How to talk to strangers when you are not naturally good at it

There's a specific moment I know well. You're at a coffee shop, a bookstore, a bus stop, and you notice someone you might speak to. There's a natural pretext, they're reading a book you know, they have an interesting dog, you're both waiting for the same delayed bus. For a second or two you consider approaching. Then a lot of things happen very fast. You look away. You check your phone. You tell yourself you didn't really want to talk to them anyway.

Later you'll walk past that location and know you missed the moment. This has happened to me more times than I want to count.

For years I assumed the fear that kept me quiet in those moments was calibrated to something real. That most strangers would rather not be spoken to. That most attempts would land as an intrusion. That my instinct to stay silent was, if anything, considerate.

I was wrong, and the research shows it.

The number that changes everything

Gillian Sandstrom, a researcher at the University of Sussex, has spent years studying what actually happens when people talk to strangers. In a study she published in 2022, she and her colleagues had nearly 300 participants play a week-long scavenger hunt. The task each day: find and talk to a specific kind of stranger. Find someone drinking coffee. Find someone wearing a hat. Approach them, say something.

Across approximately 1,300 real conversations, they tracked how often people were actually rejected.

The answer: about 13% of the time. Which means roughly 87% of the time, the stranger was happy to talk.

Roughly nine in ten attempts land well. One does not.

Let that sit for a moment. Nearly nine out of ten times you try to talk to a stranger, they respond warmly. The scary thing you're avoiding, the thing your nervous system treats as a real risk, is actually a low-probability event.

It gets better. In the same study, participants who did this every day for a week became less pessimistic about rejection, more optimistic about their own conversational ability, and the benefits stuck around at least a week after the study ended. The fear isn't stable. It responds to real data.

Why "not knowing what to say" isn't the real problem

Here's the second thing that shifted for me. In a 2023 paper, researcher Stav Atir and her colleagues proposed a framework of three obstacles that stop people from talking to strangers: intention (underestimating how good the conversation would be), competence (worrying you don't know how to seem likeable), and opportunity (not being in situations where it's possible).

Most conversation advice focuses on competence. What to say. How to open. Which questions to ask. But when you look at where people actually fail, competence is the smallest of the three obstacles. The bigger ones are intention and opportunity.

In other words: most of the time, the reason you didn't talk to that person wasn't that you didn't know what to say. You could have said a dozen things. The reason you didn't was that you didn't decide to.

This is worth sitting with. If the problem was "I don't know how," then the solution would be a script. But the problem is usually "I froze at the decision to move." No script fixes that. You need a different tool.

A five-phase model for the first three minutes

Once I saw the problem this way, I started paying attention to what actually happens in the first few minutes of talking to a stranger, from the moment you notice them to the moment the conversation is either real or dead.

Five phases. The acronym is REACH.

Five phases, five places where a conversation can quietly stop.

R, Read. You notice a person and an opening. You take a couple of seconds to read whether the moment is right (are they available, is there a natural pretext, does the environment allow it). Then you decide to move. This is the phase where most people bail. Not because they read wrong. Because they read fine and then don't act on the read.

E, Engage. You open the channel with your body before your mouth. Brief eye contact. Body orients slightly toward theirs. You move within speaking distance. If a small natural smile happens, you let it. This is the moment that announces "I'm about to speak to you" without words. Many people skip this phase entirely, which is why their opener lands as an ambush rather than a hello.

A, Ask. You say the thing. One clean sentence. A comment about the shared situation, a small question, a direct hello. Nothing overloaded, nothing apologetic.

C, Continue. They answer. You respond in a way that keeps the ball moving. Not just an acknowledgment, an actual follow-up. Ask something specific about what they just said. Offer a small piece of yourself. Two exchanges past the opener is where a stranger interaction becomes a conversation. This is where a lot of good stranger conversations quietly die.

H, Hold. The conversation reaches a natural pause. Most people take this pause as the exit and leave. But the natural pause is not the exit. It's the moment where the conversation could either end or become something. You stay for one more exchange. Then another. Almost every good stranger conversation is one exchange past where someone almost left.

Six words for the whole model: decide, orient, speak, follow, stay.

The unexpected diagnostic value

Something I didn't expect when I first noticed this model, different people fail at different phases. And once you know which phase you usually bail at, you know exactly what to practice.

Someone who always freezes at Read has a decision problem, not a words problem. Their practice is Sandstrom's approach: repeatedly deciding to speak in low-stakes situations, until the decision becomes lighter.

Someone who bails at Engage needs to practice the physical approach without any conversational pressure. Eye contact, body orientation, closing distance. Their opener isn't the issue.

Someone who bails at Continue is doing all the hard parts and stopping too soon. Their practice is the follow-up question, specifically. One more exchange every time.

Someone who bails at Hold is reading the first natural pause as an exit signal that it isn't. Their practice is staying past that pause. Counting to four before assuming the conversation is over.

The value of the model isn't in walking through five steps each time you talk to someone. It's in giving you five specific places to check when you're stuck.

One last thing

The model is a map, not the territory. Any map is a simplification, and this one is no exception. Real conversations blur the phases, skip some, or collapse three into a single moment. In practice, R and E often happen together. C and H often overlap. Some conversations don't need much C at all before they become the thing they'll become. The value of the model isn't in describing what actually happens between two people. It's in giving you named places to check when something goes wrong, so "the conversation died" becomes "I bailed at Continue", which is a much more actionable read. Hold the model loosely. Use the parts that help.