Iterative prompting

The first answer is a draft

Even experienced prompt writers rarely get exactly what they want on the first try. That's normal. Iterative prompting means treating the first response as a draft: you read it, figure out what's missing or wrong, and send a focused follow-up to fix it. A few quick rounds usually beat one heroic attempt at the "perfect" prompt.

Iteration works because most chat-based AI tools remember the conversation so far. You don't need to repeat everything — you can just say what to change.

The loop

1. Prompt — start with a reasonable first attempt.
2. Review — read the output critically. What's good? What's off?
3. Diagnose — name the specific problem: too long, wrong tone, missing a point, wrong audience, factual error.
4. Refine — send a targeted instruction that fixes that problem.
5. Repeat until it's good enough — then stop.

Vague feedback vs. targeted feedback

The single most important skill in iteration is giving specific feedback. "Make it better" forces the AI to guess what "better" means.

The improved follow-up says what to keep, what to change and why. That last part matters: "too long for a LinkedIn post" helps the AI make the right trade-offs.

A worked example

Suppose you're planning a weekend trip.

Plan a 2-day trip to Lisbon for two friends in their 20s.

The first response is a generic itinerary packed with famous sights. Rather than starting over, you refine:

Thanks. A few adjustments:
- We're on a tight budget — around €60 each per day excluding the hotel.
- We'd rather explore neighbourhoods and food spots than queue for big monuments.
- Put it in a table: Time | Activity | Rough cost.

Each round adds one layer: budget, preferences, format. That's often easier than trying to anticipate everything up front. (The prices above are illustrative — real costs vary, so check current information before you travel.)

Useful follow-up moves

• Narrow: "Focus only on point 3 and expand it."
• Adjust tone or audience: "Rewrite this for a 12-year-old."
• Change format: "Turn that into a checklist."
• Generate options: "Give me three alternative opening lines."
• Ask for critique: "What are the weakest parts of this draft?" — then ask it to fix them.
• Correct facts: "The meeting is on Thursday, not Tuesday. Update every reference."

When to start fresh

Long conversations can drift. Earlier instructions may get diluted, or the AI may cling to an approach you've moved away from. Start a new chat when:

• You've changed direction significantly.
• The AI keeps repeating a mistake you've corrected.
• The conversation has become very long and responses are getting less focused.

When you start fresh, write a new prompt that bakes in everything you learned — the budget, the tone, the format. That consolidated prompt is often your best reusable version.

Know when to stop

Iteration has diminishing returns. Decide what "good enough" means before you begin — a draft you'll edit yourself, or a finished piece? Often the fastest path is to get the AI 90% of the way and make the final tweaks by hand.