The gap between a mediocre AI output and a useful one is almost always in the prompt, not the model. Most people never learn this because mediocre output still looks like something. Here's how to ask questions that get you what you actually need.
The Problem with Vague Prompts
When you send a vague prompt, the model fills in everything you didn't specify: the audience, the tone, the depth, the structure, the perspective. It uses the most probable defaults — which are rarely what you wanted specifically. A vague prompt produces a generic output. Every time.
The Anatomy of a Good Prompt
A strong prompt has four elements. You don't always need all four, but knowing them lets you decide which to include:
1. Context
What situation is this for? Who is involved? What does the AI need to know to understand the task properly? Context is the most frequently skipped element — and the one that most improves output quality. Example: instead of "write an intro paragraph", write "write an intro paragraph for an article about AI use in small law firms, for a non-technical audience of solo practitioners who are sceptical about automation."
2. Task
What specific action do you want the AI to take? Be precise. "Help me with this" is not a task. "Rewrite this paragraph to be more direct and cut the word count by 30%" is a task. "List three objections a budget-conscious CFO would raise about this proposal" is a task.
3. Format
How should the output be structured? Bullet list, numbered steps, flowing prose, a table, a code block, a template with blanks? If you don't specify, you'll get the model's default format — usually prose with subheadings, which isn't always what you need. Specifying format saves an entire round of follow-up.
4. Constraints
What should the output avoid or limit? "Don't mention price", "keep it under 200 words", "use plain language, no jargon", "don't suggest solutions I'd need outside help to implement." Constraints are powerful because they eliminate a whole category of bad output upfront.
Examples: Before and After
Before: "Write an email about the project."
After: "Write an email to a client whose project has been delayed by one week due to a supplier issue. Tone: professional but warm. Don't be defensive about the delay — acknowledge it and focus on the revised timeline. Max 150 words."
Before: "Summarise this article."
After: "Summarise this article for someone who has 30 seconds to read it. Focus on the main finding and one practical implication. Two sentences maximum."
The Mental Model: Capable Intern
Think of the AI as a capable intern on their first week: intelligent, willing, and capable of good work — but relying entirely on your instructions. They won't read your mind. They'll try to impress you with a thorough-looking output. They'll confidently make up details they don't know. The quality of their work is a direct function of the quality of your brief. Prompt the way you'd brief a capable junior who doesn't know your context.
When to Iterate vs When to Rewrite the Prompt
If the output is close but needs adjustment: iterate with a targeted follow-up. If the output completely missed the mark: don't iterate — rewrite the prompt. The second generation of a bad prompt is usually another bad output. Identify what was missing or ambiguous, fix it in the brief, and start fresh.