After studying 10 billion sentences, a computer was able to say something funny—a skill that’s normally thought to be confined to sentient beings with high levels of social and emotional intelligence. The computer had learned to make a logical argument and even anticipate the other side’s counterargument. Yet it hadn’t learned to agree with elements of the other side’s argument, apparently because that behavior was all too rarely deployed across 400 million articles by humans. They were usually too busy preaching their arguments, prosecuting their enemies, or politicking for audience support to grant a valid point from the other side.

When I asked Harish how to improve at finding common ground, he offered a surprisingly practical tip. Most people immediately start with a straw man, poking holes in the weakest version of the other side’s case. He does the reverse: he considers the strongest version of their case, which is known as the steel man. A politician might occasionally adopt that tactic to pander or persuade, but like a good scientist, Harish does it to learn. Instead of trying to dismantle the argument that preschool is good for kids, Harish accepted that the point was valid, which allowed him to relate to his opponent’s perspective—and to the audience’s. Then it was perfectly fair and balanced for him to express his concerns about whether a subsidy would give the most underprivileged kids access to preschool.

Drawing attention to common ground and avoiding defend-attack spirals weren’t the only ways in which Harish resembled expert negotiators. He was also careful not to come on too strong.

DON’T STEP ON THEIR TOES

Harish’s next advantage stemmed from one of his disadvantages. He would never have access to as many facts as the computer. When the audience was polled afterward about who taught them more, the overwhelming majority said they learned more from the computer than from Harish. But it was Harish who succeeded in swaying their opinions. Why?

The computer piled on study after study to support a long list of reasons in favor of preschool subsidies. Like a skilled negotiator, Harish focused on just two reasons against them. He knew that making too many points could come at the cost of developing, elaborating, and reinforcing his best ones. “If you have too many arguments, you’ll dilute the power of each and every one,” he told me. “They are going to be less well explained, and I don’t know if any of them will land enough—I don’t think the audience will believe them to be important enough. Most top debaters aren’t citing a lot of information.”

Is this always the best way to approach a debate? The answer is—like pretty much everything else in social science—it depends. The ideal number of reasons varies from one circumstance to another.

There are times when preaching and prosecuting can make us more persuasive. Research suggests that the effectiveness of these approaches hinges on three key factors: how much people care about the issue, how open they are to our particular argument, and how strong-willed they are in general. If they’re not invested in the issue or they’re receptive to our perspective, more reasons can help: people tend to see quantity as a sign of quality. The more the topic matters to them, the more the quality of reasons matters. It’s when audiences are skeptical of our view, have a stake in the issue, and tend to be stubborn that piling on justifications is most likely to backfire. If they’re resistant to rethinking, more reasons simply give them more ammunition to shoot our views down.

It’s not just about the number of reasons, though. It’s also how they fit together. A university once approached me to see if I could bring in donations from alumni who had never given a dime. My colleagues and I ran an experiment testing two different messages meant to convince thousands of resistant alumni to give. One message emphasized the opportunity to do good: donating would benefit students, faculty, and staff. The other emphasized the opportunity to feel good: donors would enjoy the warm glow of giving.

The two messages were equally effective: in both cases, 6.5 percent of the stingy alumni ended up donating. Then we combined them, because two reasons are better than one.

Except they weren’t. When we put the two reasons together, the giving rate dropped below 3 percent. Each reason alone was more than twice as effective as the two combined.

The audience was already skeptical. When we gave them different kinds of reasons to donate, we triggered their awareness that someone was trying to persuade them—and they shielded themselves against it. A single line of argument feels like a conversation; multiple lines of argument can become an onslaught. The audience tuned out the preacher and summoned their best defense attorney to refute the prosecutor.

As important as the quantity and quality of reasons might be, the source matters, too. And the most convincing source is often the one closest to your audience.

A student in one of my classes, Rachel Breuhaus, noticed that although top college basketball teams have rabid fans, there are usually empty seats in their arenas. To study strategies for motivating more fans to show up, we launched an experiment in the week before an upcoming game targeting hundreds of season ticket holders. When left to their own devices, 77 percent of these supposedly die-hard fans actually made it to the game. We decided that the most persuasive message would come from the team itself, so we sent fans an email with quotes from players and coaches about how part of the home-court advantage stems from the energy of a packed house of cheering fans. It had no effect whatsoever: attendance in that group was 76 percent.

What did move the needle was an email with a different approach. We simply asked fans one question: are you planning to attend? Attendance climbed to 85 percent. The question gave fans the freedom to make their own case for going.

Psychologists have long found that the person most likely to persuade you to change your mind is you. You get to pick the reasons you find most compelling, and you come away with a real sense of ownership over them.

That’s where Harish’s final edge came in. In every round he posed more questions to contemplate. The computer spoke in declarative sentences, asking just a single question in the opening statement—and directing it at Harish, rather than at the audience. In his opening, Harish asked six different questions for the audience to ponder. Within the first minute, he asserted that just because preschools are good doesn’t mean that they should be funded by the government, and then inquired, “Why is that the case?” He went on to ask whether preschools were underprovided, whether they did help the most disadvantaged—and then why they didn’t, why they were so costly, and who they actually helped instead.

Taken together, these techniques increase the odds that during a disagreement, other people will abandon an overconfidence cycle and engage in a rethinking cycle. When we point out that there are areas where we agree and acknowledge that they have some valid points, we model confident humility and encourage them to follow suit. When we support our argument with a small number of cohesive, compelling reasons, we encourage them to start doubting their own opinion. And when we ask genuine questions, we leave them intrigued to learn more. We don’t have to convince them that we’re right—we just need to open their minds to the possibility that they might be wrong. Their natural curiosity might do the rest.

That said, these steps aren’t always enough. No matter how nicely we ask, other people don’t always want to dance. Sometimes they’re so attached to their beliefs that the mere suggestion of getting in sync feels like an ambush. What do we do then?

DR. JEKYLL AND MR. HOSTILE

Some years ago, a Wall Street firm brought me in to consult on a project to attract and retain junior analysts and associates. After two months of research I submitted a report with twenty-six data-driven recommendations. In the middle of my presentation to the leadership team, one of the members interrupted and asked, “Why don’t we just pay them more?”

I told him money alone probably wouldn’t make a difference. Many studies across a range of industries have shown that once people are earning enough to meet their basic needs, paying them more doesn’t stop them from leaving bad jobs and bad bosses. The executive started arguing with me: “That’s not what I’ve found in my experience.” I fired back in prosecutor mode: “Yes, that’s why I brought you randomized, controlled experiments with longitudinal data: to learn rigorously from many people’s experiences, not idiosyncratically from yours.”

The executive pushed back, insisting that his company was different, so I rattled off some basic statistics from his own employees. In surveys and interviews, a grand total of zero had even mentioned compensation. They were already well paid (read: overpaid), and if that could have solved the problem, it already would have.* But the executive still refused to budge. Finally I became so exasperated that I did something out of character. I shot back, “I’ve never seen a group of smart people act so dumb.”

In the hierarchy of disagreement created by computer scientist Paul Graham, the highest form of argument is refuting the central point, and the lowest is name-calling. In a matter of seconds I’d devolved from logic bully to playground bully.