  "define the term, describe the mechanism, give one example"

Length Control: How to Specify Depth

Without a length constraint, Claude calibrates response length based on its best guess at what's useful. That guess is often wrong for your specific context. There are four reliable ways to specify length:

📏
Word / Sentence Count
"In exactly 2 sentences." / "In under 100 words." / "Write 300–400 words." Highly precise. Use for constrained formats like social media, summaries, or form fields.
📋
Structural Count
"List exactly 5 bullet points." / "Write 3 paragraphs." / "Give 4 examples." Controls the number of ideas rather than the exact word count.
🎚️
Depth Level
"A high-level overview only — no implementation details." / "A thorough technical explanation including edge cases." Lets Claude choose length but constrains depth.
📖
Format Anchor
"The length of a typical tweet." / "A one-page executive summary." / "A README-style introduction." Anchors to a familiar format that implies appropriate length.

The Goldilocks Principle

Great prompts are neither too sparse nor too prescriptive. The sweet spot gives Claude a clear task, a defined audience, and key constraints — while leaving the actual craft of the response to Claude's judgment.

✅
The Goldilocks Test
A prompt is well-calibrated when you could explain it to a smart colleague in one sentence and they'd know exactly what to produce — but you haven't dictated every word choice. If your prompt reads like a fill-in-the-blank template, you've gone too far.

Before / After: Three Real Examples

See how adding specificity transforms the same request into a precise, actionable instruction:

Write a good email about our new feature.
Write a 150-word product announcement email for our new "Smart Search" feature. Audience: existing customers who use our SaaS project management tool. Goal: drive them to try the feature in their account within 7 days. Tone: friendly and excited, but professional. Include: one concrete benefit, one specific example use case, and a clear CTA button labeled "Try Smart Search Now". Do not include pricing or comparison to competitors.
Subject: Find anything in seconds — Smart Search is here ✨ Hi [Name], We just shipped something you're going to love. Smart Search lets you find any task, comment, or file across your entire workspace instantly — no more digging through project folders. Type a client's name and see every related task, file, and discussion in one view. Try it on your next project kickoff: search your client's name and pull up everything at once instead of switching between boards. [Try Smart Search Now →] It's live in your account today. Let us know what you think. The [Product] Team
Explain machine learning briefly.
Explain what machine learning is to a business executive (non-technical) who needs to understand it well enough to discuss it intelligently in a board meeting. Use exactly 3 sentences. No technical jargon — define any terms you use. Do not use analogies involving cooking or children's toys.
Machine learning is a method of building software systems that improve their own performance automatically by analyzing large amounts of data, rather than following rules that a programmer manually writes. A spam filter that learns which emails you flag is a machine learning system; a calculator that adds numbers is not. Businesses use machine learning to make faster, more accurate decisions — from predicting which customers will cancel their subscriptions to detecting fraudulent transactions before they process.
Improve this code.
Review the Python function below. Do the following, in order: 1. Fix any bugs that would cause incorrect output. 2. Improve readability with descriptive variable names. 3. Add a docstring following Google style conventions. Do NOT refactor the logic or change the function signature. Output the improved code only — no explanation needed. def calc(x, y, z): r = x * y if z > 0: r = r / z return r
def calculate_weighted_value(base_value: float, multiplier: float, divisor: float) -> float: """Calculate a weighted value with optional division. Args: base_value: The starting numeric value. multiplier: Factor to multiply the base value by. divisor: If positive, divides the product; otherwise returns the product unchanged. Returns: The calculated result as a float. """ product = base_value * multiplier if divisor > 0: return product / divisor return product

Explicit Negation: "Do X. Don't Do Y."

One of the most underused clarity techniques is explicit negation — directly stating what you do NOT want. Claude respects "don't" instructions reliably, but it can't respect implicit don'ts it doesn't know about.

Explicit Negation Patterns 
# Pattern 1: Do / Don't pairs
Write a product summary. Focus on benefits, not features.
Include concrete numbers. Do not use superlatives like "best" or "leading".

# Pattern 2: Avoidance list
Summarize this research paper.
Avoid: technical jargon, citations, methodology details.
Include: the core finding, practical implications, limitations.

# Pattern 3: Format negations
Write the cover letter.
Do not start with "I am writing to apply for..."
Do not include salary requirements.
Do not exceed one page.
⚠️
The "Don't Think of a Pink Elephant" Problem
Excessive negation can backfire. If you give Claude a list of 15 things not to do, it may focus on the negations rather than the task. Keep negations to the 2-3 most critical constraints. For complex constraints, rephrase as positive instructions when possible: instead of "don't be too formal", say "use casual, conversational language".

Practice Exercise

Here are three vague prompts. Rewrite each one applying the Golden Rules: action verb, literal instructions, explicit constraints, named audience, and stated purpose.

Rewrite These Prompts
1. "Write something about our company culture for the website."

2. "Make this paragraph better."

3. "Help me with this SQL query — it's slow."

Test your rewrites in the Playground → and compare the output quality.

✅
Chapter 2 Takeaway
Clarity is a skill, not a personality trait. Before sending any prompt, scan it for vague words (good, better, brief, improve) and replace them with precise alternatives. State what you want. State what you don't want. Name the audience and purpose. These changes take 30 seconds and consistently double output quality.