ChatGPT fake citations: how to detect them
In short
To detect a fake ChatGPT citation, verify it against authoritative databases instead of trusting its format: resolve any DOI or PMID and confirm it points to the cited work, then search the exact title and first author in Crossref, OpenAlex or PubMed. If the identifier doesn't resolve, points elsewhere, or no database returns a matching record, treat the citation as fabricated until proven otherwise.
AI assistants produce references that look flawless — real-sounding authors, well-formed DOIs, plausible journals — but have no real source behind them. A polished format is no evidence of existence. Here is a repeatable way to check.
A five-step detection checklist
- Resolve the DOI. Paste
https://doi.org/<the-doi>into your browser. If it 404s, the DOI is invalid. If it opens a different paper than cited, the citation is wrong. - Search the exact title. Put the title in quotes in a scholarly index. No result for a precise title string is a strong fabrication signal.
- Check the author–venue pairing. Confirm those authors actually published in that journal, in that year. AI frequently attaches real authors to articles they never wrote.
- Sanity-check volume, issue and pages. A journal that didn't exist in the cited year, or an impossible page range, is a red flag.
- Watch for retractions. A real paper can still be a bad source if it was retracted after publication.
Do it for a whole list at once
Running that checklist by hand for every reference is slow. CiteProve does it automatically: paste your AI-generated reference list and it resolves identifiers, searches the databases in parallel, and flags each entry it cannot verify — with a field-level diff for the ones that are real but mis-cited.
Related: ChatGPT citation checker · Why AI invents references
Related questions
- Why are ChatGPT's fake citations so convincing?
- Because a citation is a highly regular pattern — author, year, title, journal, volume, pages, DOI — and a language model is very good at reproducing patterns. The format is perfect even when no underlying source exists.
- Can I just ask ChatGPT if a citation is real?
- No. The model has no reliable index of what has been published and will often confirm a fabricated reference as real. The only dependable check is against authoritative databases.