Every personality statistic you've read comes from people describing themselves. For eleven months we did the opposite: 1,179 people handed an AI their real chat logs and tweets, and it typed them from the language alone — on four systems at once, which means for once the systems can be checked against each other.
Where this data comes from
Between 10 September 2025 and 14 August 2026, people handed TypeMyVibe their real writing — chat exports, tweets, message history — and the model typed them across MBTI, Enneagram, Big Five (SLOAN) and moral alignment from the language itself, with no quiz and no self-report.
That produced 2,177 completed analyses, which is not the number we report. Many people ran the same conversation more than once, and our own testing account alone accounted for 591 of them. Counting those would inflate the sample and over-weight whoever ran the tool most. So every figure below comes from 1,179 distinct people, one reading each — the most recent analysis per subject, with operator accounts removed entirely.
The check almost no typology study runs
Because all four systems were scored from the same text in the same pass, they can be checked against each other. MBTI's Introversion and Big Five's Reserved are supposed to track the same underlying trait. So are MBTI's Judging and Big Five's Organized. Neither scorer sees the other's output. If the method is picking up something real, those pairs should move together.
| Trait | MBTI says | Big Five says | Agree on the same person |
|---|---|---|---|
| Introversion (MBTI I / SLOAN R) | 77.4% | 77.6% | 80.2% |
| Openness (MBTI N / SLOAN I) | 81.8% | 68.9% | 73.6% |
| Agreeableness (MBTI F / SLOAN A) | 39.5% | 48.1% | 63.7% |
| Conscientiousness (MBTI J / SLOAN O) | 16.1% | 35.4% | 64.9% |
77.4% against 77.6%. Two independent scorers, reading the same conversations for different systems, disagreed on the population rate of introversion by two-tenths of a point — and agreed on 80.2% of individuals. Whatever introversion looks like in written conversation, both scorers are seeing the same thing.
Conscientiousness is the opposite story, and we would rather publish it than bury it. MBTI scoring called only 16.1% of people Judging; Big Five scoring called 35.4% of the same people Organized. That is a 19.3 point gap, with individual agreement of 64.9% — barely better than chance on a split like this.
Two types are two-thirds of everyone
In the general population, INTPs and INFPs are usually estimated in the low single digits each. Here, INTP alone is 42.3% (491 of 1,161 determinate verdicts) and INFP is 23.3% (271). Together, two of sixteen types account for 65.6% of everyone. Third place falls off a cliff: ENTP at 5.9%, ENFP at 5.9%, then ISTJ (4.7%) and ESFJ (3.3%).
The letter marginals say it louder: 77.4% Introvert, 81.8% Intuitive, 82.9% Perceiving, 63.6% Thinking. The archetypal person who asks an AI "who am I, really?" is an INxP — inward, abstract, open-ended, and just skeptical enough to distrust their own quiz answers.
Is INFJ the rarest MBTI type? The INFJ inversion
INFJ is usually named the rarest MBTI type, at about 1.5% of the US population in the national sample published in the MBTI Manual (third edition, 1998), just below ENTJ (about 1.8%) and INTJ (about 2.1%). In this study, typed from real messages rather than self-report, INFJ came out at 1.2% — rare here too, even though it is one of the types people online claim most.
Here's our favourite result. INFJ — famously "the rarest type," and by every informal measure the type the internet most loves to claim — almost never survives contact with someone's actual messages. Just 14 people in 1,161, or 1.2%, typed as INFJ. That is fewer than ESTP (27) and even ESFP (15). Self-report tests hand out INFJ generously; language analysis, which cannot be flattered, hands it out almost never. Whichever number is closer to the truth, the gap itself is the story: the distance between the types people want and the types their words support.
The pairings the forums guessed right
Typology communities have long-standing beliefs about which MBTI types map to which Enneagram numbers. INTPs are "obviously" Fives; ESFJs are "obviously" Twos. These claims circulate as received wisdom and are rarely tested against anything. Scored independently, from text alone, they hold up:
INTP → Enneagram 5
The strongest cross-system pairing in the dataset. More than half of everyone typed INTP was independently scored a Five.
ESFJ → Enneagram 2
The Helper pairing typology forums treat as self-evident, reproduced from conversational text with no MBTI input.
INFP → Enneagram 4
The Individualist mapping holds, though less tightly than INTP-5. A quarter of INFPs came out Sixes instead.
ISTJ → Enneagram 5 or 6
Split rather than singular: 34.5% Five, 27.3% Six. Two routes to the same caution.
This matters more than it looks. The Enneagram scorer had no sight of the MBTI verdict. If the readings were arbitrary these pairings would be flat; instead they reproduce, from conversational language, a structure originally worked out by theorists interviewing people. Overall, Type 5 takes 32% of the sample and Type 6 another 20% — the head triad dominates this dataset the way it dominates every typology forum.
Big Five: reserved, limbic, and up at 2am
The SLOAN marginals: 77.6% Reserved, 68.4% Limbic, 64.7% Unstructured, 68.8% Inquisitive, and an almost perfectly split 51.8% Egocentric. The most common full codes are RLUEI (167), RLUAI (151) and RLUEN (93) — quiet, emotionally intense, allergic to rigid plans. Three in four carry the R. If you have ever wondered who is awake retyping a message for the fourth time, statistically it is this crowd, and they know it.
The internet is chaotic — we measured it
Typed from real messages, 59.5% land in some flavour of Chaotic — Chaotic Good alone is 32.2%, Chaotic Neutral another 21.3% — while all three Lawful alignments together muster 9.5%. Worth noting for anyone who saw an earlier version of this page: on the de-duplicated sample, Lawful Good (7.3%) slightly outnumbers Chaotic Evil (6.0%). The earlier figures counted repeat runs, and that one reversed when we stopped. (Take the alignment test if you want to know your own odds.)
The more obscure the system, the harder the model leans
Attitudinal Psyche — a Russian typology far less represented in any training corpus than MBTI — produced the most lopsided output of all. A single code, FVEL, took 41% of its 315 readings, against a top MBTI type at 42% of 1,161 and a top Enneagram type at 32%. For a well-documented system, concentration plausibly reflects the sample. For a thinly-documented one it more likely reflects a scorer falling back on a familiar answer — worth knowing before you take any obscure-typology result from an AI at face value.
What this actually tells us
Three honest readings. One: typology is a hobby with a house style — introspective, abstract, slightly anxious — and any personality product's early adopters will look like this. Two: text-based analysis genuinely distributes types differently than self-report quizzes, because it scores what you do in language rather than what you believe about yourself; the INFJ inversion is the cleanest evidence, and the near-perfect introversion replication is the strongest reason to think it is measuring something. Three: it is not uniformly reliable, and the conscientiousness gap shows exactly where it is not.
Methodology in one line: completed analyses from 10 Sep 2025 to 14 Aug 2026, de-duplicated to one reading per person, operator accounts excluded, determinate verdicts only, aggregates only, no demographic weighting. If you are writing about this study, cite it as "TypeMyVibe cross-system text-typing dataset, August 2026 (n = 1,179)" and link this page — and if you want a cut of the data we did not publish, ask us.