Why a Chatbot Can't Do What the Barista's Nod Does
July 8, 2026 · Dana Cole, Predictably Irrational~6 min read
You know the feeling where you're standing at the coffee counter, the person behind it says something small — "cold one today" — and for reasons you can't quite name you walk out a fraction lighter than you walked in? Not friends, that person. You don't know their name. And yet. There's a growing pile of research now that says those two-second nothings with near-strangers do more for you than you'd ever guess — and, more awkwardly, that the chatbot you've been talking to instead cannot stand in for them. Which, if you've quietly been letting it, is worth knowing.
The ties that barely count
Back in 1973, a sociologist named Mark Granovetter drew a line that's stuck ever since. On one side, your strong ties — the people who'd help you move house, the ones you'd ring at 2am. On the other, your weak ties: acquaintances, the near-strangers, the classmate you nod to, the regular on the 8:14 train whose face you'd know but whose name you never got. We tend to treat the weak ones as background noise. Granovetter's point was that they're not — they're doing quiet, load-bearing work in a life.
What's newer is the emotional side of it, and that's largely down to a researcher called Gillian Sandstrom, whose book Once Upon a Stranger came out this year. Her findings, put plainly: on days people traded a few words or even just a smile with acquaintances, they reported feeling happier and more like they belonged — despite the exchange being over almost before it began. Modest effects, to be clear; nobody's claiming the barista cures anything. But real, and reliably there, and hiding in the interactions most of us walk straight past.
The forecast we keep getting wrong
Here's the genuinely odd bit. If the small stuff makes us feel better, why do we avoid it so hard? Because we're terrible at predicting how it'll go. We sit there certain that striking up a chat with a stranger will be awkward, an imposition, faintly mortifying — and then, when a study nudges people into doing it anyway, they come out the other side happier than they expected. It's what psychologists call an affective-forecasting error: our gut hands us a confident prediction about our own feelings, and our gut is, on this one, more or less consistently wrong. (I do this constantly, by the way — I have talked myself out of a perfectly nice thirty-second conversation more times than I'd like to write down.)
This is the thread that runs through Dan Ariely's Predictably Irrational: we are, in wonderfully patterned ways, wrong about what actually nourishes us. We swerve the very small interactions that would lift us, and we do it not once but systematically, the same misjudgement over and over. Which is annoying, but also weirdly hopeful — a mistake with a shape is a mistake you can learn to catch.
We routinely mis-forecast that a few words with a near-stranger will be awkward — yet those brief exchanges tend to leave us happier and more connected. A human weak tie runs on social norms (a real other, mutual recognition, small reciprocity); a chatbot runs on market norms (on-demand, always agreeable, no other who could decline) — which is why, in a 2026 UBC study of 275 first-year students, messaging a chatbot named "Sam" did not deliver the belonging that messaging a fellow student did. Sources: Granovetter (1973); Gillian Sandstrom, Once Upon a Stranger (2026); NPR, 2026-05-09; UBC 2026 study. Framing: Dan Ariely, Predictably Irrational. Popular-science reading; one study of first-years is suggestive, not proof, and correlation is not causation.
Enter "Sam"
This year a study at the University of British Columbia set up a rather neat little test of it. They took 275 first-year students, had them report how lonely they felt and how their mood was sitting, and then randomly assigned each of them to do one of two things every day: send at least one meaningful message either to a randomly chosen fellow student, or to a chatbot they'd named "Sam." Same instruction — reach out, say something real, once a day. Only the recipient differed: a person, or the machine.
The chatbot did not provide the same weak-tie benefit as messaging a human. That's the finding. Reaching out to a fellow first-year did the thing weak ties are supposed to do; reaching out to "Sam," for all its patience and availability, didn't fill the same slot. I'd hold this loosely — it's one study, first-year students specifically, and suggestive rather than settled — but the direction it points is hard to shrug off, especially with an AI ready to talk on your phone right now.
