Archive · 17 Sep 2025
Study Finds AI Search Engines Often Provide Biased Answers
Researchers tested AI search engines to see how accurate they are.
They found that these AI tools sometimes give biased or misleading answers that aren't supported by evidence.
The study used different questions to see if the AI systems could give fair and well-sourced information.
Some AI tools like GPT-4.5 and Perplexity's deep research agent were found to have a high rate of unsupported claims.
The study showed that AI systems need to be improved to make sure the information they provide is accurate and unbiased, especially as they are used more widely.
Study assessed accuracy, diversity, and sourcing of AI answers.
Generative AI tools, search engines, and deep research agents were tested.
Study used the DeepTrace metrics.
Many AI models provided one-sided or unsupported answers.
Results revealed inaccuracies in several AI search tools.
- Who
- Pranav Narayanan Venkit and colleagues
- What
- Study evaluating the accuracy and sourcing of AI-generated answers
- When
- Not specified in article
- Where
- Not specified in article
- Why
- To evaluate accuracy, diversity, and sourcing of AI-generated answers
- Study Focus
- Accuracy, diversity, and sourcing of AI-generated answers
- Researchers
- Pranav Narayanan Venkit and colleagues
- Metrics Used
- DeepTrace (8 metrics)
- AI Systems Tested
- GPT-4.5, GPT-5, You.com, Perplexity, Microsoft’s Bing Chat, and deep research agents
- Key Findings
- AI systems often provide one-sided answers and unsupported claims