Does this text sound desperate?

You have written something warm and now you are shortening it. Cutting the second sentence, deleting the bit that said you were looking forward to it, wondering whether two messages in a row is too many. I co-founded Subtext, and this specific edit, the one where someone takes the warmth out of a message they already wrote, is the most common thing I see.

The research says you are making a mistake in two directions at once, which is the most useful thing I have read on this subject. People consistently overestimate how badly warmth, effort and vulnerability land. They also underestimate how badly low effort lands. And the standard advice for not seeming desperate, which is shorter messages, slower replies and less visible enthusiasm, is a set of effort-reduction strategies. The advice recommends the exact behaviour the evidence penalises.

Warmth lands better than you think

Five independent research programmes point the same way, which is unusual in this field.

The liking gap. Boothby, Cooney, Sandstrom and Clark found people systematically underestimate how much their conversation partners liked them¹. It replicated in the lab with strangers, among first-year students living together across an academic year, and among members of the public at a workshop.

Vulnerability. Bruk, Scholl and Bless ran seven studies and found people rate their own displays of vulnerability, including confessing romantic feelings, more negatively than observers rate identical displays in other people². You are harsher on your own message than anyone else will be.

Reaching out. Liu, Rim, Min and Min ran experiments with more than 5,900 participants and found people underestimate how much others appreciate being contacted, with the effect larger when the contact is more of a surprise³. Two size caveats belong next to that: the authors flagged their own focal experiment as underpowered at d = 0.32, and an independent replication with 742 participants reproduced the effect at d = 0.25. Real, replicated, and modest. The message you think is an imposition is usually received as a nice surprise, just not as dramatically as the headline versions suggest.

Gratitude. Kumar and Epley had people write gratitude letters, predict how recipients would react, then measured actual reactions⁴. Expressers underestimated how positive recipients would feel and overestimated how awkward it would be. The important bit is the mechanism: expected awkwardness predicted whether people were willing to express gratitude at all. The miscalibration does not just make you nervous, it stops you sending the message.

The illusion of transparency. Gilovich, Savitsky and Medvec showed people overestimate how visible their internal states are to others⁵. If you feel keen while typing, you assume the keenness is showing. It usually is not.

That last one cuts both ways and I want to be fair about it. It supports the idea that a sender who hears their own message as desperate is hearing an internal state the recipient cannot detect. It equally implies you cannot be confident that your intended warmth is arriving. Both follow from the same finding, and it is the reason Subtext exists as a second pair of eyes rather than a rulebook: neither the anxious reading nor the confident one is something you can fully trust from inside your own head.

Low effort lands worse than you think

This is the mirror image, and it is the more actionable one.

Fang, Zhang and Maglio ran eight preregistered studies with 5,306 participants, spanning vignettes, a Discord field experiment with roughly 1,900 users, interactive speed dating run during Valentine’s week, and archival analysis of Tinder conversation histories across 37 countries⁶.

The self-report numbers are the ones that should stop you. 99 per cent of respondents used texting abbreviations, and 84 per cent did not believe others would dislike them. In a later study, 80 per cent predicted others would be indifferent and 4.2 per cent expected a positive reaction.

Recipients rated abbreviating senders as less sincere and were measurably less likely to write back. In the Tinder data, higher netspeak scores predicted shorter conversations. The effect was mediated by perceived sender effort, and familiarity between the two people did not weaken it.

So the reason to worry is the opposite of the one you have. Nobody is scoring your enthusiasm. They are noticing whether you put anything into it.

This is the asymmetry Subtext is built around. Most people using it are trying to make a message safer, and the useful thing it does is tell them when the safer version has become the emptier one. Those are the same edit more often than anyone expects.

Where eagerness genuinely costs you

Two zones. Both narrower than the folk version.

Indiscriminate interest, spread across many people. Eastwick, Finkel, Mochon and Ariely used a speed-dating design with 156 undergraduates across seven sessions⁷. People who expressed romantic desire for most of their partners were less likely to have it reciprocated. Finkel’s summary of the mechanism is that people who like everyone may exude desperation in a romantic context, in a way they do not in friendship.

Read the finding precisely. What was penalised was liking everybody, not warmth towards one person. Selectivity is the variable, and it is not the same thing as coolness. It is also the one thing on this page that a tool cannot see from a single message, which is why Subtext reads a thread rather than a line.

Reassurance seeking that does not stop when reassurance arrives. The research construct is excessive reassurance seeking: persistently seeking assurance of being lovable regardless of the assurance already given⁸. Starr and Davila meta-analysed 38 studies covering 6,973 people and found it correlated with depression at r = .32⁹.

The number that matters for this article is the other one in the same paper. Across 16 studies, the correlation between reassurance seeking and actual interpersonal rejection was only .14. So the behaviour is robustly tied to how the person seeking feels, and only modestly tied to whether anyone actually pulls away. Which cuts against the folk assumption that visible neediness reliably drives people off. The distinction Subtext tries to surface is between a message asking something real and a message asking the same thing again, because those look nearly identical on the screen and do very different things to the person reading them.

