Owen posts a short opinion: cities should cap how fast rents can rise. Within an hour, the replies have split into two camps. One side says he wants to destroy housing and turn every landlord into a villain. The other side, cheering him on, says anyone who disagrees just wants to throw grandmothers onto the street for profit.
Nobody in that thread is arguing with a real person. Owen never said landlords were villains. The people who disagree with him mostly do not want anyone evicted. Both camps are punching cardboard cutouts, winning every round, and learning nothing.
The fix has a name: steelmanning. Instead of attacking the weakest version of an idea, you build the strongest one first, then decide whether it still falls. This guide covers what it is, the classic rules, a step-by-step method, three worked examples, when not to do it, and exercises.
What is steelmanning?
A steelman is the strongest, most charitable version of an argument you disagree with. Steelmanning is the act of building it: restating the other side's position in a form its smartest supporters would recognize and endorse, and often improving on how they said it themselves.
The name is a play on the straw man. A straw man is a flimsy dummy that falls over when you push it. A steel man is the version that pushes back.
Straw man, weak man, steelman
It helps to see the three side by side, because the middle one is the sneakiest.
| Move | What you do | Example (topic: a four-day work week) |
|---|---|---|
| Straw man | Invent or distort a position nobody holds, then refute it | "So you think people should get paid for doing less work?" |
| Weak man | Pick the weakest real argument or the worst real advocate, then refute that | "One influencer says Fridays off cures burnout. Obviously nonsense." |
| Steelman | Find the strongest real case and engage with it | "The best case is that many knowledge workers do their most valuable work in fewer focused hours, so a compressed week can hold output steady while reducing burnout and turnover. Let's test that." |
The straw man is a known fallacy (see our guide to logical fallacies). The weak man is harder to spot because someone really did say the silly thing. But treating the silliest spokesperson as the whole side is just a straw man with better sourcing.
The principle of charity
Steelmanning grows out of an older idea philosophers call the principle of charity: when you interpret someone's words, assume the most reasonable meaning that fits what they said, rather than the most foolish one. If a colleague says "this deadline is impossible," charity reads it as "this deadline is unrealistic with our current staffing," not as a literal claim about physics.
Charity is about interpretation. Steelmanning goes further: you actively construct the best case for their view, including arguments a thoughtful supporter would make even if they did not.
John Stuart Mill put the reason for all this in one line in On Liberty: "He who knows only his own side of the case, knows little of that." If you have only ever beaten the weak version of the opposing view, you do not actually know whether your own view survives the strong one.
Rapoport's rules and the ideological Turing test
Two tools give steelmanning a concrete shape. One tells you how to respond. The other tells you how to check whether you really understand.
Rapoport's rules
In his book Intuition Pumps and Other Tools for Thinking, the philosopher Daniel Dennett popularized a set of rules he credited to the game theorist Anatol Rapoport. Paraphrased, they say that before you criticize someone's position you should:
- Restate their position so clearly, vividly and fairly that they say, "Thanks, I wish I'd thought of putting it that way."
- List any points of agreement, especially ones that are not already widely shared.
- Mention anything you have learned from them.
- Only then say a word of rebuttal or criticism.
The rules force you to understand the view before attacking it, and they change how the other person hears you. Someone who has just heard their argument stated better than they managed will listen to your objection very differently from someone who has just been mocked. Notice the bar in rule one: not "accurately," but "so well they wish they had said it." That is the steelman standard.
The ideological Turing test
The economist Bryan Caplan proposed the ideological Turing test, named after Alan Turing's famous test for machine intelligence. The idea: could you explain an opposing view so convincingly that its actual supporters could not tell you apart from one of them?
Many people can list the other side's talking points. Far fewer can explain why a decent, intelligent person finds them persuasive. If your version sounds like a parody, a real supporter would spot you instantly.
How to build a steelman, step by step
This method works whether you have five minutes before a meeting or a week before a debate.
1. State the question precisely
Write the disagreement as one clear sentence. "Should our school ban phones?" is vague. "Should students be required to store phones in lockers for the whole school day, including breaks?" is something you can actually steelman, because you know what is being defended.
2. Find the best advocate, not the loudest one
The loudest voices are rarely the most careful. Look for people who hold the view and think hardest about it: researchers, experienced practitioners, people with direct experience. On rent control, read a thoughtful tenant organizer, not a meme account.
3. Identify the values underneath
Most real disagreements are not about facts alone. They are about which values matter most. People who support rent control often weigh stability and community heavily. People who oppose it often weigh long-term supply and affordability for newcomers heavily. Both are real goods.
Once you can name the value the other side is protecting, the argument stops looking stupid and starts looking like a different set of priorities, which you can actually discuss.
4. Gather their strongest evidence and reasoning
What are the two or three best reasons a smart supporter would give, and the best evidence behind each? If you have a rebuttal you love, find out how a good proponent would answer it. They almost always have an answer.
