Check a bibliography: identify errors and fabricated sources
A systematic workflow for comparing DOIs, titles, authors and years, documenting discrepancies and following up unresolved references.
Why a well-formatted reference can still be wrong
Errors do not only come from generative AI. Typographical mistakes, incorrect years, swapped authors, confusion between editions and publication dates, or metadata combined from two real works can produce plausible references that do not identify the original publication.
A general web search may repeat the same mistake or return a similar title. Compare identifying fields with DOI registries, specialist databases, repositories and library catalogues.
Five steps for a reliable check
1. Separate title, authors, year, journal or publisher, DOI and ISBN. Comparing fields makes differences visible despite citation styles and name variants.
2. Resolve persistent identifiers first. A working DOI or ISBN is a starting point; the returned title, authors and year must also match the cited work.
3. Compare the core fields together. Punctuation, subtitles and expanded first names may vary. A different year or authorship needs an explanation, including possible online-first publication or a different edition.
4. Choose databases by publication type, subject and language. Articles need research indexes; books may need regional library catalogues; preprints and research data need repositories.
5. Record the databases searched, links to the strongest records and every material discrepancy. A report makes the evidence easier to revisit.
What “not found” means
No match does not prove fabrication. Older books, local series, institutional reports and recent publications may be absent from the selected databases. A badly transcribed title can also hide a real record.
Distinguish a confirmed reference, a plausible record with metadata differences, and an unresolved reference. The second can support a specific correction; the third needs further research or a question to the person who cited it.
When manual review remains necessary
Edited volumes, historical editions, translations and incomplete metadata can produce several plausible records. Open the originals and check edition, pages and context. A bibliographic match does not establish that the text supports your claim.
For theses and institutional reviews, retain which source supplied which metadata and when. These details let another reader assess the same finding.
Three findings, three next steps
These are illustrative workflows, not measured test results. If the DOI matches but the year differs, check online publication, print edition and the version cited before correcting it. If the title matches but the DOI points elsewhere, research the title and authors independently. If no reliable record appears, try further catalogues and spelling variants and keep the finding open.
Original sources
Check with visible evidence
Veracite compares references with twelve databases and shows matching records, discrepancies and unresolved findings.