An expert evaluation of Evrópuvefur, a government-funded AI service run by the University of Iceland, reveals a significant coverage-trust trade-off between curated retrieval and open web search. The study analyzed 449 AI-generated answers regarding EU accession, comparing the reliability of sources cited from each method.
- In 35% of web-search answers (65 of 187), at least one cited source was flagged as untrustworthy or irrelevant, whereas curated sources were rarely flagged and only for being out of date.
- Open web search provided broader coverage by answering more questions, but the curated corpus prioritized trustworthiness, declining to respond when it lacked sufficient information.
- The system failed to cite strong local sources like RÚV in any of the 287 web-search answers reviewed.
- Prompt-level steering had minimal impact, with a trusted-domain list increasing citations to listed domains from only 12% to 21%.
- Fluency and topical fit were found not to predict source trustworthiness.
The authors argue that source trustworthiness is a measurable but largely invisible dimension of information quality in public AI services, highlighting the need for transparency-oriented responses.