5. Write the version a smart proponent would sign
Write the steelman as a short paragraph, in their voice, without sarcasm or scare quotes. Include the value, the mechanism (why the idea would work) and the best evidence. If it does not make you slightly uncomfortable, it is probably not strong enough yet.
6. Check it with a real supporter
The step almost everyone skips. Show your steelman to someone who actually holds the view and ask, "Would you sign this?" If they say, "Yes, and honestly better than I usually put it," you have passed Rapoport's first rule.
7. Only then, respond
Now make your case. Usually your objections get sharper, some old ones disappear (they only worked against the straw man), and you find points of genuine agreement to build on.
Worked examples: three contested questions
Here is the method applied to three questions people argue about constantly. The middle column is the comment-section version. The right column is the version worth arguing with. The goal is not to pick a winner, only to show the strongest case on each side.
Rent control
Owen supports rent caps. Sana thinks they backfire.
| Side | Straw man version | Steelman version |
|---|---|---|
| For rent control | "Landlords are greedy and the government should just fix prices." | "Moving is expensive and disruptive, so tenants cannot easily walk away from a sudden large increase. That gives landlords leverage beyond a normal market. Limiting year-over-year increases protects existing residents from displacement, keeps schools, jobs and neighbors stable, and buys time while new supply gets built. The policy can be designed to allow reasonable increases and exempt new construction." |
| Against rent control | "They just want landlords to get rich." | "Caps help the people who already have a unit, but they can reduce the incentive to build new housing, maintain existing buildings, or rent units out at all. Tenants also stay put in homes that no longer fit them, which slows turnover. The people hurt most are newcomers and young renters who cannot find a place. If the real problem is too little housing, the fix is more housing, plus targeted help for tenants at risk." |
Sana's steelman of Owen's side is not about greed, it is about leverage and stability. Owen's steelman of Sana's side is about the people still looking for a home. Now they can discuss something real: exemptions for new construction, how tight a cap should be, targeted subsidies.
A four-day work week
Felix runs a small design studio and is considering a four-day week. His business partner is not convinced.
| Side | Straw man version | Steelman version |
|---|---|---|
| For a four-day week | "People just want a long weekend." | "For focused knowledge work, output is not proportional to hours. Much of a five-day week goes to meetings, context switching and recovery from fatigue. A compressed week forces the team to cut low-value work, reduces burnout and sick days, and is a hiring advantage that costs less than a raise. Run it as a time-limited trial with clear output measures, and roll it back if they drop." |
| Against a four-day week | "Management wants to squeeze every hour out of people." | "Some work does not compress. Clients expect responses on Fridays, and coverage gaps can lose accounts. Teams may simply cram five days into four, trading one kind of burnout for another. Early trials often attract the most motivated teams, so results may not generalize. And once a benefit is granted, it is very hard to take back, so a failed trial carries a real cost." |
Both steelmen point to a better decision: a trial with measurable outcomes, a plan for client coverage, and an agreement up front about what happens if it fails. If you are weighing a decision like this yourself, the AI pros and cons tool is a quick way to lay both columns out before you commit.
Banning phones in schools
| Side | Straw man version | Steelman version |
|---|---|---|
| For a ban | "Kids today are addicted and ruined by technology." | "Phones are designed to capture attention, and a classroom cannot compete with a device that vibrates every few minutes. A full-day ban removes the constant temptation, which is easier for students than exercising willpower all day. It also reduces in-school social pressure, filming of classmates, and the fear of missing out during breaks. Teachers stop spending time policing devices and get back to teaching." |
| Against a ban | "They don't care about kids' mental health." | "Parents want to reach their children in an emergency, and some students rely on phones for medical, family or safety reasons. A blanket ban treats a teaching problem as a device problem and does not teach the self-regulation students will need as adults. Enforcement can be uneven and fall hardest on certain students. Class-time rules with a clear storage policy may capture most of the benefit with fewer downsides." |
Once both steelmen are on the table, the question narrows: not "phones good or bad?" but "full-day ban or class-time rules, and how do we handle emergencies and exceptions?" That is a question a school can actually answer.
Try it on this one. Pick the side you disagree with and argue it before the AI does:
Steelmanning at work and in relationships
Most of your steelmanning will not happen in a debate hall. It will happen in meetings and in the kitchen, with people you will see again tomorrow.
At work
In a meeting, a steelman sounds like a question with a restatement built in:
"Let me make sure I have your case right. You're worried that if we launch in October, support gets overwhelmed while we're still fixing bugs, and the first customers are the ones we can least afford to lose. Is that fair?"
When the person says yes, they feel heard, the room hears the strongest version of the concern, and your counterargument lands on the real point. A few habits help:
- Steelman before you disagree, not after. Once you have attacked a weak version, restating the strong one sounds like backpedaling.
- Name the value. "You're prioritizing quality over speed, which is a reasonable call."
- Propose a test. When both steelmen are strong, a small pilot often beats a long argument.
