Module 1 · Lesson 1
Why Prompts Matter
The same model, wildly different results.
8 min read
You typed "write about climate change" and got three paragraphs of beige. You typed it into the same model your colleague used to draft a policy briefing that landed in front of city council. Same model. Different prompts. You blamed the tool.
The tool was not the problem.
Why this matters
Most people think of LLMs as "smarter search." They are not. They are brilliant, literal-minded collaborators who have never met you. Every prompt is a cold open. The skill you are about to learn will determine 80% of the value you ever extract from every AI tool you touch — for the rest of your career.
The reframe
A search engine pattern-matches keywords against an index. Its job is to find things. A language model is doing something categorically different: it is trying to satisfy whatever task the shape of your words implies. If the words imply nothing specific, the output will be specifically nothing.
Consider what "write about climate change" actually implies.
| Variable | What the prompt specifies | What the model fills in |
|---|---|---|
| Audience | — | Generic web reader |
| Format | — | Five-paragraph explainer |
| Length | — | ~600 words |
| Angle | — | Neutral overview |
| Tone | — | Textbook, hedged |
| Purpose | — | "Inform, roughly" |
Every blank gets filled with a statistical average. The statistical average of "climate change writing" is a middle-school explainer with Wikipedia energy. That is not a failure of intelligence. That is the model doing exactly what you asked.
Before vs. After
| The vague prompt | The specific prompt |
|---|---|
| "Write about climate change." | "Write a 200-word briefing for a city mayor on how climate change will affect municipal infrastructure costs over the next decade. Tone: direct, no jargon." |
| Output: generic five-paragraph essay. | Output: briefing you could almost send. |
| Edit time: 20+ minutes. | Edit time: 2 minutes. |
The three variables doing the work here are context, task, and constraints.
- Context answers: who is asking, and why?
- Task answers: what format, length, and purpose?
- Constraints answer: what should it avoid, simplify, or refuse?
Strip any one and ambiguity rushes back in. The model is not dumb. It has nothing to work with except what you hand it.
The economics of two extra minutes
Two extra minutes spent on specification routinely saves twenty minutes of editing an output that went the wrong direction. That is not a heuristic — it is a ratio the people who work seriously with these tools recognize from experience.
The mental model
Imagine you have hired a brilliant, knowledgeable contractor who has never met you. They can write, code, analyze, synthesize, argue. They will do what you said, not what you meant. If you walk up and say "build the thing," they will build a thing. If you want the thing, you have to describe it.
Rule of thumb
If a capable stranger could read your prompt cold and know exactly what you want, why, and what success looks like — you wrote a prompt. If they would ask you three clarifying questions, you wrote a wish.
Try it yourself
30-second exercise
Open any LLM and send it this prompt:
"Help me with my email."
Read what you get back. Then send it this prompt:
"I need to decline a vendor renewal without burning the relationship. The vendor is pushy and has sent three follow-ups. Keep it to four sentences, professional but firm, no apologies."
Compare the two outputs. The same model. A different brief. One is nearly sendable. One is a shrug.
What you should notice
The first prompt produces a question back to you, a generic offer to help, or a template so broad it is useless. The second produces something you could almost send immediately.
The difference is not the model's capability. It is:
- A specified goal (decline renewal)
- A constraint on length (four sentences)
- A tonal instruction (firm, no apologies)
- Situational context (pushy vendor, three follow-ups)
Notice how the second prompt gave the model a problem to solve rather than a category to occupy.
This is not mystical
Prompt engineering is not a priesthood. It is the same discipline as writing a good creative brief or a clear design spec — state the audience, the deliverable, the constraints, the tone. Most people do not bother because it feels like extra work.
It is not extra work. It is the work. The rest of this course is going to show you how to do it fast, on repeat, for every task you hand to an AI from here forward.
Key takeaway
A prompt is a brief, not a search query. Specify context, task, and constraints — or the model will average across every possible interpretation and you will hate the result.
The deeper question
If a model produces a better output when given more context, is the model actually smarter — or are you just doing more of the cognitive work yourself, and handing the model a narrower problem to solve?
That is not a rhetorical question. It is the question you will carry through the rest of this course, and through every AI interaction for the rest of your career.
Check your understanding
According to the argument in this lesson, why does a vague prompt like 'write about climate change' produce a generic result?
The lesson frames the relationship between user and LLM as briefing 'a brilliant but literal-minded collaborator.' What is the core implication of this framing?
A prompt is a brief, not a search query. Specify audience, task, and constraints — or the model will average across every possible interpretation and you will hate the result.
Paste your last vague prompt into the Prompt Workshop. It will surface the missing context, task, and constraint variables and rewrite it into something specific.
Refine a prompt. Get the diff, rationale, and tips.