Playing hard to get is mostly misreported

The founding study is the most misdescribed thing in dating advice.

Walster, Walster, Piliavin and Schmidt ran six experiments¹⁰. Five failed to find that a hard-to-get woman was preferred. Only the sixth found support, and only for selective unavailability: hard for other men to get, warmly receptive to the participant. The paper gets cited as proof that aloofness works, when the authors could only produce the effect after adding the selectivity twist.

Later work sharpens the same point. Dai, Dong and Jia found playing hard to get simultaneously raised wanting, meaning motivation to pursue, and lowered liking, meaning how much the person actually enjoyed the target¹¹. And it only worked when the target was already psychologically committed to pursuing. Without prior commitment, playing hard to get lowered both. Male-only Hong Kong student samples, so a real limit on generalising.

Birnbaum, Zholtack and Reis found being hard to get raised perceived mate value and sexual desire, and Reis is explicit about the boundary: if it makes you seem disinterested or arrogant, it backfires¹².

The supportable claim is narrow. Signalling selectivity and some scarcity can raise perceived mate value, conditional on prior interest. It does not support being cold, slow or withholding towards the specific person you actually want. Popular advice collapses “selective” into “uniformly unavailable”, and those are different things with different evidence behind them. Nothing about playing hard to get is a texting strategy Subtext will help you execute, since the version with any support behind it is about who you are warm to, not about withholding warmth from the one person you are actually writing to.

The uncertainty question, which is genuinely unresolved

Two studies, opposite directions, and I would not pretend to resolve it.

Whitchurch, Wilson and Gilbert had 47 female undergraduates view fabricated profiles and told them the men had rated them highest, average, or that it was uncertain¹³. The uncertain condition produced the highest attraction. That result is the basis for a lot of advice about keeping people guessing.

Take it lightly. Small, female-only, single study, no interaction with anyone, and the authors themselves flagged their mediation as tentative because the confidence interval did not exclude zero.

Birnbaum and colleagues ran six studies and found the reverse¹⁴. In their first study, a prospective partner was rated more sexually desirable under certainty than under uncertainty. Their proposed mechanism is self-protection: inhibiting desire shields you from investing in something whose future is unclear.

The plausible reconciliation, which is an interpretation rather than a demonstrated finding, is that rejection risk is doing the work. Browsing a profile carries no possibility of personal rejection, so uncertainty is a low-stakes puzzle that makes you think about someone more. In live interaction, uncertainty signals possible rejection, and defensive distancing kicks in. Texting someone you are actually talking to is the second situation.

Does the tone you are worried about even transmit?

Mostly yes, which is reassuring in a way most articles on this subject are not.

Pollmann and Roos asked both people¹⁵. Participants uploaded a real message thread where they had asked a friend to do something, rated the reply’s warmth, and then the friend who wrote it was asked what they had meant. Across 171 matched pairs, intended and perceived warmth were statistically indistinguishable, and an equivalence test ruled out any gap larger than half a point on a nine-point scale.

They also tested message length, emoji, reader gender, reader age, closeness and reader neuroticism as moderators. None of them predicted misreading.

The limits the authors state, which matter here: the messages were mostly positive, the topics were benign, and everyone already knew each other. So this does not overturn the finding that senders overestimate how well tone travels¹⁶. It says that in ordinary friendly messages, the reading is roughly right. It says nothing about the harder case: a newer relationship, a heavier topic, or a moment when you genuinely cannot tell how something will land, which is closer to what Subtext is actually there for.

What to actually do

Do not cut the warm sentence. Every finding in the first section says you are pricing it wrong, and the effort research says the shorter version carries a documented cost. It usually looks like this:

Edited down “That was fun. Let’s do it again sometime.”

Shorter, not safer · the effort-reduction move the evidence above penalises · the line cut is exactly the warmth that lands better than you think

What was actually there “That was really fun, I haven’t laughed that hard in ages. Would love to do it again soon if you’re around.”

The second one is the one worth sending.

Do not abbreviate to seem casual. It is the single behaviour in this literature with clear evidence of reducing replies, and almost nobody thinks it will.

Watch selectivity rather than intensity. The one place eagerness genuinely costs you is spreading it across everyone. Being warm to one person is a different behaviour from being warm to twelve, and only the second one was penalised.

Notice whether reassurance is landing. If you have asked and received an answer and the question comes back the next day in different words, that is the loop, and it is the thing worth attending to. It is also a pattern where talking to someone, a friend or a professional, does more than any wording change will.

Subtext will tell you when a message reads warmer than you intended, and more usefully, when the version you have edited down has lost the thing that made it worth sending.

Free to start, on your phone or in the browser.

---

Think I have read a study wrong, or know research on this I have missed? Tell me on LinkedIn.

Samet Durgun is the co-founder of Subtext, an app that catches the emotional tone of your messages and rewrites them in your own voice. He’s based in Berlin.

