AI BriefingFudan University Study Finds Qwen AI Search Engine Avoids Same-Company Bias and Fake Concepts
·2026.10.24 01:09
Key point
Qwen attached citations to 98.3% of answers and was the only engine to ignore a planted fake concept, outperforming ChatGPT and DeepSeek in source reliability.
Details
- Fudan University study accepted at NDSS 2027 found Qwen's top cited source was Baidu (10.1%), avoiding the same-company favoritism seen in Doubao (25.9% to ByteDance) and Wenxin (24.4% to Baidu).
- Qwen attached citations to 98.3% of answers, matching ChatGPT and exceeding DeepSeek (86.7%) and Google's AI (80.8%).
- In a test where researchers planted posts about a made-up concept, 8 of 10 engines cited it within a week, but Qwen never cited it.
- Qwen cited 96% of sites with Chinese titles, the highest rate among tested engines, and was the only engine alongside ChatGPT to ignore up to 20 planted posts on Sohu.
- A secondary study by Aidso noted Qwen's website provides about 20 links per answer compared to about 11 on the app, making the web version better for source density.
- Data is from spring 2026, and Qwen did pick up some planted opinion and product posts in the experiments.
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.
Related
Research demonstrates backdoor injection in open LLMs using under $50 in computeApple Releases LensVLM-9B for Processing Compressed Text Images
Challenges in On-Policy RL for Agentic Workflows: Harness Consistency and Tree Training