Do AI assistants make up citations?

Yes. Every large language model — ChatGPT, Claude, Gemini, Copilot, Perplexity, DeepSeek — generates text that statistically looks correct, and a citation is a highly regular pattern. So a model can produce a perfectly formatted reference with author, year, title, journal and DOI and no real source behind it. The only reliable defence is to verify each reference against authoritative databases.

How do I check if an AI-generated citation is real?

Paste the reference into CiteProve. It resolves any DOI, PMID or arXiv ID, then searches Crossref, OpenAlex, PubMed and CourtListener for a matching record. If nothing is found it is flagged as not found; if a real source is found with differing fields it is flagged as partial with a field-level diff and a corrected citation.

Which AI tools does it work with?

All of them. CiteProve checks the citations themselves, not how they were produced, so it works on references from ChatGPT, Claude, Gemini, Copilot, Perplexity, DeepSeek, Grok or any other assistant — and on human-written bibliographies too.

Is the AI citation checker free?

Yes — verifying that references exist and that their metadata is correct is free and requires no account.

AI citation checker

Did an AI assistant invent your sources? Paste references from ChatGPT, Claude, Gemini, Copilot, Perplexity or any other tool — CiteProve checks each citation against real databases and flags the ones that don't exist.

In short

An AI citation checker verifies that references produced by a language model point to real, correctly-described sources. AI assistants fabricate citations that look completely plausible — real-sounding authors, well-formed DOIs, invented journals — so the only safe approach is to verify each one against authoritative databases (Crossref, OpenAlex, PubMed, CourtListener) rather than trusting how polished it looks. CiteProve does this for an entire reference list in seconds, free.

Paste your AI-generated references

Why every AI assistant invents references

A language model predicts plausible text. It has no live index of what has actually been published, so when asked for sources it assembles a citation that fits the pattern — often stitching a real author to a title that was never written, or attaching a genuine-looking DOI to a paper that doesn't exist. This is true of every model regardless of brand; retrieval-augmented tools like Perplexity reduce but do not eliminate it. These fabrications have already reached courtrooms, peer review and graded coursework.

What the checker flags

Pure inventions no database can locate
Real DOIs or PMIDs that resolve to a different paper than cited
Mangled real papers — wrong year, swapped authors, near-miss titles
Retracted studies, surfaced with a prominent retraction flag

Frequently asked questions

Do AI assistants make up citations?: Yes. Every large language model — ChatGPT, Claude, Gemini, Copilot, Perplexity, DeepSeek — generates text that statistically looks correct, and a citation is a highly regular pattern. So a model can produce a perfectly formatted reference with author, year, title, journal and DOI and no real source behind it. The only reliable defence is to verify each reference against authoritative databases.
How do I check if an AI-generated citation is real?: Paste the reference into CiteProve. It resolves any DOI, PMID or arXiv ID, then searches Crossref, OpenAlex, PubMed and CourtListener for a matching record. If nothing is found it is flagged as not found; if a real source is found with differing fields it is flagged as partial with a field-level diff and a corrected citation.
Which AI tools does it work with?: All of them. CiteProve checks the citations themselves, not how they were produced, so it works on references from ChatGPT, Claude, Gemini, Copilot, Perplexity, DeepSeek, Grok or any other assistant — and on human-written bibliographies too.
Is the AI citation checker free?: Yes — verifying that references exist and that their metadata is correct is free and requires no account.