I used to think good prompt engineering meant writing the perfect prompt on the first try.
Craft something brilliant, hit enter, get exactly what you want. Like a magic spell.
That’s not how it works. And once I accepted that, I got much better at using AI.
The First Prompt Is Just the Beginning
Your first prompt is a starting point, not a finish line.
Even experts don’t nail it on the first try. They treat prompt engineering as a conversation—a back-and-forth where each response informs the next request.
The AI gives you something. You see what’s missing or wrong. You adjust. Repeat until you get what you need.
This isn’t a failure of technique. It’s the technique.
Why First Prompts Miss the Mark
Several things work against first-try success:
You don’t know exactly what you want until you see options. You think you want “a professional email,” but once you read AI’s version, you realize you actually wanted something shorter, warmer, or more direct.
AI interprets your words differently than you meant them. “Casual” to you might mean “relaxed but professional.” To AI, it might mean “using slang and emojis.” You don’t discover this mismatch until you see the output.
Complex requests have many variables. Tone, length, format, level of detail, what to include, what to omit—you can’t specify everything upfront. Some things only become clear through iteration.
The gap between concept and execution is real. You have a vision in your head. Translating that into words AI can act on takes refinement.
None of this is your fault. It’s just how the process works.
The Iteration Mindset
Here’s the mental shift that helps:
Don’t try to write the perfect prompt. Try to start a productive conversation.
Your first prompt should be good enough to get a useful response—something you can react to and build from. Then you iterate.
This is faster than agonizing over the perfect initial prompt. And it produces better results, because you’re making adjustments based on real output rather than imagining what might work.
Practical Iteration Phrases
These are my most-used follow-up phrases. Copy them directly:
Adjusting Length
- “Make it shorter. Cut by half.”
- “Too brief. Expand with more detail.”
- “This should fit in a single paragraph.”
- “Break this into shorter sections.”
Adjusting Tone
- “Too formal. Make it conversational.”
- “Too casual. More professional, please.”
- “Warmer. Like you’re talking to a friend.”
- “More confident. Less hedging.”
Adjusting Content
- “Focus more on [specific aspect].”
- “Remove the part about [topic].”
- “Add examples.”
- “Include specific numbers/data.”
- “Less theory, more actionable advice.”
Adjusting Format
- “Present this as bullet points.”
- “Convert to a numbered list.”
- “Use headers to organize this.”
- “Put this in a table.”
- “Format as code.”
Redirecting
- “That’s not quite what I meant. I want…”
- “Good, but I need this to be more about X and less about Y.”
- “Start over with this angle instead: [new direction].”
- “Keep the structure but change the examples.”
Getting Closer
- “Almost. The second paragraph is perfect—make the first paragraph match that tone.”
- “The format is right, but make the content more specific.”
- “Keep everything except the conclusion. Rewrite just that part.”
The Three-Iteration Rule
Most tasks converge within three iterations:
Iteration 1: Get the basic output. See what AI produces with your initial prompt.
Iteration 2: Adjust the obvious issues. Fix tone, length, or format problems.
Iteration 3: Fine-tune the details. Small tweaks to get it just right.
If you’re beyond three iterations and still not close, something’s wrong with your approach—either the task is too complex for a single prompt, or you need to restart with a different angle.
Breaking Down vs. Iterating
Sometimes iteration isn’t the answer. Sometimes you need to break the task apart.
Iterate when:
- The output is in the right ballpark
- You’re adjusting one or two things
- Small changes will fix the problem
Break apart when:
- The output is completely wrong
- Multiple major things need to change
- The request is trying to do too much
For complex tasks, try: “Let’s do this step by step. First, just give me [first piece].”
Then iterate on that piece until it’s right. Then move to the next piece.
Selective Feedback
Be specific about what’s working and what isn’t.
Vague: “This isn’t right.”
Better: “The tone is perfect, but the examples don’t fit my industry. Use examples from B2B software instead of retail.”
Even better: “Keep paragraphs 1 and 3 exactly as they are. Rewrite paragraph 2 to focus on implementation challenges rather than benefits.”
The more specific your feedback, the less AI has to guess about what to change.
The “Yes, And” Technique
Instead of replacing the whole output, build on what’s good.
Prompt: “Write an intro paragraph for my blog post about remote work.”
AI writes something decent but generic.
Follow-up: “Good start. Now make it more personal—mention a specific frustrating moment from remote work that readers will relate to.”
AI adds the personal element.
Follow-up: “Better. Now punch up the last sentence to make it more compelling.”
Each iteration adds a layer. You’re sculpting the response rather than starting over.
When to Start Over
Sometimes iteration isn’t worth it. Start fresh when:
- You’re on iteration 5+ and still frustrated
- The fundamental approach is wrong (not just the details)
- You’ve learned something from the output that changes what you want
- The conversation context has gotten cluttered
Starting over isn’t failure. It’s using what you learned to make a better first attempt.
When you restart, take what you learned: “Actually, what I want is [clearer description based on what you now know].”
The Hidden Benefit of Iteration
Iteration teaches you what you actually want.
Often, you don’t fully know what you’re looking for until you see what you’re not looking for. The AI’s first attempt clarifies your own thinking.
“Oh, I don’t want a formal proposal. I want more of a casual pitch.”
“Actually, I need this to be less about features and more about the problem we solve.”
“I thought I wanted comprehensive, but really I just need the key points.”
This discovery process is valuable. Don’t fight it—use it.
Practical Workflow
Here’s how I approach most AI tasks now:
Write a reasonable first prompt. Don’t overthink it. Include the basics: task, context, any important constraints.
Read the output with questions: What’s right? What’s wrong? What’s missing?
Give specific feedback. Focus on the biggest issue first.
Repeat steps 2-3 until you’re satisfied or realize you need a different approach.
If stuck, restart with the clarity you’ve gained.
This usually takes 2-4 exchanges. Rarely more.
Your New Mantra
Stop trying to write the perfect prompt.
Start trying to have a productive conversation.
The first message is just the opening. What matters is what you do with the response—how you guide, adjust, and refine until you get what you need.
Your first prompt will never work.
And that’s exactly how it’s supposed to be.