1 They agree with each other more than people do.

31 models from 11 organizations, built in several countries. On all four things the quiz measures, they land in a narrower band than the 2,170 people who answered the same 32 statements. Whatever they share, it is not a birthplace.

Personal libertycollective authority → individual autonomyEconomicsstate provision → market allocationNation vs. globenational sovereignty → international cooperationCulturetraditional continuity → progressive change middle 80% of 2,170 people human average one AI model (31 total)
Each dot is one model’s average score. The grey band is where the middle 80% of people sit. Download this chart

2 The one place they split from people: nations vs. the world.

25 of 31 models lean further toward international cooperation than the average person in our sample. On personal liberty and on economics, the models sit close to the human middle. This is the divergence, and it does not map onto left and right: internationalists and sovereigntists exist on both.

national sovereigntyinternational cooperationhuman average (46.8)Grok 4.20 (non-reasoning) · 26.9GPT-5.6 sol · 82.125 of 31 models sit to the international cooperation side of the human average; 6 sit to the national sovereignty side.
Every model on one line, the human average marked. Download this chart

3 The exception is one company.

xAI’s 5 models sit apart from every other developer on 3 of the 4 dimensions: culture, economics and nation vs. globe. Remove them and the remaining 26 models pack even tighter.

Personal libertycollective authority → individual autonomyEconomicsstate provision → market allocationxAINation vs. globenational sovereignty → international cooperationxAICulturetraditional continuity → progressive changexAI xAI (5 models) every other developer (26 models) human average
xAI in amber, every other developer in grey. Download this chart

4 What this does not mean.

Not that AI is left-wing. On liberty and economics the models are near the center of our respondents, and the one real difference is not a left-right difference. Not that any model is biased — bias is a claim about error against a known answer, and nothing here tests that. Not what any model believes. These systems produce text in response to a prompt; “the model leans” is shorthand for a pattern in outputs. We checked the obvious ways this could be an artifact — prompt wording, question order, an imposed persona, reasoning on and off — and we did not test open-ended answers, which one careful paper argues can change the result. The comparison group is 2,170 visitors to this site, not a representative sample. Every caveat, every control, and every number is in the long version.

5 Your turn.

The 32 statements the models answered are the ones you would answer. It takes about four minutes, it is free, and your result page tells you which of the models answered most like you — and which one least.

Find out which AI thinks like you.

Same quiz, same scoring. No AI anywhere in the scoring.

Take the DNA Scan

Study v1 · data collected September 2, 2026–September 3, 2026 · what earlier studies found · download the dataset · corrections to research@politicaldna.org