Sources

Numbered in the order they appear above. Where a figure could not be verified against the primary source, the text says so. Checked 28 August 2026.

1. Boothby, E. J., Cooney, G., Sandstrom, G. M., and Clark, M. S. (2018). The Liking Gap in Conversations. Psychological Science, 29(11), 1742 to 1756. https://clarkrelationshiplab.yale.edu/sites/default/files/files/BoothbyCooneySandstromClark2018.pdf
2. Bruk, A., Scholl, S. G., and Bless, H. (2018). Beautiful mess effect: Self-other differences in evaluation of showing vulnerability. Journal of Personality and Social Psychology, 115(2), 192 to 205. https://psycnet.apa.org/record/2018-32285-002
3. Liu, P. J., Rim, S., Min, L., and Min, K. E. (2023). The Surprise of Reaching Out. Journal of Personality and Social Psychology, 124(4), 754 to 771. More than 5,900 participants across experiments; the focal experiment was flagged as underpowered by its own authors at d = 0.32, and an independent replication with 742 participants found d = 0.25. https://psycnet.apa.org/record/2022-64599-001
4. Kumar, A., and Epley, N. (2018). Undervaluing Gratitude: Expressers Misunderstand the Consequences of Showing Appreciation. Psychological Science. https://doi.org/10.1177/0956797618772506
5. Gilovich, T., Savitsky, K., and Medvec, V. H. (1998). The illusion of transparency. Journal of Personality and Social Psychology, 75(2), 332 to 346. https://psycnet.apa.org/record/1998-04730-003
6. Fang, D., Zhang, Y., and Maglio, S. J. (2025). Shortcuts to Insincerity: Texting Abbreviations Seem Insincere and Not Worth Answering. Journal of Experimental Psychology: General, 154(1), 39 to 57. Eight preregistered studies, 5,306 participants. https://www.apa.org/pubs/journals/releases/xge-xge0001684.pdf
7. Eastwick, P. W., Finkel, E. J., Mochon, D., and Ariely, D. (2007). Selective versus unselective romantic desire. Psychological Science, 18(4), 317 to 319. Seven speed-dating sessions, 156 undergraduates. https://pubmed.ncbi.nlm.nih.gov/17470256/
8. Joiner, T. E., Metalsky, G. I., Katz, J., and Beach, S. R. H. (1999). Depression and Excessive Reassurance Seeking. Psychological Inquiry, 10(3), 269 to 278. https://www.jstor.org/stable/1449387
9. Starr, L. R., and Davila, J. (2008). Excessive Reassurance Seeking, Depression, and Interpersonal Rejection: A Meta-Analytic Review. Journal of Abnormal Psychology, 117(4), 762 to 775. https://www.psych.rochester.edu/research/starrlab/wp-content/uploads/2014/08/meta-analysis.pdf
10. Walster, E., Walster, G. W., Piliavin, J., and Schmidt, L. (1973). Playing hard to get: Understanding an elusive phenomenon. Journal of Personality and Social Psychology, 26(1), 113 to 121. Five of six experiments failed to find the effect. https://psycnet.apa.org/record/1973-24304-001
11. Dai, X., Dong, P., and Jia, J. S. (2014). When does playing hard to get increase romantic attraction? Journal of Experimental Psychology: General, 143(2), 521 to 526. Male-only Hong Kong student samples. https://doi.org/10.1037/a0032989
12. Birnbaum, G. E., Zholtack, K., and Reis, H. T. (2020). No pain, no gain: Perceived partner mate value mediates the desire-inducing effect of being hard to get. Journal of Social and Personal Relationships, 37(8 to 9), 2510 to 2528. Per-study samples not verified. https://journals.sagepub.com/doi/10.1177/0265407520927469
13. Whitchurch, E. R., Wilson, T. D., and Gilbert, D. T. (2011). He loves me, he loves me not: Uncertainty can increase romantic attraction. Psychological Science, 22(2), 172 to 175. 47 female undergraduates; the authors flag their own mediation as tentative. https://journals.sagepub.com/doi/10.1177/0956797610393745
14. Birnbaum, G. E., Kanat-Maymon, Y., Mizrahi, M., Barniv, A., Govinden, S., and Reis, H. T. (2018). Are you into me? Uncertainty and sexual desire. Computers in Human Behavior, 85, 372 to 384. Six studies; points in the opposite direction to source 13. https://www.sciencedirect.com/science/article/abs/pii/S0747563218301663
15. Pollmann, M. M. H., and Roos, C. A. (2025). “I get u”. People correctly interpret the tone of text messages and emails. Computers in Human Behavior Reports, 18, article 100689. A companion open-access title rather than the parent journal; recent and not yet replicated. https://doi.org/10.1016/j.chbr.2025.100689
16. Kruger, J., Epley, N., Parker, J., and Ng, Z.-W. (2005). Egocentrism Over E-Mail. Journal of Personality and Social Psychology, 89(6), 925 to 936. https://web-docs.stern.nyu.edu/pa/kruger_email_ego.pdf