Prompt Engineering FoundationsIterating and formatting outputs · Lesson 5 of 16

Iterating: the art of the follow-up

Article · 10 min · 7 min lecture

Video lecture

Iterating: the art of the follow-up

11 chapters · about 7 min · captions · full transcript

Watch now

Chapters

The first answer is a draft

Professionals rarely accept the first result. They treat it as a draft and steer. Iterating well is often faster than trying to write a perfect prompt up front.

Five follow-up moves

1. Adjust. Change one thing at a time so you can see the effect.

Good. Make it 30% shorter and replace the second paragraph with a
customer example.

2. Point to specifics. Quote the part you want changed.

The line "we are passionate about quality" feels generic. Replace it
with something specific about our hand-stitched seams.

3. Ask for options. Instead of one version, ask for several and pick.

Give me 3 alternative opening lines: one funny, one emotional, one direct.

4. Explain what is wrong and why. The AI learns your taste within the conversation.

This sounds too salesy for our audience of nurses, who are sceptical of
marketing. Make it more practical and humble.

5. Combine. Take the best parts of different versions.

Use the opening from version 2 and the ending from version 3.

When to start fresh

Long conversations can drift. The AI may hold on to earlier instructions you have abandoned, or mix details from different drafts. Start a new chat when:

  • you have changed direction significantly
  • the answers are getting worse or repetitive
  • the conversation has become very long

When you start fresh, write a new, better prompt that includes everything you learned: "Write X, for Y, in style Z, avoiding A and B." That improved prompt is now reusable.

Ask the AI to critique itself

A simple, effective habit:

Before I use this, review your draft against my requirements:
under 150 words, mentions the Friday deadline, friendly tone.
List anything that doesn't meet them, then fix it.

This catches missed requirements. It is not perfect, especially for facts, so you still need to check anything important yourself.

Ask the AI to improve your prompt

You can also ask for help with the prompt itself:

Here is a prompt I use to write weekly team updates. Suggest how to
make it clearer and more specific, then show me the improved version.

This is a great way to learn, because you see what the AI thinks is missing.

Worked example: iterating an event invitation

A community organiser in Karachi drafts a volunteer invitation.

Round 1: "Write an invitation for volunteers for a beach clean-up on Saturday." Result: generic and long.

Round 2: "Make it under 80 words, suitable for WhatsApp, and add the time (7-10am), meeting point (Sea View, near the camel rides) and what to bring (water, hat, gloves provided)." Result: much better, but a bit formal.

Round 3: "Make it more energetic and friendly, with one emoji, and add a line asking people to reply 'IN' to confirm." Result: ready to send.

Round 4 (saving for later): "Now write me a reusable prompt I can use for future clean-up invitations, with blanks for date, time and location."

In four short messages, the organiser has a finished message and a template for next time.

Save what works

Keep a simple document of prompts that worked well, with a note of what they are for. Over time you build a personal library, and your first attempts get better and better. Many AI tools also let you save custom instructions or reusable prompts; check your tool's settings.

Common iteration mistakes

  • Changing many things at once, so you cannot tell what helped.
  • Saying "no, that's wrong" without saying what is wrong.
  • Continuing a confused conversation instead of starting fresh with a better prompt.
  • Forgetting to check facts after several rewrites, since errors can creep in.

Hands-on: the three-message workout

Practise iteration deliberately with this routine on any real draft:

Message 1 (the brief):
Write a 120-word LinkedIn post announcing that I passed my Google Ads
certification. Audience: small business owners in the UK who might hire
me. Tone: confident, not boastful.

Message 2 (point to specifics):
The opening "I'm thrilled to announce" is overused. Give me 3 alternative
first lines that start with a benefit for the reader.

Message 3 (combine and finish):
Use option 2 as the opening, keep the rest, and end with a question that
invites comments. Then list what you changed.

Asking the AI to list its changes makes it easy to review edits instead of re-reading everything.

Editing in place

Many assistants now open longer drafts in a side panel or document view (for example a canvas or artifact-style editor) where you can highlight a sentence and ask for a change to just that part. Use it for long documents: targeted edits keep the parts you already like and avoid the whole text being rewritten each time.

Branching instead of arguing

If an answer goes in the wrong direction, you do not need to argue with it for five messages. Most assistants let you edit your earlier message and regenerate, which creates a clean branch of the conversation. Fix the prompt at the point things went wrong rather than piling corrections on top.

When iteration is not the answer

If you have sent four follow-ups and the answer is still not right, stop and diagnose. Usually one of three things is missing:

  • Information the AI does not have (add context or paste material).
  • A clear picture of good (add an example).
  • The right tool or mode (for a hard calculation or multi-step plan, try a thinking mode; for current facts, turn on web search).

Then start a fresh chat with an improved prompt.

How to measure your iteration skill

Keep a tally for a week: how many messages did it take to reach a usable result? Also note which follow-up moves worked best for you. Most people find that "point to specifics" and "explain why" do most of the work, and that their first prompts improve as they learn what they usually forget.

Key takeaways

  • Treat the first answer as a draft; iterate by adjusting, pointing to specifics, asking for options, explaining why, and combining.
  • Start a fresh chat with an improved prompt when a conversation drifts or gets long.
  • Ask the AI to check its draft against your requirements, and still verify facts yourself.
  • Save prompts that work to build a personal library.

Check your understanding

Quick questions to lock in the lesson. They don’t count towards your certificate.

  1. An answer is 'too salesy'. What is the most effective follow-up?
  2. When is starting a new chat a good idea?
  3. Why change one thing at a time when iterating?

Put it into practice

Take one draft the AI gave you and improve it in exactly three follow-up messages. Then ask the AI to write a reusable template prompt from the final version.

Enrol for free to save your progress

Reading is always free. Enrol to keep your place, take the final assessment and earn a verifiable certificate.