Citation Hallucination Pattern Library

Citation Hallucination Patterns

A citation hallucination is a reference that looks academically plausible but fails verification. It may contain a fabricated DOI, a real DOI paired with the wrong title, real authors attached to a paper they did not write, or a real source that does not support the claim. The safest review checks identifiers, metadata, Scholar visibility, and claim support together.

What citation hallucination means

A hallucinated citation is not just a formatting problem. It is a broken evidence claim. The reference may look complete in APA, MLA, Chicago, IEEE, Vancouver, or another style while the underlying academic record is missing, mismatched, or irrelevant.

This is why citation review should move beyond style checking. The important question is whether the cited work exists, whether the metadata fields agree, and whether the source actually supports the surrounding sentence or paragraph.

Common hallucination patterns

AI-generated references often fail in repeatable ways. A useful review process names the pattern, then checks the evidence needed to confirm or reject the citation.

Pattern
What it looks like
What to check
Fabricated DOI
The citation includes a DOI-shaped identifier, but the DOI does not resolve or points nowhere.
Resolve the DOI first, then compare the resolved title, authors, year, and venue.
Chimera citation
Real-looking authors, journals, years, or topics are combined into a paper that does not exist.
Search the exact title and compare the full author-paper pairing against structured records.
Wrong author-paper pairing
The authors are real researchers, but they are attached to a title they did not publish.
Check author lists in CrossRef, PubMed, OpenAlex, publisher pages, and Scholar visibility results.
Real DOI, wrong title
The DOI resolves, but the title in the reference belongs to a different paper.
Treat DOI resolution as the start, then compare title similarity and metadata agreement.
Real journal, fake article
The journal exists and the topic sounds plausible, but the specific article cannot be found.
Verify journal issue, year, volume, pages, title, and publisher record together.
Wrong publication year
The citation points to a real work, but the year reflects a preprint, online-first date, or invented date.
Compare preprint, online publication, issue publication, and database metadata dates.
Unsupported claim citation
The source exists, but it does not support the sentence, claim, or paragraph where it is cited.
Read the source and assess claim support, not just bibliographic existence.
Bibliography padding
A reference list includes plausible but irrelevant, duplicate, incomplete, or weakly related sources.
Check source relevance, duplicates, missing fields, and whether each citation is used in context.

How Citely reviews these patterns

Citely treats each citation as an evidence bundle. One real field is not enough. A reference becomes more trustworthy when the identifier, metadata, Scholar visibility, and source context agree.

Identifier resolution

DOI, PMID, arXiv ID, and publisher URLs are checked first because they can point to canonical records.

Metadata agreement

Title, authors, year, venue, volume, issue, pages, and publisher fields are compared across records.

Scholar visibility

Google Scholar exact-title and similar-title results are useful evidence, but not final proof.

Pattern consistency

Citely looks for known hallucination shapes such as DOI-title mismatch or real journal fake article.

Human source judgment

Researchers still decide whether a real source supports the claim and meets disciplinary standards.

Manual checks that catch most fake citations

1

Resolve every DOI, PMID, arXiv ID, or publisher URL.

2

Search the exact title in structured databases and Google Scholar.

3

Compare title, authors, year, journal, volume, issue, and pages field by field.

4

Look for exact-title Scholar visibility and similar-title variants.

5

Read the source when the citation is being used to support a specific claim.

6

Flag references that only look plausible but do not align across evidence layers.

Example failure

A reference may cite a real journal and a DOI-shaped string, but the DOI does not resolve and the exact title has no matching structured record. A similar Scholar result may exist, but with different authors and a different year.

That case should not be marked as verified. It is better treated as suspicious until the title, author list, year, venue, and identifier can be reconciled against academic records.

Why Google Scholar is useful but not decisive

Google Scholar can reveal whether a title, PDF, repository copy, or citation trace appears in the academic search ecosystem. This is especially helpful for papers without DOI records or records that are missing from structured metadata sources.

But Scholar results can be duplicated, incomplete, versioned, or weakly matched. Citely therefore uses Scholar visibility as evidence, then compares it with DOI resolution, title-author agreement, year, venue, and structured academic records.

FAQ

What is a citation hallucination?

A citation hallucination is a reference that looks academically plausible but contains invented, mismatched, incomplete, or unsupported bibliographic information. It may include a fake DOI, a real DOI paired with the wrong title, real authors attached to a paper they did not write, or a source that exists but does not support the surrounding claim.

Why do AI tools hallucinate citations?

Large language models generate likely text patterns rather than reliably resolving academic records. Because citations have recognizable formats, an AI tool can produce titles, authors, years, journals, and DOI-shaped strings that look credible even when the specific paper does not exist.

Can a citation be fake even if the journal is real?

Yes. A hallucinated citation can use a real journal name, real authors, or a real research topic while inventing the article itself. Verification should check the full combination of title, authors, year, venue, and identifier, not just whether one field is real.

Is a Google Scholar result enough to verify a citation?

No. Google Scholar visibility is useful evidence, especially for discovery, PDFs, repositories, and citation traces. It should be compared with DOI resolution, structured metadata, title-author agreement, publication year, and venue information before treating a citation as reliable.

What is the safest way to check an AI-generated reference list?

Resolve identifiers first, compare the resolved metadata against the citation text, search exact titles when no DOI exists, review Scholar visibility as supporting evidence, and read sources that are used to support important claims. Citely automates much of this evidence-gathering workflow.

Check references before hallucinated citations spread

Paste a bibliography or AI-generated reference list into Citely to identify fabricated, mismatched, incomplete, and suspicious citations before submission.