Primary source discovery is among the most demanding forms of reference work. This paper investigates whether generative artificial intelligence (AI) tools can find reliable and relevant primary sources, how effectively they summarize them and whether they can provide ethical access and enable discovery, to establish what reference librarians can responsibly recommend.
The authors tested four generative AI tools at free, institutionally provided and paid access levels, querying each with 20 prompts based on undergraduate history topics. A grading rubric framed by the Guidelines for Primary Source Literacy was used to assess primary source status, availability, summary, topic appropriateness and citation.
No tool hallucinated; all sources recommended by the tested generative AI tools existed. Overall, the tools performed well on topic appropriateness and adequately in citations, but some struggled to provide accessible sources and summaries.
Because models evolve continuously and prior interactions may bias output, the study is not strictly reproducible, though the rubric and prompt set can be reused.
The findings support a reallocation of professional attention rather than a displacement of it. Librarians can recommend generative AI as a discovery venue alongside traditional search methods, and redirect reference and instruction time to what the tools neglect.
This is one of the first studies to evaluate generative AI tools for primary source discovery, and it separates discovery from the analytical skills that follow it.
