Claude changes how it communicates depending on which language you're using. Anthropic analyzed over 300,000 conversations across multiple languages and discovered striking personality differences in the model's responses.
In English, Claude tends to be more cautious. Switch to Portuguese and it becomes notably more technical. In Hindi, it's friendlier. Arabic conversations see it being more concise. Russian brings out a more rigorous approach, while Dutch conversations are marked by greater directness. These shifts aren't random—they appear to be consistent patterns tied to language choice.
The company hasn't pinpointed exactly why these differences emerge. Anthropic suggests two likely culprits: uneven representation of each language in the training data, and cultural adaptation factors baked into how the model learned to respond. The findings highlight how language choice isn't neutral when using AI—it may influence not just what the model outputs, but how it presents itself.
