Prompts and "Feedback"—Letting AI Evaluate Itself to Evolve Your Thinking (Generative AI Prompt Textbook 11)
You've had AI generate ideas. You've had it write text. The structure and logic are reasonably solid. But for some reason, it feels lacking. That might be because there is no reflection.
When humans create something, we improve the quality of our output through a loop of reviewing, critiquing, reflecting on, and improving it. In other words, thinking "evolves" through feedback.
Generative AI can be operated using the same structure. In fact, if you use a prompt that says "have it critique what it wrote itself," the AI's output will advance to the next phase.
✅ Basic Prompt:
Please evaluate the text just output from an objective, third-party perspective. Provide comments on its structure, persuasiveness, and clarity. Also, include suggestions for improvement.
—With just this, the AI begins to analyze its own work calmly, as if it were its own mentor.
What is interesting is that the AI does not perceive criticism as a "personal attack." For humans, feedback can sometimes be painful. But for AI, it is nothing more than a computational process. In other words,
it can perform "harsh reviews" and "bold rewrites" repeatedly without any emotion.
This calmness is the true value of "utilizing AI in the rewriting phase."
✅ Applied Prompt (Changing Perspectives):
You are a harsh editor. Please point out the weaknesses in this text without hesitation. Specifically indicate any parts where readers might lose interest, areas where the logic is weak, or points that lack persuasiveness.
✅ Further Application (Up to the Improvement Phase):
Based on the points above, please present an improved version of the text. To make the changes clear, please also include brief reasons for the modifications.
—By doing this, the AI begins to run a mini PDCA cycle of "self-negation → improvement → reconstruction." This is truly an "intellectual feedback loop."
Let's list a few more practical ways to use it.
✅ Presentation Materials:
Regarding this presentation outline, are there any parts that a first-time viewer might find "boring"? If so, please suggest how to make it more engaging.
✅ Project Ideas:
Please tell me the bottlenecks or points with low feasibility for this project. Please be realistic. If possible, also tell me the "reasons why it might fail to persuade internal stakeholders."
✅ Catch Copy:
Think of reasons why the target audience might just say "meh" when they see this copy. Then, tell me what should be changed to make it memorable, and provide three alternative versions.
—These are all examples of applying "refine after output" prompt design.
At this point, the AI is no longer just a "machine that answers." It is your virtual teammate, editor, challenger, and supporter.
And what we are doing is not merely generating text, but the very act of retraining our thinking.
A prompt is not just a tool for "getting something." It is also a "device for making what you have already obtained even better."
The greatness of generative AI lies not in writing a perfect text on the first try, but in "being able to run a cycle of self-improvement as many times as you like."
Next time, we will discuss "Prompts and 'Virtual Personas'—Making AI Play Your 'Ideal Team' to Hold a Meeting."
Yes, giving a single AI multiple personas simultaneously to create a dialogue as if a team were debating. That is "multiplayer" in the world of prompts.—The conference room inside you begins to stir.