Disagreeing Well

What was measured

Yeomans, Minson, Collins, Chen and Gino built an interpretable machine-learning model of conversational receptiveness: the linguistic markers that signal you are genuinely willing to engage with an opposing view.

In contentious policy discussions, government executives rated as more receptive — by the algorithm and by their partners, but notably not by their own self-assessments — were judged better teammates, better advisers and better representatives.

People seen as receptive by the person they were disagreeing with were better liked, credited with better judgement, and more likely to change the other person's mind.

The self-assessment gap is the practical finding. Receptiveness is something you perform for a listener, not something you feel. Feeling open is not the same as sounding open, and only one of the two was rewarded.

The move

Aim at being heard as receptive, not at being receptive

The gap between your intention and the signal you actually send is where most arguments are lost. Assume your good faith is invisible until you say something that demonstrates it.

Executives rated as more receptive by their counterparts were seen as better teammates and advisers, and more persuasive. Their own self-ratings of receptiveness did not track the effect.

The field test

The same work looked at Wikipedia editors — a population that conducts long, public, written disagreements as a matter of routine.

Editors who wrote more receptive posts were less likely to receive personal attacks in reply. Receptiveness early in an exchange forestalled escalation later in it.

That is the part worth keeping: it is not only that receptive people are liked more, it is that receptive openings change what comes back at them.

The move

The opening move sets the ceiling

How the first message is written predicts how the last one will be. You have most of your influence over the exchange before you have made your own case at all.

Wikipedia editors who wrote more receptive posts were less prone to receiving personal attacks from editors who disagreed with them.

Limits — Observational within the Wikipedia corpus rather than randomised, so the direction of causation is inferred, not proven.

Move 01 — say it back

The receptive markers are unglamorous: acknowledge the other position explicitly, state points of agreement before disagreement, hedge your own claims, and phrase things positively rather than negatively.

Saying their case back is the cheapest way to hit several of those at once, and it is falsifiable in a way that nodding is not — they can tell you whether you got it right.

The move

You have not earned the right to disagree yet

Put their position in your own mouth, well enough that they would sign it, before you say a word of yours. If they correct you, you have just learned you were about to argue with the wrong thing.

Acknowledging the opposing view, stating agreement explicitly and hedging were among the linguistic markers that predicted being rated receptive — and receptiveness predicted liking, perceived judgement and persuasion.

Move 02 — ask how

Fernbach, Rogers, Fox and Sloman found that asking people to explain in detail how a policy would actually work punctured their illusion of understanding, and moderated their position.

The effect was specific to mechanism. Asking people to list reasons for their preference did not do it — only asking them to explain how the thing functions. In a third experiment, generating a mechanistic explanation also reduced donations to related advocacy groups.

Why is a request for a justification, which people are fluent at producing. How is a request for a model, which they usually discover they do not have.

The move

How, not why

Ask it as a genuine question about the machinery, and answer it about your own position too. The point is not to catch them out — if you cannot explain your own mechanism either, that is the finding.

Asking people to generate a mechanistic explanation of a policy reduced both their judged understanding and the extremity of their position. Asking them to enumerate reasons did not.

Limits — See the next section — a 2022 replication found this holds only for consequentialist positions.

Where it stops working

Sloman and Vives revisited the finding in 2022 and it did not fully replicate. Explanation reliably reduced how much people thought they understood — that part held.

But the moderation of attitudes appeared only for issues judged on consequences. For positions grounded in protected values — the moral, non-negotiable kind — explaining changed nothing.

This is the most useful thing in the lesson. It tells you which disagreements the tool is for, and it tells you that when someone will not budge under a mechanism question, the likeliest reason is that you are not having a disagreement about consequences at all.

The move

Work out which kind of disagreement you are in

If a mechanism question moves them, it was about consequences. If it does nothing, it is about values, and repeating the question is no longer curiosity — it is pressure. That is the moment to stop asking and start listening.

Explanation reduced judged understanding as originally reported, but reduced attitude extremity only for consequentialist issues, not for those based on protected values.

Move 03 — stop trying to win

Broockman and Kalla sent 56 canvassers to the doors of 501 voters for conversations built around non-judgemental exchange and active perspective-taking — roughly ten minutes each.

In a randomised trial, those conversations substantially reduced transphobia, and the effect was still measurable three months later. The authors note the decrease was larger than the average American's decrease in homophobia across the fourteen years from 1998 to 2012.

It is worth knowing why this study is careful. It was run in the wake of a fabricated result on a similar question, by researchers who had helped expose that fabrication, and it was pre-registered with the data made public. It is the honest version of a finding that once was not.

The conversations contained no argument. Canvassers asked, listened, and shared. Nobody was trying to win.

The move

Trade winning for being someone they would talk to again

You will almost never change a mind inside one exchange. What you can do is end it as a person worth returning to, which is the only condition under which the change ever happens.

A single roughly ten-minute door-to-door conversation encouraging active perspective-taking substantially reduced transphobia, with effects persisting at least three months in a randomised field experiment.

Limits — One issue, one city, one canvassing organisation. Later work has found the effect real but harder to reproduce at scale.

If you use it to get something, it stops working. The canvassing conversations were effective because the listening was genuine and the canvasser had no move waiting behind it. Receptiveness deployed as a lever is a slower argument, and people can feel the difference — the Yeomans work found the person on the other side is a better judge of your receptiveness than you are.

Which means the failure mode here is not being too soft. It is performing openness while keeping score, and then wondering why nobody moved.

None of this obliges you to entertain everything. Deciding a position is not worth engaging is a legitimate answer, and much more honest than pretending to weigh it.