Contradictory Requirements

A prompt with contradictory requirements asks the model to satisfy two goals that cannot both be fully achieved at the same time. Unlike vagueness — where the problem is too little information — contradictions give the model conflicting information that forces it to make an arbitrary trade-off.

The model will not tell you the requirements contradict. It will produce something, but that something will silently sacrifice one of your requirements to satisfy the other. You may not notice until you look closely at the output.

"Write a brief overview that comprehensively covers all aspects of cloud computing."

Brief means short. Comprehensive means thorough. For a topic as broad as cloud computing, these two goals cannot coexist. Something must be sacrificed.

The model will typically prioritize breadth (many topics touched) over depth, producing a shallow list-like overview. But it will call this 'comprehensive' in the opening line, creating a false sense of completeness.

Fix: Decide which matters more, then reduce the scope. "Brief overview of the top 3 cloud computing models" achieves both.

"Write a formal report that's also fun and engaging."

Formal writing follows conventions: measured language, third person, structured sections, no colloquialisms. Fun and engaging writing typically uses humor, first or second person, surprising angles, and emotional hooks. These styles actively resist each other.

Fix: Choose a point on the formality spectrum. "Write in a professional but conversational tone — not stiff, but not casual either." This is a coherent instruction; the contradiction is resolved by picking a middle ground and naming it precisely.

"Explain quantum entanglement simply and in full detail."

Simplification requires abstraction — hiding complexity and omitting nuance. Full detail requires the opposite. For technical topics, there is no way to be both maximally simple and maximally detailed simultaneously.

Fix: Specify the target audience and their background, then let depth follow from that decision. "Explain quantum entanglement for a physics undergraduate — assume they know wave functions but not QFT." This is specific and achievable because the audience definition implies the right balance.

Before sending a prompt with multiple instructions, do a quick contradiction scan:

1. List all instructions and constraints
2. For each pair, ask: can these be satisfied simultaneously without trade-offs?
3. If a pair conflicts, decide which takes priority
4. State the priority explicitly in the prompt

This takes 30-60 seconds for a complex prompt and prevents the frustrating experience of getting output that you realize only in hindsight cannot possibly satisfy what you asked for.

Not all contradictions are absolute opposites. Some instructions have partial overlap where satisfying one makes the other harder:

• "No jargon" + "technically precise" — technical precision often requires technical vocabulary
• "Objective" + "persuasive" — persuasion implies selective presentation of facts
• "Creative" + "consistent with brand voice" — creative constraints reduce creative range

Partial overlaps produce partially correct outputs that feel slightly off. The model compromises on both rather than fully achieving either.

The cleanest resolution for a contradiction is an explicit priority instruction:

• "Be as concise as possible while covering these 5 points. If you cannot cover a point fully in the word limit, summarize it in one sentence rather than omitting it."
• "Aim for a professional tone, but if you must choose between stiff-formal and casual-engaging, lean toward engaging."
• "Prioritize accuracy over simplicity. When a concept requires technical vocabulary, use it and define it the first time."

Priority instructions remove the model's need to guess how to handle the trade-off.

For automated prompt generation pipelines, you can add a contradiction pre-check using a second model call before generating the actual content:

Tone contradictions are particularly common because many tone descriptors exist on spectrums that are implicitly opposed:

• Formal ←————————→ Casual
• Authoritative ←————→ Approachable
• Urgent ←——————————→ Calm
• Concise ←—————————→ Explanatory

Combining terms from opposite ends of the same spectrum creates a contradiction. Instead of using two opposing terms, locate where on the spectrum you want to be and use a term for that specific point: "conversational but professional" or "direct without being cold."

Scope contradictions arise when the content scope and the length budget are incompatible:

• "Cover all 10 features in 100 words" — 10 words per feature is not enough for meaningful coverage
• "Provide a comprehensive analysis with examples for each point in a 2-minute read"
• "Full product comparison across 8 dimensions for 5 products in one paragraph"

These are mathematical contradictions. The fix is always to either reduce scope or increase the length budget. When both feel non-negotiable, the task needs to be split into multiple outputs.

A prompt can contradict itself about the audience:

• "Write for both beginners and experts" — the explanation depth that works for one is wrong for the other
• "Write for a technical team and their non-technical management" — same problem
• "Accessible to all readers regardless of background" — impossibly broad audience specification

Fix: Pick the primary audience and write for them. If you need two versions, run the template twice with different audience variables rather than trying to produce one output that serves both audiences poorly.

You need to write a product announcement for a new enterprise software tool. You want it to be: (1) exciting and enthusiastic, (2) formal and appropriate for a CEO audience, (3) under 80 words, (4) cover all 7 new features. Which contradiction is most critical to resolve?

Contradictory requirements are instructions that cannot both be fully satisfied simultaneously. Unlike vagueness, contradictions give the model too much conflicting information. The model will satisfy the contradiction silently, sacrificing one goal for the other without telling you.

The classic pairs are: brief+comprehensive, formal+fun, simple+detailed. Detect contradictions by listing all instructions and checking pairs. Resolve them by picking a priority, finding a middle-ground term, or splitting the task into multiple outputs.

In the next lesson, you will tackle missing context — the third major category of prompt mistakes.