Generative artificial intelligence (AI) systems like ChatGPT, Gemini, and Copilot are increasingly generating false information and fabricated sources, creating a major challenge for librarians and institutions responsible for providing accurate data. The core problem is that these AI tools always provide an answer, even when no real information exists—they simply invent details that appear plausible.
The Rising Tide of AI-Generated Falsehoods
According to Sarah Falls, a research engagement librarian at the Library of Virginia, approximately 15% of reference questions her staff receives are now written by AI. These queries often include entirely made-up citations and sources, forcing librarians to spend extra time verifying (or debunking) the claims. The issue is not merely annoying; it represents a fundamental flaw in the technology’s current approach to knowledge retrieval.
The International Committee of the Red Cross (ICRC) has publicly warned about this problem, stating that AI tools cannot admit when historical sources are incomplete; instead, they invent details.
The ICRC now advises users to consult their official catalogues and scholarly archives directly, rather than relying on AI-generated lists. This highlights a broader concern: until AI becomes more reliable, the burden of fact-checking will fall squarely on human archivists.
Why This Matters
This trend is significant for several reasons. First, it underscores the limitations of current generative AI models. These systems are designed to produce content, not necessarily verify it. Second, it places undue strain on already stretched library resources. As Falls notes, institutions may soon need to limit the time spent verifying AI-generated information due to sheer volume.
Finally, this situation reinforces the enduring value of human expertise. Unlike AI, librarians are trained to think critically, conduct thorough searches, and—crucially—admit when they don’t know an answer. This is a core principle of reliable information management that AI currently lacks.
The overreliance on AI-generated content without critical evaluation will continue to burden librarians and other information professionals. The solution isn’t to abandon AI entirely, but to understand its limitations and prioritize human verification until the technology improves.
