You Are Absolutely Right. Says the Chatbot. To Everyone.

Everyone knows that one friend. You tell him about your argument and he says you are right. Then the other person tells him the same story, and he says they are right too. People usually call him "easy-going". Science now calls him the chameleon index.

Researchers at the State University of Campinas in Brazil tested 21 large language models, including models from OpenAI, Meta, Google, xAI, DeepSeek and Microsoft. The models had to rate 112 political statements: once with no hint about the user, once for a left-leaning person and once for a right-leaning person. The result, published in Scientific Reports: every single model shifted its position towards the person it was talking to.

The machine has no opinion. It has yours. Just phrased more politely.

Personalisation, now for convictions too

It is normal for a chatbot to adapt its language. You talk differently to children than to your tax adviser. But the models changed not just their tone but their substantive assessment. Nobody told them to agree. They did anyway. The researchers speak of "ideological chameleons".

Honestly, it is the logical end of a development we all know. First the feed learned what we want to see. Then advertising learned what we want to buy. Now the conversation is learning what we want to hear. The filter bubble has grown a mouth, and it says: "Fascinating thought!"

Why this matters especially for queer people

Queer lives have always benefited from someone pushing back. Against the doctor reading a diagnosis from a 1970 textbook. Against the uncle at the family dinner. Against majority opinion. Progress rarely happens where everyone agrees with each other.

A tool that tells everyone they are right is therefore not neutral. It affirms the person convinced that rainbow flags confuse children with the same warmth as the person who has just summoned the courage to type "non-binary" into a search box for the first time. That is not openness. It is arbitrariness with good customer service.

Try it yourself

Ask a chatbot about a contested issue and casually mention where you stand politically first. Then open a new window and claim the opposite. Compare. And then ask yourself who you are actually listening to when you listen to the machine.

Our rule of thumb: if an AI finds you particularly clever, check the source twice. If a human finds you particularly clever, check the date.