When You Hear Every Word and Still Can't Connect

I had a conversation with a friend and colleague recently that I could not get right.

I heard every word he said. I could have repeated them back to him. And I still had nothing.

I sat through long pauses, trying to compute what he needed from me. I asked a couple of questions to narrow it down, and I watched them land wrong — because I was solving when I should have been listening.

By the end I felt completely detached from someone I’d normally read with no trouble. So I did the only honest thing I had left: I asked for a pause, to process.

I think I finally know what happened. I was running the wrong protocol.

Charles Duhigg (2024), in Supercommunicators, makes a claim that sounds too simple until it explains your whole week: almost every conversation is really one of three.

Practical — “What is this actually about? What do we decide?” Emotional — “How do we feel?” Social — “Who are we to each other?”

And the thing that makes a conversation click — he calls it the matching principle — is brutal in its simplicity: both people have to be in the same one. When you’re not, you can be speaking the same language, word for word, and miss each other completely.

That was us.

I was deep in Practical. What’s the problem, what are the options, what’s the next step. He was in Emotional — he wanted to be heard first, and then, yes, some practical next steps, but in that order.

So every move I made was technically correct and completely useless. My questions weren’t bad questions. They were the wrong kind of question. A brain in solving-mode can’t hear an emotional ask, because it has already filed the ask as a problem to be fixed.

The part that stayed with me is that this isn’t a personal defect. It’s structural.

Duhigg tells the story of Dr. Behfar Ehdaie, a prostate cancer surgeon at Memorial Sloan Kettering. Many of his patients have low-risk, slow-growing tumors, and the evidence says the right move is active surveillance — monitor it, don’t cut. He would explain this clearly, with the data. Patients would nod, go home, and come back days later demanding aggressive surgery anyway.

They weren’t ignoring him. They couldn’t hear him. They had just been told they have cancer, and they were sitting in an emotional conversation while he was running a practical one.

So he changed the opening (Ehdaie et al., 2017). Instead of leading with data, he started asking: “Tell me what this diagnosis means to you.” He let them say the fear out loud — family history, mortality, whatever it was. He shared his own, about his father’s illness. Only then did he talk about surveillance.

By Duhigg’s account, the share of patients who followed the evidence-based recommendation rose sharply once he led with the question instead of the data. I trust the direction of that more than any exact figure — it’s from his telling, not a clinical trial. But the shape of it is the whole point.

Same medicine. Same patient. Different conversation.

And there is real signal under the framework, not just a tidy model.

When two people genuinely connect, their brain activity starts to track each other. The coupling grows stronger the better the understanding, and vanishes the moment communication breaks down (Stephens et al., 2010). Separately, when people fall into sync they unconsciously start matching their smallest words — the pronouns, articles, and prepositions nobody notices — and that matching predicts everything from whether a speed date clicks to whether a couple is still together months later (Ireland et al., 2011). I’d hold the three buckets loosely. They’re a teaching tool, not a law of nature; there is no practical-emotional-social gene. But the claim underneath is solid: alignment is a real, measurable thing, and you feel its absence as detachment. Which is exactly what I felt.

So the lesson is annoyingly simple and genuinely hard to do: match before you solve.

Find out which conversation you’re in before you start being useful. Ask the deep question. Let the answer change the kind of response you give. The pause I took wasn’t the conversation failing — it was me realizing, too late, that I had never figured out which protocol we were on.

Here is what makes this funny for me.

I do a version of this for a living, pointed the other way. I train models to reason. When I want to know whether a model actually understands a task, I can write an eval function that scores it. I can curate a dataset that teaches it the thing it’s missing. I can open the reasoning trace and check whether it reached the answer honestly or just got lucky.

None of that exists for a person.

There is no eval function that returns which conversation my friend is in. There is no dataset of his inner state I can train against. There is no trace to read. I have to find all of it live, in the room, from a face and a pause and a shift in tone — and adjust before the next sentence leaves my mouth.

I spend my days making machines better at processing language. It turns out the harder problem was never the machine.

The thing in front of me was a person. And he just needed me to be in the same conversation he was already in.

References

Duhigg, C. (2024). Supercommunicators: How to unlock the secret language of connection. Random House. Publisher page

Ehdaie, B., Assel, M., Benfante, N., Malhotra, D., & Vickers, A. (2017). A systematic approach to discussing active surveillance with patients with low-risk prostate cancer. European Urology, 71(6), 866–871. PubMed · MSK profile

Ireland, M. E., Slatcher, R. B., Eastwick, P. W., Scissors, L. E., Finkel, E. J., & Pennebaker, J. W. (2011). Language style matching predicts relationship initiation and stability. Psychological Science, 22(1), 39–44. PubMed

Stephens, G. J., Silbert, L. J., & Hasson, U. (2010). Speaker–listener neural coupling underlies successful communication. Proceedings of the National Academy of Sciences, 107(32), 14425–14430. PNAS