Prompts and "Contrast"—Asking for Differences to Bring Out the Essence (Generative AI Prompt Textbook 10)
Words themselves do not hold meaning. Or, to be more precise, it might be more accurate to say “they only hold meaning when compared.”
When you say "fast," it is only in comparison to "slow," and even if you are told about an "innovative service," it won't click unless you know how it differs from "conventional methods." In other words, if the "differences" are not highlighted, information remains nothing more than a mere explanation..
What happens when you apply this understanding of structure to generative AI?
The conclusion is clear.Simply incorporating "contrast" into your prompt will suddenly make the AI's output sharper.
✅ A common, mediocre prompt:
Please tell me about the future of education.
✅ A prompt incorporating a contrast structure:
You are an education critic. Compare "traditional classroom-based education" and "individually optimized education utilizing generative AI" based on three points: specific scenarios, effects, and evaluation criteria. If possible, calmly include the disadvantages of each as well.
—When you do this, the AI's output transforms from "vaguely nice-sounding talk" into "text that considers differences in perspectives and institutional design."
Why is "contrast" effective?
It is because AI is extremely strong with "comparative structures." For a generative AI that has learned from millions of comparison articles, evaluation reports, and review content, a request to "compare A and B" becomes a task with a clear structure and goal.
In other words, by presenting a framework of "what is the same and what is different," you can provide the AI with an "axis of thought."
✅ Prompt example (business context):
Compare this product concept with competitors A and B, and explain it so that the points of differentiation are clearly conveyed. Please limit the perspective to user experience and brand message.
—By specifying the "axis" of comparison, the AI begins to focus on the difference in value rather than just a comparison of specifications.
Also, contrasts in time or emotional axes, such as "future vs. past," "ideal vs. reality," and "expectations vs. results," can also be used.
✅ Prompt example (storytelling):
You are a writer. Write a short episode comparing "how we worked before AI" and "how we work after generative AI became commonplace" from the perspective of one person. If possible, highlight the differences through descriptions of daily life.
—This kind of contrast structure also creates an "axis" within the story. The reader naturally begins to follow "what has changed."
You can also use it at a higher level of abstraction. For example, as a tool for thinking about values or philosophical conflicts.
✅ Prompt example (thought experiment):
Regarding the question, "Does efficiency through AI increase human happiness?", please fairly present the arguments for and against. Also, clearly define the "happiness" that each side claims.
—This is a technique where you have the AI speak from different "standpoints" to recreate the structure of a dialogue by itself. It can also be used for debate practice.
I would like to emphasize once again here that "contrast" is not merely for "finding differences," but is a technique for "highlighting the essence through those differences."
To say something is "different," you must also look at "what is the same.". In other words, comparison is analysis, and it is also the visualization of structure.
If you incorporate this structure into your prompt, the AI's output will become logical. And within that logic, "nuance" will begin to dwell.
In other words, a prompt is not only a "technique for asking," but also a "technique for separating." And by
"making it compare," the AI will draw the shape of things more clearly.
Next time, we will move on to "Prompts and 'Feedback'—How to have the AI evaluate itself and encourage improvement." Beyond just asking,
how do you utilize the "aftermath of the answer"? Let's move on to the next phase of dialogue, where you have the AI review its own output.