Research agents and chatbots attach citations that look right but fail in three ways: the link is broken or never existed, the paper is invented or has the wrong authors, or the source is real but does not support the sentence it is attached to. Reports, papers and literature reviews have gone out with these errors.
Models generate references the same way they generate text, so a citation can be plausible without being retrieved. Deep research agents produce many citations per answer, and one 2026 study found they hallucinate links more often than search-augmented models.
Ask for the exact supporting sentence for every claim. Check every link in a fresh session or with a script that records whether it loads and whether the sentence is there. Resolve every DOI or arXiv ID against a metadata source such as Crossref, OpenAlex or arXiv and compare title and authors. Read the sources behind any claim you publish or act on. For literature work, ground the agent in papers you hold instead of asking it to list papers from memory.
In 636 citations from 84 short literature reviews, 55 percent of GPT-3.5 citations and 18 percent of GPT-4 citations were fabricated, and many real ones had wrong details.
Of 21 citations in a ChatGPT-4o literature review, 5 were correct, 8 had errors and 8 could not be found in PubMed.
Across 1,600 queries to eight AI search tools, more than 60 percent of answers got the source wrong. Several tools cited made-up or broken URLs or syndicated copies.
The authors found 3 to 13 percent of citation URLs were likely never real and 5 to 18 percent did not resolve. A link-checking tool let models cut non-resolving links to under 1 percent.
A government report made with help from Azure OpenAI cited research that did not exist and a made-up court quote. The firm agreed to a partial refund.
GPTZero reports 100 hallucinated citations in more than 50 accepted papers, including invented authors and arXiv IDs, each checked by a person against publication databases.