Iterative Prompting: Refining Step by Step
You can guide an AI conversation by refining, redirecting and asking follow-ups.
Why this lesson matters
Most people treat AI like a takeaway counter: you place one order and hope it's right. If it isn't, you start all over again.
But the real power of AI is in the conversation. You refine, steer, give feedback, and build towards the result step by step. This is called iterative prompting, and it's how professionals use AI.
Don't cram everything into one prompt
A common mistake: you stuff your whole assignment, all the context, all the constraints and all the examples into one enormous prompt. The result? The AI loses the thread and delivers a messy answer.
Better: build your prompt up in steps.
| Approach | Risk | Example |
|---|---|---|
| Everything in one prompt | AI misses details, inconsistent result | 500-word prompt → vague answer |
| Step by step | AI focuses per step, higher quality | 3-4 short prompts → refined result |
Example: writing a business case
Instead of one mega-prompt, do this:
Step 1: Give context
I work as an IT manager at an insurer with 2000 employees. We're considering supporting our customer service with AI chatbots.Step 2: Ask for a structure
Which parts should a business case for this project contain?Step 3: Work out each part
Now write the 'cost-benefit analysis' part. Use these assumptions: implementation costs €150K, expected saving €80K per year on FTE costs, payback period wanted within 2 years.Step 4: Refine
Keep reading
Leave your name and email address and you can read the rest
You get the whole lesson right away, and every other lesson stays open after that. No password, no confirmation email.
The tone is too optimistic. Add risks and mitigating measures too. The board is sceptical about AI, keep that in mind.Working iteratively is faster, even if it feels like extra steps. You get feedback on the direction as early as step 2 and can adjust before you've worked everything out.
Chain-of-thought: making it think step by step
Sometimes you want the AI to think first. This is called chain-of-thought prompting.
Without chain-of-thought:
What's the best price for our new product?→ AI gives a number straight away, without reasoning.
With chain-of-thought:
Help me set the price for our new SaaS product. Think step by step:
1. Which factors determine the price?
2. What are common pricing models in B2B SaaS?
3. Given our costs (€50K/year) and target group (SMEs, 50-200 employees), which price is realistic?
Show your reasoning.→ AI goes through each step, shows its reasoning, and arrives at a well-founded conclusion.
Use phrases like "Think step by step", "Show your reasoning", or "Go through the options first before you draw a conclusion". This triggers chain-of-thought in most AI models.
Giving feedback to AI
The AI's answer is almost never perfect on the first attempt. And that's fine. It only gets really good once you give feedback.
Types of feedback that work well:
| Feedback type | Example |
|---|---|
| Adjust the tone | "Make it more formal / more informal / more enthusiastic" |
| Shift the focus | "Put more emphasis on the risks" |
| Correct mistakes | "Revenue was €4M, not €5M" |
| Shorten or expand | "This is too long. Cut it in half." |
| Change the style | "Write it as bullet points instead of running text" |
Vague feedback doesn't work. "Make it better" tells the AI nothing. Say exactly what needs to change: which part, in which direction, and why.
When to keep going vs. start a new conversation
After a few messages back and forth you sometimes notice the AI "gets stuck": it repeats itself, keeps making the same mistakes, or loses the thread.
That's because every AI model has a context window: a limit on how much text it can process at once. The longer the conversation, the harder it is for the AI to keep track of everything.
Start a new conversation when:
- The AI repeats itself despite corrections
- You switch topics
- The conversation is longer than 15-20 messages
- The quality noticeably drops
Keep going when:
- You're refining and the AI still understands the context
- You're building a complex document in steps
- Earlier messages contain relevant context
Quiz
You've asked the AI to write a report, but the tone is too informal. What's the best follow-up prompt?
When is it smart to start a new conversation instead of continuing?
Want to keep your progress, get the practice files and sign up for the free evening? Create a free account. All you do is click a link in your email.
Create a free account