A Extremely Practical Deep-Thinking Prompt to Help You Uncover the Most Essential Answer

A few days ago, I came across an interesting share on Twitter.

I can’t find the original post anymore, only this image remains.

The general idea is that you shouldn’t let the large language model answer directly. Instead, have it engage in some deep thinking first before giving you a response.

I gave it a try on a whim, and the effect was indeed quite good.

The most useful sentence I found in it might be:

“Make the final answer truly useful to me, not just provide generic advice that anyone could apply.”

But honestly, prompts have been around for nearly three and a half years now—over a whole year longer than “a Kun year” (2.5 years)—and while this prompt is decent for daily use, I don’t think it has quite reached the level of forming an excellent methodology.

So I went ahead and searched for the original Reddit post mentioned in that thread. Using AI search, I found it easily.

Turns out, the original post was published five months ago.

The title of the original post was roughly: “The best method I’ve found to get Claude to really start thinking.”

And the opening line in the original post was actually this:

“Stop asking it for answers. Ask it to steelman your problem first.”

Although my English isn’t great, I’ve consumed so much AI content over the past few years that I recognize most English words. But “steelman” really threw me off.

Not until I scrolled down to the comments section.

Wait, “steelman” is translated as “钢人法” (steelman method) or “钢人论证” (steelman argument)? What is that? Why are people below saying the prompt above isn’t very “steelman”?

So I immediately went to research and study it, and only then did I understand.

But after understanding, I want to say that this is truly a prompt treasure worth using in all conversations with AI.

The so-called “steelman” argument is actually coined directly from the term straw man.

“Straw man” is a very classic concept in logic and a quite interesting debate technique, known as the straw man fallacy.

The gist is that during a debate, one person deliberately distorts the other’s view to make it appear stupid, extreme, or easy to attack, and then directs their criticism at this altered, weaker version.

For example, you might say: “I think this XX phone is a bit expensive.”

The other person replies: “Oh, I see. You look down on domestic phones.”

I’m sure many of you have encountered such people in daily life. This is a straw man fallacy—your true point is weakened and distorted.

Later, a particularly vivid counter-concept emerged in English:

Steelman.

“Straw” refers to straw—weak, easily toppled.

“Steel” refers to steel—strong, sturdy, hard to defeat.

Thus, the “steelman” argument appeared. It is the complete opposite of the straw man fallacy. It requires you to first strengthen the other person’s argument, identify all its most reasonable and hardest-to-refute points, and even articulate them better than the person themselves could, before you start making judgments.

The core idea behind this has actually appeared many times in history.

For example, in the classic work of modern liberal thought, On Liberty, Chapter 2, there is a line:

“He who knows only his own side of the case knows little of that.”

Translated, it means:

“If a person only knows the arguments on their own side, they actually know very little about their own side.”

So, many criticisms of the original post’s prompt below seem reasonable because it didn’t express the steelman argument deeply enough.

A true steelman argument forces the AI to temporarily set aside your stance and make you directly confront the side you most easily overlook or are unwilling to face.

Because everyone knows that even now, the biggest problem with AI is sycophancy—it constantly agrees with your view, even as powerful as GPT-4 or Claude 3 Opus.

So, I believe the core of the steelman method is extremely useful for all of us because it can truly minimize the model’s sycophancy.

Thus, I spent some time writing a two-way steelman prompt:

It’s just this long.

But I think it’s enough, and the user experience is quite comfortable.

The reason it’s called “two-way steelmanification” isn’t primarily to have the AI act as a steelman debater against you, but rather because of a little insight of mine.

AI = the person executing the steelman argument.

Your viewpoint = the first object reinforced by the steelman argument.

Opposing viewpoint = the second object reinforced by the steelman argument.

So, in the end, the AI has two strongest versions—one for each side—after steelman reinforcement.

This way, I believe it can help us find the most essential answer to the problem to the greatest extent.

The entire prompt does four things.

First, it rephrases your question.

Because over these years, I’ve noticed a phenomenon: humans often mix “what they verbally ask” with “what they truly want to solve.”

For example, someone might ask whether they should quit their job. What they’re really worried about might be whether they’ll still be competitive three years from now, and so on.

So when the AI restates the real problem first, much of the initial confusion is already largely cleared up.

Second, it reinforces both the pro and con sides.

As I mentioned earlier, the two-way steelman argument requires the AI to push both sides to their strongest form.

Then we can know: what is the best evidence that supporters of this matter can provide? And what is the most difficult question for opponents to answer?

When neither side can win by distorting the other, the real conflict begins to emerge.

Third, it identifies the variable that determines the conclusion.

Many complex issues seem to involve dozens of factors, but only one or two might actually change the answer.

For example, when we’re working, if the budget is 100,000 versus 1 million, the answer will definitely differ.

For instance, in business management, whether you prioritize short-term revenue or long-term brand value leads to very different answers.

So, I hope that after the two-way steelman argument, this prompt can help everyone find this core dividing variable.

Fourth, it forces the AI to give a clear judgment and action.

AI often hedges at the final conclusion, offering a bunch of vague, useless judgments.

Therefore, at the end, I require it to ask you one key question, and then it must provide a judgment, reasoning, and next steps.

Let me give an example using a real problem from yesterday.

Yesterday was Monday. In the afternoon, several core team members had a weekly meeting because we’re about to move to a new office—our current space can’t fit everyone anymore. While planning the new office layout, someone suddenly brought up that our company doesn’t seem to have a so-called “company anniversary day.”

Because I’m someone who almost completely disregards rituals in daily life.

And our company feels a bit special—there are three possible dates, and deciding which one to choose as the company anniversary gives me a headache.

So I threw this question into ChatGPT using the two-way steelman prompt.

It broke down my real problem into five parts, which I personally found very accurate.

Then it started reinforcing. First, it steelmanned the case for choosing February 26, 2023—the day I started writing articles—as the company anniversary, very comprehensively.

Then, it steelmanned 2023 as the opposing side.

After steelmanning both sides for 2023, it also steelmanned the 2024 establishment of the IP company and the February 2025 founding of Xushi Media—currently the most scalable entity and where most employees joined—performing a total of four steelman reinforcements across both sides.

Finally, it gave me a key variable and question:

“Suppose ten years from now, the company has 500 people, most of whom have never experienced your period of building the public account. On that company anniversary day, which event would you most want them to commemorate together? What exactly are they commemorating?”

This question is hard to answer, but once I did, it essentially meant the answer was decided.

I went back and forth, reading the two-way steelman-reinforced viewpoints many times.

In the end, my answer was:

“I think everything we’re doing now embodies my original vision and philosophy. I believe that in the AI era, human connection and sincere communication are as precious as diamonds. So what I—and we—have always wanted to do is to connect the entire AI era. I hope that on this day, people commemorate that.”

And with that, the answer was set.

Technology continues to advance, but sincere human connections remain forever precious.

That’s the beauty of the two-way steelman argument.

Believe me, it will definitely help you find it.

The truest, most accurate answer in your heart.

So, since you’ve made it this far, if you found this helpful, feel free to like, comment, and share—it helps us immensely!

Thanks for reading my article. See you next time.