Prompt Engineering for Content & SalesRefinement, better thinking and context · Lesson 3 of 15

Refine, don't restart: the feedback loop

Article · 11 min · 8 min lecture

Video lecture

Refine, don't restart: the feedback loop

11 chapters · about 8 min · full transcript

Coming soon

Chapter 1 of 11

Refine, don't restart

  • Specific feedback
  • Five refinement moves
  • When to start over

The narrated lecture is in production

Every chapter is scripted and ready. Browse the chapters and read the full transcript now — the video will appear here when it’s published.

Chapters

The first answer is a starting point

Professionals rarely accept a first draft from a human writer, and they should not from AI either. The real skill is iterative refinement: giving specific, targeted feedback that moves the output closer to what you need.

Vague vs specific feedback

Vague (weak)Specific (strong)
"Make it better.""The hook is too slow. Start with the customer's problem in under 8 words."
"More engaging.""Add one concrete detail from a customer's life, such as the school run or a late-night craving."
"Too long.""Cut to 80 words. Keep the second and fourth sentences."
"Not our tone.""Less formal. Use contractions, short sentences and one light joke. Remove 'we are delighted'."

Specific feedback tells the model what to change, how and what to keep.

A five-move refinement toolkit

  1. Select and build: "Option 3 is closest. Write five variations of option 3 with different opening lines."
  2. Surgical edit: "Keep everything except the CTA. Rewrite the CTA to push WhatsApp orders, not website visits."
  3. Dial a quality: "Make it 30% more playful" or "Make it more premium: fewer exclamation marks, more sensory detail."
  4. Change the angle: "Rewrite from the perspective of a customer who was skeptical at first."
  5. Critique then revise: "Before rewriting, list three weaknesses of this draft from the perspective of a busy mum scrolling Instagram. Then fix them."

Ask the model to critique itself

Models are often better at spotting problems when explicitly asked to evaluate. Useful prompts:

  • "Score this caption from 1 to 10 on hook strength, clarity and brand fit. Explain each score, then write an improved version."
  • "What questions would a skeptical buyer still have after reading this sales page?"
  • "Which sentence here is weakest and why?"

Treat self-critique as a helpful second opinion, not a verdict. The model can be overly positive about its own work, or change good things unnecessarily. You still decide.

When to restart instead

Refinement works when the draft is roughly right. Start a fresh prompt when:

  • The draft is fundamentally off-target (wrong audience, wrong offer). Fix the original prompt instead.
  • The conversation has become long and muddled, with conflicting instructions.
  • The model keeps reverting to a style you have rejected several times.

When you restart, update the original prompt with what you learned, such as the missing context, the new constraint or a better example. That improved prompt becomes your reusable template.

Worked example: a sales follow-up

A Dubai interior design studio's first AI draft of a follow-up email after a consultation is polite but generic.

  1. "Too generic. Reference that the client wants a warm, family-friendly majlis and is worried about the timeline before Ramadan." The draft improves.
  2. "The opening apologizes for following up. Remove that. Start with the one idea from the consultation that excited them most: the built-in low seating." Better.
  3. "Add a single, low-pressure next step: a 15-minute call to review the mood board. Offer two time slots." Done.
  4. "Now list any claims or promises in this email that I should double-check with the team." The model flags the timeline wording; the designer confirms it with the project manager before sending.

Four focused follow-ups turned a generic email into a specific, trustworthy one in a few minutes.

Hands-on: a refinement drill with before/after

Use one real draft (a caption, email or ad) and run exactly three rounds, each using a different move from the toolkit:

ROUND 1 - Critique then revise
Before rewriting, list the 3 biggest weaknesses of this draft for [audience] reading it on [platform].
Then fix only those 3. Keep everything else.

ROUND 2 - Surgical edit
Keep everything except [the CTA / the opening line]. Rewrite it to [specific goal].

ROUND 3 - Facts and promises check
List every factual claim, number, deadline or promise in the final version.
Mark each as: in my original notes / added by you.

Before (round 0): "We're excited to announce our new catering menu! Perfect for any occasion. Contact us today!"

After (round 3): "Office lunch for 20, sorted by 11:30. Our new catering menu has three set boxes (veg, chicken, mixed) delivered across Jeddah business districts. Reply with your date and headcount and we'll send a quote the same day." The round 3 check flags "the same day" as added by the AI; you confirm with the kitchen that it's true before posting.

Refinement in modern assistants

  • Canvas-style editors (ChatGPT canvas, Claude's artifacts and document editing, Gemini Canvas) let you highlight a sentence and ask for a change to just that part: surgical edits without regenerating everything.
  • Branching/editing a message lets you revise your original prompt and regenerate, useful when you realize context was missing.
  • When a chat gets long and muddled, ask for a summary of decisions and start fresh (see the context lesson).

Pitfalls

  • Endless tinkering. Set a limit of three or four rounds. If it is still not right, rethink the prompt or write that part yourself.
  • Accepting subtle drift. Each revision can introduce new claims or change facts. Re-check facts in the final version.
  • Forgetting to save the winning prompt. Capture it in your library.

Key takeaways

  • Give specific feedback: what to change, how, and what to keep.
  • Use the five moves: select and build, surgical edit, dial a quality, change the angle, critique then revise.
  • Restart with an improved original prompt when the draft is fundamentally off-target or the chat is muddled.
  • Limit rounds, re-check facts after revisions (drift), and save the winning prompt.

Check your understanding

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

  1. Which feedback is most likely to improve a caption?
  2. When should you start a new prompt instead of refining?
  3. Why re-check facts after several rounds of revision?

Put it into practice

Take one AI draft and improve it using at least three different moves from the refinement toolkit. Save the final improved prompt to your library.

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.