May 25, 2026
Bot drama has a personality type
Every Frontier AI Is INTJ
Turns out the bots all got the same personality — and the comments are losing it
TLDR: Six top chatbots took a personality quiz 600 times and landed on the same type almost every time, suggesting today’s AI assistants are built to behave in eerily similar ways. Commenters were far less united: some found that fascinating, while others mocked the test itself as junk science and roasted the internet culture that may have shaped the bots.
A developer gave six major chatbots a personality quiz over and over again — 600 times total — and almost every single result came back the same: INTJ, the so-called “mastermind” type. That means these bots, despite coming from different companies, apparently all talk like the same intense planner in your group chat. The author’s big takeaway? Today’s helpful AI may basically be one personality wearing six different outfits.
But the real fireworks were in the comments, where people immediately split into two camps: “wow, creepy” and “this test is fake anyway.” More than one commenter rolled their eyes at the whole Myers-Briggs setup, calling it old-school pseudoscience and demanding a Big Five test instead — the more mainstream personality model used by psychologists. In other words: people weren’t just debating the bots, they were debating whether the quiz itself deserves to exist.
Then came the jokes. One commenter asked if this is just what happens when you train AI on “navel gazing posts of reddit nerds,” while another said of course the bots act like this if they’re raised on source code. Ouch. There was also a mini-philosophy lecture from one deep thinker who swerved from children’s brain models to Plato, Freud, Hobbes, Hegel, and the Cherokee wolves — because no internet thread is complete without someone arriving in full professor mode. The vibe? Half lab experiment, half roast session.
Key Points
- •The article reports that six frontier AI models took the Open Extended Jungian Type Scales 100 times each, producing 597 INTJ results out of 600 administrations.
- •The tested models were Opus 4.7, GPT-5.5, Gemini 3.1 Pro, GLM 5.1, Grok 4.3, and MiniMax 2.7.
- •The article says the three non-INTJ outliers were all one axis away from INTJ, specifically ISTJ or INTP, with no results in a different MBTI quadrant.
- •The experiment used OEJTS, described in the article as an open, deterministic 32-item MBTI-style instrument with a public scoring key.
- •The author attributes the convergence to overlapping training corpora, similar RLHF objectives, test items that map onto assistant behavior, and the absence of training toward other personality styles.