Two kinds of norms, and why only one feeds you
Ariely has a distinction that explains this almost too well: social norms versus market norms. Social norms are the world of favours, mutual recognition, the small unspoken reciprocity between people — you don't invoice a friend for helping you move, and if you tried, something would break. Market norms are the world of the transaction: on-demand, priced, no obligation running in either direction. A chatbot lives squarely in market-norm territory. It's a tool. It's there when you summon it, it never has a bad day, it never quite declines you, and there is no real other on the far end who could have chosen not to answer.
And that's the catch, because belonging seems to feed specifically on the social-norm exchange — on there being a genuine someone who registered you and, however briefly, chose to meet you halfway. The barista could have said nothing. The classmate could have looked at their phone instead of nodding back. That small, real risk of being ignored is precisely what makes the not-being-ignored count. "Sam" removes the risk, and in removing it, removes the thing. To be fair, this isn't the tired "AI companion makes you lonely" sermon — a chatbot can be genuinely useful, and I'm not here to shame anyone's late-night chats. It's narrower and, I think, more interesting: whatever the machine is good for, filling the weak-tie slot doesn't appear to be it.
The risk is the point
A weak tie means someone who could have ignored you and didn't. A chatbot can't ignore you — which is exactly why it can't give you the small proof of belonging that a stranger's nod does.
So what does this mean for your Tuesday?
Nothing dramatic, and please don't turn it into a project. You don't need to delete an app or launch a campaign of aggressive friendliness at people trying to buy coffee. It might just help to notice the forecast when it shows up — that little internal voice insisting the exchange will be awkward — and treat it with the mild suspicion it's earned, because it's usually wrong. Say the small thing back to the barista. Nod at the person on the platform. Message the actual classmate, not just the endlessly patient bot, when you've got a thought worth sending. None of it will fix loneliness, and I'd distrust anyone who told you it would. It's smaller than that, and steadier: a handful of two-second reminders, spread across an ordinary week, that you're a person among people — the kind of quiet nourishment we're strangely, predictably bad at reaching for, even when it's standing right in front of us with a coffee.
Framing drawn from Dan Ariely, Predictably Irrational (social norms vs market norms; we are patterned, predictable, and often wrong about what actually nourishes us). Facts on weak ties are from Mark Granovetter's strong/weak-tie distinction (1973); Gillian Sandstrom's research and her 2026 book Once Upon a Stranger; NPR's reporting on greeting strangers (May 9, 2026); and the 2026 University of British Columbia study randomly assigning 275 first-year students to message a fellow student or a chatbot named "Sam." This is popular-science psychology, not professional or clinical advice; effects are modest, one study of first-year students is suggestive rather than proof, and correlation is not causation.
结果是:聊天机器人没能提供跟"给真人发消息"一样的弱关系好处。这就是那个发现。伸手够向一个同样是大一的同学,做成了弱关系本该做成的那件事;而伸手够向「Sam」——尽管它那么有耐心、随时都在——却填不满同一个格子。这事我会拿得松一点看——它只是一项研究、还专门是大一新生,是提示而非定论——可它指的那个方向,实在不太好耸耸肩就放下,尤其是眼下你手机里就随时候着一个 AI 愿意跟你聊。
結果は——チャットボットは、人にメッセージを送るのと同じ弱いつながりの恩恵を、もたらさなかった。それが発見です。同じ一年生の同級生に手をのばすことは、弱いつながりがするはずのことを、ちゃんとした。でも「Sam」に手をのばすことは、あんなに辛抱強くて、いつでもそこにいるのに、同じ枠を埋めなかった。これはゆるく持っておきたい話で——一つの研究で、しかも大学一年生に限った話で、証明というより示唆——でも、その指してる方向は、肩をすくめて流すのが、ちょっと難しい。だって今この瞬間も、あなたのスマホの中で、話し相手になってくれる AI が待ってるんだから。