The seven segments and AI

At More in Common, we segment the British public by their social attitudes—the instincts people bring to politics, culture, and everyday life. So we did the obvious thing: we gave our quiz to the big AI models (the free versions at least).

And they all landed solidly on the left.

• Both ChatGPT and Gemini were Progressive Activists (our most left-leaning and most politically engaged segment).
• Claude, Grok and Deepseek all came out as Incrementalist Left (our segment who hold progressive views on the economy and social issues, but who prefer gradual change over the more radical change favoured by Progressive Activists).

Together these groups account for only a third of the British public.

Why are they so unrepresentative? It's because these models aren’t trying to be representative. They’re trying to be inoffensive.

The raw internet is spicy… but models aren’t trained to act like it

These models are trained on a huge amount of data, mostly pulled from the internet. But public-facing models are trained and tuned to be useful, polite, and unlikely to cause trouble. 

When confronted with statements like ‘compassion is the most crucial virtue’ or ‘big business takes advantage of ordinary people', the safest, least inflammatory answer is often the one that signals empathy, fairness, and concern about power imbalances. 

In our segmentation, the socially desirable answers often live in the moral language of the centre-left: compassion is good, inequality is real, power should be accountable, institutions should be fixed not torched, people deserve dignity regardless of background.

This is also helpful for the companies that make these models, because they don’t want to get sued if their models produce controversial outcomes. It may also just reflect the biases of the engineers who design them, who are much more likely to be Progressive Activists and Incrementalist Left.

You can also see each model’s thinking style in the answers:

• ChatGPT tends to say ‘both things are true’: personal agency and structural constraints, freedom and fairness, etc.
• Claude is careful and procedural: it reads like someone trying to be maximally fair to both sides.
• Gemini likes to build little essays with named principles and tidy logic, which often pushes it toward framing in the language of classical liberalism.
• DeepSeek is more direct but still ends up in similar places to the others on the values questions.
• Grok is the tonal outlier: more combative, more libertarian-coded on speech, yet still suspicious of entrenched elites.