Tell your friends you appreciate them. It lands better than you think

The hesitation has been measured directly. In a set of experiments published in 2018, people wrote a letter of gratitude to someone and predicted how that person would react; the recipients then reported how they actually felt. The writers got it wrong in a consistent direction. They underestimated how positive the recipient would feel, overestimated how awkward the recipient would find it, and underestimated how surprised the recipient would be.

The proposed mechanism is a mismatch in what each side is paying attention to. The person writing evaluates themselves on competence: is this well put, does it sound strange, is it too much. The person reading is not assessing the writing at all. They are registering that someone thought of them warmly enough to say so. Because the writer is grading a performance nobody else is watching, the cost they expect is inflated and the benefit is discounted.

The limits matter. These were largely one-off exchanges in study conditions, measured immediately afterwards. Nothing in them says that repeated expressions keep landing the same way, and there is no long-run data on what happens to a friendship afterwards. What they do establish is that the moment before you send it is a poor guide to how it will be received.

This is not a quirk of thank-you notes. In a separate series of experiments, people who reached out to someone unprompted predicted how the recipient would feel, and recipients appreciated the contact more than the senders expected. The gap grew the more surprising the contact was. Telling a friend out of nowhere that you value them is about as unprompted as contact gets, which puts it where the underestimate is largest. Those studies looked at single moments, not friendships over time, and nobody has shown that acting on the finding makes a friendship last.

The pattern holds for conversation too. When people were assigned to have either shallow or deep conversations, they predicted the deep ones would be more awkward and less connecting than they turned out to be, and they were most wrong about the most substantive topics. One proposed reason is that people underestimate how interested others are in them. Those conversations were assigned, often between strangers who knew they were in a study, so the stakes were lower than in life, and it does not follow that any given conversation will go well.

Put together, the three findings describe one consistent forecast: that warmth, contact and depth will cost more and land worse than they do. "It would be weird to say that" is what that forecast sounds like from the inside.

The relationship literature gives a working definition of what the other person is listening for. Intimacy, operationally, is feeling understood, validated and cared for, and perceived partner responsiveness, the sense that someone is paying attention to you, predicts the behaviours that keep a relationship going, gratitude among them. The word doing the work is perceived. What moves a relationship is not what you felt but what the other person registered.

A meta-analysis of relationship maintenance found the same thing from another angle: relationship quality tracked a person’s perception of their partner’s upkeep more closely than the partner’s own report of doing it. This is a framework resting on mostly correlational work, and responsiveness and closeness feed each other, so no direction is established. But it explains why appreciation that stays in your head does nothing for the friend it is about.

It also tells you what to say. "You mean a lot to me" is warm, but it could be sent to anyone. "I still think about how you drove over the night my flight was cancelled" could only come from someone who noticed. Referencing something specific is one of the few reliable ways to make attention visible, because it cannot be produced without having paid it.

The first is turning it into a balance. Research on communal and exchange relationships distinguishes two norms: in exchange relationships, benefits are given to create or repay an obligation; in communal ones, they follow from concern for the other person. Where closeness is wanted, prompt repayment of a benefit reduces attraction rather than increasing it. The studies were mostly short laboratory interactions, and real friendships mix the two norms. Our reading is that appreciation phrased as squaring up ("I owe you one", "I’ll make it up to you") borrows the exchange norm, while appreciation phrased as noticing does not.

The second is handing it to someone, or something, else. Messages known to involve AI were judged less trustworthy and less authentic, with larger penalties for greater involvement. The penalty attaches to what the reader believes produced the words, not to the words themselves. This is a young literature and the norms are shifting quickly, so the size of the penalty may not hold. But a polished message solves the problem the gratitude research says the reader does not have.

Neither of these is about getting the wording right. The reader is not marking it. They are reading it for evidence that it came from you and was about them.

Pick one friend and one specific thing. Not a summary of the friendship, a moment: something they did, said or remembered that mattered to you. Write two or three sentences about it, and resist the urge to redraft until it sounds eloquent, because eloquence is what you are grading and not what they are reading.

Send it without a reason. You do not need a birthday or an occasion, and the research on unprompted contact says the lack of one is part of why it lands. If the forecast says it will be awkward, treat that as the measured bias it is, not as information about your friend.

None of this needs an app. It needs the habit of noticing the moments worth thanking someone for while they are fresh, a note of them somewhere you will see again, and the decision to send the message when your own forecast tells you not to.