TL;DR of the paper, generated automatically.

A new benchmark finds current LLMs can't reliably show users diverse multilingual narratives.

• Benchmark: 1,333 news articles in 23 languages, on 23 contested topics.
• Tested: 14 LLMs across 24 query languages.
• Result: models leave gaps in competing accounts, even when diversity is requested.