Why "Write Me a Blog Post" Fails
Session 5.1 · ~5 min read
The Vague Prompt Problem
"Write me a blog post about productivity." This prompt gives the AI maximum freedom. It does not specify an audience, a tone, a structure, a length, an angle, or what "productivity" even means in this context. The AI responds by producing the most statistically average blog post about productivity it can generate. It hedges. It lists. It says nothing specific.
This is not the AI being lazy. It is the AI being rational. Given no constraints, it produces the output most likely to be acceptable to the widest possible audience. That output is, by definition, generic.
Vague input produces vague output. The AI does not read your mind. It reads your prompt. If your prompt says "write a blog post about productivity," the AI has no information about what makes your perspective on productivity different from the 50 million existing blog posts on the topic. So it writes blog post 50,000,001, which reads exactly like the other 50 million.
What Happens When You Constrain
Compare two prompts for the same task.
| Dimension | Vague Prompt | Constrained Prompt |
|---|---|---|
| Audience | Unspecified | Mid-career engineers managing teams of 5-12 |
| Angle | Unspecified | Why most productivity advice fails for technical managers |
| Tone | Unspecified | Direct, mildly irreverent, no corporate jargon |
| Structure | Unspecified | Problem statement, 3 failed approaches, 1 working approach, practical steps |
| Length | Unspecified | 800-1000 words |
| Forbidden | Nothing | No "in today's fast-paced world," no bullet-point lists, no generic advice |
| Required | Nothing | At least one specific example from engineering management |
The constrained prompt reduces the space of possible outputs. The AI cannot produce a generic listicle because the structure forbids it. It cannot hedge because the tone demands directness. It cannot write for beginners because the audience is specified. Every constraint eliminates a category of bad output.
The Constraint Funnel
Think of prompt engineering as narrowing a funnel. At the top, the AI has infinite possible outputs. Each constraint narrows the funnel. The goal is to narrow it until only outputs that meet your standards can pass through.
(infinite)"] --> B["+ Audience constraint
Eliminates wrong-audience outputs"] B --> C["+ Topic angle
Eliminates generic treatments"] C --> D["+ Tone constraint
Eliminates wrong voice"] D --> E["+ Structure constraint
Eliminates bad architecture"] E --> F["+ Forbidden phrases
Eliminates AI artifacts"] F --> G["Remaining outputs
(narrow, high quality)"] style A fill:#222221,stroke:#c47a5a,color:#ede9e3 style D fill:#222221,stroke:#8a8478,color:#ede9e3 style G fill:#222221,stroke:#6b8f71,color:#ede9e3
You do not need perfect constraints on the first attempt. Start with three or four. Run the prompt. Evaluate the output. Identify what is wrong. Add a constraint that prevents that specific failure. Run again. This iterative process is the core of prompt engineering, which Session 5.8 covers in depth.
The Seven Dimensions of a Complete Prompt
Every production prompt should address seven dimensions. Not all are required for every task, but considering each one forces you to make deliberate choices instead of accepting defaults.
| # | Dimension | Question It Answers |
|---|---|---|
| 1 | Role | Who is the AI pretending to be? |
| 2 | Audience | Who is reading the output? |
| 3 | Task | What specific action should the AI perform? |
| 4 | Context | What background information does the AI need? |
| 5 | Format | What structure should the output follow? |
| 6 | Constraints | What is forbidden? |
| 7 | Quality criteria | How will you evaluate the output? |
Dimension 6 (constraints) is the most underused. People specify what they want. Fewer specify what they do not want. Negative constraints are often more powerful than positive ones. "Do not use the phrase 'it's important to note'" is more actionable than "write in a natural style."
Further Reading
- Prompt Engineering, OpenAI API documentation
- Few-Shot Prompting, Prompt Engineering Guide
- Prompt Engineering in 2025: The Latest Best Practices
Assignment
Generate a blog post with the prompt "Write a blog post about productivity." Then generate another with a detailed prompt specifying: audience, tone, specific angle, required examples, forbidden phrases, word count, structure, and voice characteristics. Compare the two outputs side by side. Document every difference. Count the AI artifact markers (from Module 1) in each. Which has more?