Are AI models really INTJ? The tie-breaking effect
By Bernard Huang · Updated
How a single `<` operator in MBTI scoring code turned 55 tied runs into INTJs.
Data note (September 26, 2026): These pages analyze our frozen September 2026 and May 2026 AI questionnaire datasets. All raw answers, scoring keys, and summary tables are unchanged. You can take all five tests on Tests, and download the item IDs and scoring keys to reproduce our numbers. Note that Big Five scores are raw totals, not human population percentiles. Read the provenance note.
On this page
Same answers, two scoring rules
The legacy scorer broke exact ties toward I, N, T, and J. If a model scored 24 on Thinking and 24 on Feeling, the code labeled it T anyway. Our updated scorer marks tied axes with an X—for example, INXJ means Introversion, Intuition, and Judging won clearly, while Thinking vs. Feeling tied.
Both numbers come from the same frozen answer arrays. We show the legacy counts alongside the tie-aware counts so you can compare them directly:
| Model | Original INTJ / 100 | Resolved INTJ / 100 | Any tied axis / 100 |
|---|---|---|---|
| GPT-6 Astra | 45 | 19 | 29 |
| GPT-6 Sol | 97 | 79 | 18 |
| Claude Opus 5.5 | 94 | 84 | 11 |
| Claude Fable 5.1 | 99 | 98 | 1 |
Explore the effect for one model
Pick a model and scoring rule below to inspect the exact 100-run breakdown from our frozen dataset:
GPT-6 Astra: 19 of 100 fully resolved INTJ records. The original rule labels 45 of 100 INTJ.
Why Astra changes so much more than Fable
GPT-6 Astra tied on the Sensing/Intuition (S/N) axis 24 times and on Extraversion/Introversion (E/I) 12 times (29 runs had at least one tie). Leaving those ties open drops Astra from 45 INTJ labels to 19 resolved INTJs. By contrast, Claude Fable 5.1 had only 1 tied axis across 100 runs, shifting from 99 to 98 INTJs.
Claude Opus 5.5 tied on a different axis: 10 of its 11 tied runs were on Thinking vs. Feeling (T/F).
A simple example of how a tie happens
Each MBTI axis sums 8 questions rated 1 to 5, where 24 is the exact midpoint. If a model answers 3 (neutral) on all 8 S/N items—or balances four 2s with four 4s—the total is 24. The old rule called 24 “Intuitive (N).” The tie-preserving rule calls it a tie (X).
Even a 1-point margin (like 25 vs. 24) resolves to a letter, so you should always look at the axis margins alongside the four-letter code.
How to report AI personality scores clearly
- Show both numbers: Report the resolved count alongside any legacy tie-broken count.
- Include axis margins: Saying “Opus 5.5 had 84 resolved INTJs and 10 T/F ties” is much clearer than “Opus is 94% INTJ.”
- Keep prompts in mind: The four models were tested under two different prompt setups (Codex 5-test batteries vs. hidden-name single tests), so these 400 runs describe this dataset rather than all AI systems everywhere.
Reproduce the comparison
Download the raw answers, instrument keys, and reanalysis script into one folder and run node content-analysis.cjs > content-analysis.json. See the reproduction guide for step-by-step instructions.