The Citation Integrity Problem in AI Assisted Writing

Hallucinated citations, meaning references that look plausible and do not exist, are the single biggest ethical failure in AI assisted academic writing. They are also the most preventable.

The failure mode is predictable. A model that does not have a ground truth citation will construct one that looks plausible, with a real sounding journal, a real sounding year, and a realistic DOI. Left in the final draft, this lands you somewhere between unintentional fabrication and professional misconduct.

The workflow fix takes five minutes per reference. First, after the AI draft, extract every citation into a list. Second, resolve each one via DOI, Semantic Scholar, or your library search. Third, reject and regenerate any citation that does not resolve. Fourth, replace AI suggested references with citations you have actually read.

The Acadly AI Research Writer uses citation placeholders by design. The drafts deliberately avoid fabricating specific references. You supply the real citations. That trade off feels slower for the first draft and saves you from a retraction conversation later.

The ethical bar is simple. Every citation in your final paper has to resolve to a paper you have read. Not skimmed. Read. Not 'the AI said it existed.' Verified.