COLUMN

Beyond "that's so me": the three measures at the centre of Aimii

Last updated: 2026-08-25 · About 4 min read

Why a type feels accurate

In 1948 the psychologist Bertram Forer gave each of his students a personality profile written, he told them, for them alone. They rated it 4.26 out of 5 for accuracy. Every student had the same page — assembled from an astrology book out of sentences that fit anybody^[1].

Type systems like the MBTI have the same softness underneath. Sixteen types assume personality arrives in two humps; measurement keeps finding one distribution, thickest in the middle^[2]. The boundary is not a gap in the data, it is a line somebody drew — which is why, retaking it within five weeks, about half of people come out a different type^[3]. Not because they changed. Because most people stand near a line.

If a result once felt accurate, that feeling was real. It may just have been accurate about everybody.

Aimii reads what you actually sent

Not answers to questions. Messages already sent — which is why the sentences come out differently.

Not “you can be cautious, though you also have a bolder side” but “of the eleven times you sent two messages in a row, three were followed by silence for a day”.

The second cannot fit everybody. If those eleven moments are not in your conversation, it never appears. And if it is wrong, you can prove it wrong — the moments are counted and shown, so you can look at the three and say no. There is nothing to do that with in a personality test.

Aimii picks up other things too — who does the asking, how long replies take, how often either of you asks a question, how long the messages run. But three of them sit at the centre, and all three are things relationship science spent decades confirming. When 43 longitudinal studies of some 11,000 couples were run through machine learning in 2020, what best explained satisfaction was not personality at all — it was what was actually happening between the two people^[9].

1. Do they take up what you said?

Reis and colleagues call it perceived partner responsiveness^[4][5]: feeling understood, valued, cared for. It is among the most consistently supported ideas in the field.

The difference is small on the page and decisive to receive. You say your hands shook during a presentation. “Nice one” is warm and touches none of it. “That shaking-hands thing — did it settle after? You were worried about the last slide” takes four seconds longer to type and is a different message entirely. AI judges which one a reply is, and judges warmth separately, because a kind answer that engages with nothing is a real thing.

2. After it goes wrong, does someone bring it back?

Gottman's repair attempt: the apology, the softening, the joke, reaching out again. The finding is consistent — whether you recover matters more than how often you argue^[6].

The hard part is knowing what counts as friction. A bad day is not friction. A late train is not friction. AI decides that first, and only then who reached back and whether it landed.

3. Is one of you pushing while the other goes quiet?

Demand–withdraw, from Christensen and Heavey. One presses for an answer, the other goes quiet, and the pressing hardens because of the quiet. A meta-analysis of 74 studies and some 14,000 people finds it tracking dissatisfaction^[7][8] — and it is a shape two people make together, not a label for either of them.

A rule sees three messages and nine hours of silence. It cannot see that the three were jokes, that the nine hours were a night shift, or that this pair are simply slow with each other. That is the judgement, and it is why it is done by a model reading the words rather than arithmetic on timestamps.

Most people are quietly surprised

Because something turns up that they had not noticed: that the asking was always theirs, that they were the faster replier, or that nothing much was wrong after all.

The conversation is read on your own device, and only a few dozen selected lines are sent. Three readings a day are free.

References

1. Forer, B. R. (1949). The fallacy of personal validation: A classroom demonstration of gullibility. Journal of Abnormal and Social Psychology, 44(1), 118–123. doi:10.1037/h0059240
2. McCrae, R. R., & Costa, P. T. (1989). Reinterpreting the Myers-Briggs Type Indicator from the perspective of the five-factor model of personality. Journal of Personality, 57(1), 17–40. doi:10.1111/j.1467-6494.1989.tb00759.x
3. Pittenger, D. J. (2005). Cautionary comments regarding the Myers-Briggs Type Indicator. Consulting Psychology Journal: Practice and Research, 57(3), 210–221. doi:10.1037/1065-9293.57.3.210
4. Reis, H. T., Clark, M. S., & Holmes, J. G. (2004). Perceived partner responsiveness as an organizing construct in the study of intimacy and closeness. In D. J. Mashek & A. Aron (Eds.), The Handbook of Closeness and Intimacy (pp. 201–225). Lawrence Erlbaum.
5. Gable, S. L., & Reis, H. T. (2010). Good news! Capitalizing on positive events in an interpersonal context. Advances in Experimental Social Psychology, 42, 195–257.
6. Gottman, J. M., & Levenson, R. W. (1992). Marital processes predictive of later dissolution: Behavior, physiology, and health. Journal of Personality and Social Psychology, 63(2), 221–233. doi:10.1037/0022-3514.63.2.221
7. Christensen, A., & Heavey, C. L. (1990). Gender and social structure in the demand/withdraw pattern of marital conflict. Journal of Personality and Social Psychology, 59(1), 73–81. doi:10.1037/0022-3514.59.1.73
8. Schrodt, P., Witt, P. L., & Shimkowski, J. R. (2014). A meta-analytical review of the demand/withdraw pattern of interaction and its associations with individual, relational, and communicative outcomes. Communication Monographs, 81(1), 28–58. doi:10.1080/03637751.2013.813632
9. Joel, S., Eastwick, P. W., Allison, C. J., et al. (2020). Machine learning uncovers the most robust self-report predictors of relationship quality across 43 longitudinal couples studies. PNAS, 117(32), 19061–19071. doi:10.1073/pnas.1917036117