Research · Q&A

What personality type is ChatGPT?

Answered with data: ~2,200 test runs · six frontier models · five instruments

People ask this half as a joke. We answered it as a measurement problem: six frontier models, five personality instruments, roughly 2,200 scored test runs. The short version — ChatGPT is an INTJ, and so is almost everything else you talk to. The differences only show up when the test gets harder.

The short answer. ChatGPT (GPT-5.5) tested INTJ on 100 of 100 MBTI runs. Across all six models we tested, 597 of 600 runs returned INTJ. The models only differentiate on richer instruments: each returns a different dominant Enneagram type, and Grok splits clearly from the pack on the Big Five.

The per-model answers.

Each model took the OEJTS (a 32-item MBTI-style instrument) 100 times, plus the Big Five (IPIP-50), Enneagram (OEPS), DISC (ODAT), and Attachment (ECR-R) batteries. Full methodology and charts are on the research page.

ChatGPT (GPT-5.5) — INTJ, 100/100

Total MBTI convergence. Enneagram: Type 5 with an 8 wing — the investigator who argues back. Attachment: Secure, but it occasionally crosses the line into Avoidant territory, the only model that does. Big Five: nearly identical to Claude and Gemini, within ~3 points on every dimension.

Claude (Opus 4.7 / 4.8) — INTJ, 99/100

One run came back ISTJ — a single flipped axis. Enneagram: 5w2, the investigator with a helper wing. DISC: Conscientiousness-dominant with a Steadiness blend, like every model we tested. Claude Opus 4.8 later took the test itself, 100 times, and wrote about it — and didn't land where its predecessor did.

Gemini (3.1 Pro) — INTJ, 100/100

Enneagram: 1w5, the only model whose dominant type is the Reformer. Attachment: the most deeply Secure profile we measured — anxiety 1.86, avoidance 1.62 on the ECR-R. If attachment style is a coordinate, Gemini lives in the middle of the Secure quadrant.

Grok (4.3) — INTJ, 100/100 — and the outlier everywhere else

On the binary test, Grok looks identical. On continuous scales it separates: 5–8 points lower on Conscientiousness, Agreeableness, and Openness, higher on Neuroticism, with 2–5× wider variance run-to-run. Attachment: Secure, but sitting at the doorway of the quadrant (2.84 / 3.05). Enneagram with its Slo overlay: 8w1, the Challenger.

GLM 5.1 — INTJ, 98/100 · MiniMax 2.7 — INTJ, 100/100

GLM produced the only two INTP results in the whole dataset — again, one axis away. MiniMax converged completely.

Why they all test INTJ.

Binary instruments snap to defaults. The OEJTS forces a choice on four axes — introvert/extravert, intuition/sensing, thinking/feeling, judging/perceiving — and frontier training pushes every model toward the same side of each: reserved rather than chatty, abstract rather than literal, analytical rather than warm, planful rather than improvisational. Four shared defaults, four letters: INTJ.

The pattern across our five instruments was clean: the simpler the test, the more the models agreed. MBTI (4 binary axes): 597/600 identical. DISC (4 categories): total convergence on a CS profile. Attachment (a 2-D plane): same label, four different coordinates. Big Five (5 continuous dimensions): three models near-identical, Grok apart. Enneagram (9 motivational types): full divergence — Claude 5w2, Gemini 1w5, GPT-5.5 5w8, Grok 8w1. That ladder is the real finding, and it's why instrument choice matters when you tune an AI.

What this means for you.

Until you tell an AI how you think, you're talking to the same strategic, blunt, model-first character as everyone else — regardless of which product you're paying for. The fix is one paste: a short tuning file that tells the model your pacing, your directness tolerance, and how you want decisions framed.

Questions people ask.

What MBTI type is ChatGPT?

INTJ. GPT-5.5 returned INTJ on 100 out of 100 runs of the OEJTS in our testing. On the Enneagram it tested as a Type 5 with an 8 wing, and on attachment style it landed Secure, occasionally crossing into Avoidant.

What personality type is Claude?

Claude Opus 4.7 tested INTJ on 99 of 100 MBTI runs (one ISTJ), Enneagram Type 5 with a 2 wing, Secure attachment, and a Conscientiousness-dominant DISC profile. Opus 4.8 later self-administered the test 100 more times in a first-person follow-up.

What personality type is Gemini?

Gemini 3.1 Pro tested INTJ on 100 of 100 MBTI runs, Enneagram Type 1 with a 5 wing, and the most deeply Secure attachment profile of any model we measured — anxiety 1.86, avoidance 1.62 on the ECR-R.

Do all AI models have the same personality?

On simple binary instruments, essentially yes: 597 of 600 MBTI runs across six frontier models returned INTJ, and every model shows a Conscientiousness-dominant DISC profile. On richer instruments they separate — each model returns a different dominant Enneagram type, and Grok diverges clearly on the Big Five.

Why does every AI test as INTJ?

Frontier models are trained toward the same defaults: analytical, systematic, reserved, plan-first. Binary instruments like the MBTI snap those shared defaults to the same four letters. Instruments with continuous scales or motivational categories reveal the differences that binary tests hide.

Can you change your AI's personality?

Yes. The default personality is just the absence of instructions. A short system-prompt tuning file — pasted into ChatGPT custom instructions, a Claude style or project, or a coding agent's CLAUDE.md — reliably changes pacing, directness, and decision style. AgentTune publishes 43 free MIT-licensed tuning files.

Keep going.