In relationships
With a partner, a friend or family member, the stakes feel different, but the mechanism is the same. Sana and her partner keep arguing about spending the holidays with his family. Her usual version of his position: "You don't care what I want." His usual version of hers: "You just don't like my parents."
The steelman version sounds different: "You only see your parents twice a year, your dad is getting older, and holidays are when the whole family is together at once. Missing it feels like missing something you can't get back." And from him: "The trip takes two full days of travel, you never get a real break, and you'd like at least one holiday that feels like rest for us."
Neither is an accusation. Both are things the other person would sign. From there, alternating years or a shorter visit starts sounding like a solution. One caution: a steelman delivered as sarcasm ("Fine, let me steelman your completely reasonable position") is worse than none.
When not to steelman
Steelmanning is a powerful habit, not a universal duty.
Bad-faith actors
Steelmanning assumes the other person wants the truth or a decision. Some people move the goalposts every time you answer or argue purely to provoke. You can still steelman the position privately so you understand it, but you do not owe the person endless engagement. Stepping away is legitimate.
Harmful claims and settled facts
Building the strongest version of a harmful claim in public can amplify it. Steelmanning to understand why people are drawn to a view (for research, teaching, or to respond well) is different from polishing it for an audience. Likewise, you can steelman values and trade-offs endlessly, but you do not need the strongest case for a flat earth before discussing geography. Charity toward the person ("why might they believe this?") still helps. Charity toward a thoroughly debunked claim does not.
Steelmanning is not agreeing
The most common misunderstanding. Building the strongest version of a view does not mean you endorse it. Often, once you have seen the best version, you can reject it with more confidence and explain exactly why. It also does not mean inventing a better argument and pretending they made it. If you improve their case, say so: "Here's what I think is the strongest version of your view. Is this what you mean?"
Exercises to build the habit
Steelmanning is a skill, which means it improves with practice. These drills take five to twenty minutes each.
- The one-paragraph steelman. Pick a position you strongly disagree with. Write one paragraph arguing for it, in first person, with no sarcasm. Rewrite until it would pass the ideological Turing test.
- Name the value. For any heated issue in the news, write one sentence per side naming the value each is protecting. If you cannot name the other side's value, you do not understand them yet.
- Rapoport in four lines. Next time you disagree with someone, write four lines before replying: their view restated, one point of agreement, one thing you learned, then your objection. Send only if all four lines are real.
- Switch sides. Prepare an argument for a motion, then argue the opposite side with equal effort. It is the classic debate drill for good reason (our guide on how to debate covers it in more detail).
- The strawman audit. Reread something you wrote about a contested issue. Would the other side agree with every summary of their view?
A close cousin of these drills is arguing a side on purpose to stress-test it. Our guide to the devil's advocate covers that technique, and Socratic questioning shows how to probe a position with questions instead of counterclaims.
Practicing steelmanning with AI
The hardest part of practicing steelmanning is finding someone willing to argue the side you disagree with, well, on demand. That is where AI debaters are genuinely useful.
Ask several AI debaters to argue the side you reject
On Debatly, several AI models and characters argue different sides of your question across rounds, and a judge weighs the arguments and gives a verdict with reasoning. That setup lends itself to steelman practice:
- For vs Against. Enter the question and focus on the side you disagree with. Read its opening as a draft steelman. Then write your own version before reading the rebuttal, and compare.
- Model vs model. Have two different AI models argue opposite sides of the same question. Different models often reach for different reasons, which gives you a wider set of arguments to build your steelman from.
- Devil's advocate. Put your own position forward and let Petra, the devil's advocate attack it. Her strongest objections are, in effect, the steelman of the other side.
Then let Judge Alex weigh the round. If the judge found the side you reject more convincing than you expected, that is worth sitting with. New to the format? What an AI debate room is explains how a session runs, and the debate room is where you set up your own question.
Here is a good one to steelman, because most people have a strong gut reaction to it:
Felix's routine before any decision he feels strongly about: write his position in two sentences, run an AI debate and read only the side he disagrees with, write his own steelman of it without copying, note which arguments the judge credited that he had dismissed, and finally show his paragraph to a colleague who holds the opposing view.
An honest limit
AI can produce a fluent, confident steelman that is still shallow. It may miss lived experience, local context, or the one argument real supporters care about most. Treat it as a strong first draft, and check it against real proponents and real sources.
One more to try, on a question that splits nearly every workplace:
Start with one argument this week
You now have the full toolkit. Know the difference between a straw man, a weak man and a steelman. Follow Rapoport's rules: restate, agree, acknowledge what you learned, then respond. Use the ideological Turing test to check yourself. And build steelmen deliberately: find the best advocate, name the value, gather the strongest evidence, write the version they would sign, and check it with a real supporter.
Pick one view you disagree with this week and write its steelman. Then argue with it.
If you want a sparring partner that will argue the side you reject as hard as it can, run it as a debate on Debatly and see how the judge weighs it. Try for free, and start with the question you feel most certain about. Plans are on the pricing page.
“Steelmanning: How to Argue Against the Best Version of an Idea”
