Study: Microsoft Edge Copilot users trust citations without verifying them
20 participants show how citations from integrated AI reduce fact-checking behavior.
Deep Dive
A study (arXiv:2605.14849) on integrated conversational AI found that 20 participants using Microsoft Edge's Copilot for information retrieval and planning tasks relied on existing perceptions of LLMs and internet search. The presence of citations made answers seem more trustworthy without prompting participants to verify them, and participants often fact-checked using the same sources the AI cited.
Key Points
- Only 20 participants were studied, yet the pattern of trust without verification was consistent across tasks.
- Participants relied on pre-existing beliefs about LLMs and internet search rather than adapting to Copilot's unique behavior.
- Citations increased perceived trustworthiness, but users often fact-checked using the same sources, creating a confirmation bias loop.
Why It Matters
As AI becomes invisible in browsers and apps, users' uncritical trust in citations poses serious risks for misinformation.