DEV Community

draftkit
draftkit

Posted on

Stop Writing AI Prompts Like Essays: 5 Structural Rules That Fix Everything

Most AI prompt advice is fluffy. "Be specific!" "Give context!" That's not actionable. Here are 5 structural rules you can apply immediately.


Rule 1: Constraints > Instructions

Telling the AI what NOT to do is more effective than telling it what to do.

Bad: "Write a professional email"
Good: "Write an email. No 'I hope this finds you well.' No 'reaching out.' No 'circle back.' Under 120 words. First line must reference something specific."

Negative constraints eliminate the generic patterns AI falls into.

Rule 2: Format specifies quality

If you want bullet points, say so. If you want a table, say so. If you want exactly 3 options, say "exactly 3."

Output exactly 3 options, each with:
- Name (5 words max)
- Pros (2 bullets)
- Cons (2 bullets)
- Cost estimate
- Recommended for: [specific scenario]
Enter fullscreen mode Exit fullscreen mode

Vague format = vague output. Specific format = usable output.

Rule 3: One prompt, one job

Don't ask the AI to "write a blog post and create social media copy and design a landing page." That produces mediocre everything.

Split it:

  1. Prompt 1: Write the blog post
  2. Prompt 2: Based on the blog post above, write 3 tweets
  3. Prompt 3: Based on the blog post, write landing page copy

Sequential prompts beat kitchen-sink prompts every time.

Rule 4: Edge cases reveal quality

Test your prompt with unusual inputs:

  • Empty input ("what if there's no data?")
  • Very long input ("does it still follow the format?")
  • Conflicting requirements ("does it handle contradictions gracefully?")

If your prompt breaks on edge cases, it's not done.

Rule 5: The rubric test

Before reading the output, write down what "good" looks like:

Good output:
- Leads with the problem, not the product
- Includes at least one specific number
- Under 280 characters
- No emoji spam
- Ends with a question or CTA
Enter fullscreen mode Exit fullscreen mode

Then check the output against your rubric. If it fails, fix the PROMPT, not the output.


These 5 rules are why my prompt collections produce consistently good output. Every prompt was built with these rules and tested against real use cases.

$9 starter pack: https://innovate01.gumroad.com/l/qevvy

Top comments (0)