They Think You Don't Want the Real Answer: What the Research Says About Empathy and Asking

Written By Rob Volpe

The most repeated piece of empathy advice in circulation is to walk a mile in someone else's shoes. Imagine how they feel. When researchers tested whether that actually makes people more accurate about what's in another person's head, it failed.

Tal Eyal, Mary Steffel, and Nicholas Epley published 25 experiments in the Journal of Personality and Social Psychology in 2018, with roughly 2,600 participants reading emotions from faces, predicting a partner's preferences, and detecting fake smiles and lies. Being instructed to take the other person's perspective produced no consistent gain in accuracy, and sometimes reduced it while raising confidence. One approach worked: asking the person. The paper's title says it directly. Accurately understanding another mind requires getting perspective, not taking perspective.

Imagining is not asking. That distinction is the hinge. Imagining draws its material from the person doing the imagining, and it produces an answer that arrives fluent and confident because it was assembled from familiar parts, with no mechanism for checking it against the source. Artificial intelligence has made the same operation easy to see. As the sociologist Zeynep Tufekci has argued, large language models are not reasoning machines but plausibility engines, producing the most likely next thing. Imagining another person's mind works the same way, and yields the same kind of result: plausible, not verified.

So why do people ask so rarely? A 2022 study by Michael Kardas, Amit Kumar, and Nicholas Epley found that participants expected deeper conversations to be awkward and expected the other person not to care much. The conversations then went better than predicted, the deeper ones especially. The error was a consistent underestimation of how much other people care about what one would tell them. When researchers corrected that expectation, participants chose deeper topics, which means the belief was not passive. It was steering what people were willing to say and ask.

The proof, for me, is a Cincinnati Reds game I never attended. I was nine, I was certain my parents would say no, and I turned down the invitation without asking them. They would have said yes. I understood that day for decades as a failure to read my parents. It wasn't. The mistake was predicting them at all when my mother was a two-minute walk away. The full story, and a more recent version involving someone I've interviewed for six years and still got wrong, is in this week's issue of Reading Between the Lines.

The turn is what this means for anyone deciding on someone else's behalf, which includes every business that studies its customers. A decision made about a person without asking them is a decision made about someone who doesn't exist: a synthetic twin assembled from the decider's own assumptions. My industry has a literal version of this in the debate over synthetic respondents. They can tell you what someone like your customer would most likely say. They can't tell you what your customer would say if asked.

Key takeaways

  • Eyal, Steffel, and Epley (2018) found across 25 experiments that instructing people to take another's perspective did not improve accuracy about that person's thoughts and feelings, and sometimes reduced it while increasing confidence.

  • What did improve accuracy was getting perspective: asking the person directly.

  • Kardas, Kumar, and Epley (2022) found people underestimate how much others care about what they would share, and that correcting this expectation leads people into deeper conversations.

  • Empathy is not prediction. Forecasting behavior is what a plausibility engine does well. Understanding a person well enough to support them requires information only that person holds.

Common questions

  • What is the difference between perspective-taking and perspective-getting? Perspective-taking is imagining another person's point of view from the outside. Perspective-getting is obtaining it from them, by asking and listening. The research supports the second and not the first.

  • Why don't people have deeper conversations if they go so well? Because they expect the opposite, and those expectations are consistently wrong. The expectation itself is what keeps conversations shallow.

  • Is empathy the same as predicting what someone will do? No. Prediction estimates likely behavior from prior information. Empathy is understanding someone's feelings and perspective well enough to support them or reach a better outcome together. A system can be excellent at the first with no access to the second.

Read the full story: the missed ballgame, the AI that couldn't verify its own inner workings, and the interview that proved six years of data still isn't the same as asking.

Watch the Companion Video

This post extends the sixth episode of my podcast, What Does It Mean to Be Human?, with Maria Ross.

Additional Links

Rob Volpe
Next

Exposure Is Not the Same as Contact (Here's the Psychology)