Reference Checker
Check references against Crossref, OpenAlex and PubMed for metadata mismatches and recorded retractions. First 20 checks free; review the evidence.
Your reference list
Paper extraction runs in your browser. Reference text and identifiers are sent to metadata providers through our server. Drafts, reviewed corrections and results are saved for this page in this tab; the allowance count is saved in your browser.
Write with traceable sources in AnswerThis
Find papers and write with linked sources. Review the evidence behind each claim. Opens in a new tab; copy your results first. Your paper and references do not transfer automatically.
Prepare a list that can be checked
Paste one complete reference per line, keeping the title, creators, year and DOI together. A DOI gives the strongest publication identity; a title-only entry can be harder to distinguish from another work. The checker compares recognizable bibliographic fields with publication records. It does not read the cited paper to determine whether its methods are sound or whether it supports the sentence in which you cited it.
You can also upload a text-based PDF, TXT or Markdown paper, up to 5 MB and 100 pages. The paper needs a detectable References or Bibliography heading. Review the extracted list before starting lookups, since headers, wrapped lines or unusual layouts can affect extraction. Scanned PDFs need OCR and are not supported here. A reviewed list is the input to the check; uploading a paper does not verify all its claims.
Worked example: review a year mismatch
Consider the supplied line “Harris et al. (2019). Array programming with NumPy. Nature. https://doi.org/10.1038/s41586-020-2649-2”. In the documented example record, that DOI identifies the NumPy paper published in 2020. A comparison can therefore flag “Year: 2019 → 2020” as details that differ. This is an illustrative metadata comparison, not a fresh provider query. Open the publication and confirm the year before editing the reference.
With the same identity and a supplied year of 2020, Found means the record matches the fields that were checked. It does not prove every author initial, page range or journal abbreviation was verified. If the list instead contains an unresolvable title, Not found means no confident database match was obtained. Preserve the original line and investigate the publisher or catalog rather than deleting it as a supposed fabrication.
Interpret identity, differences and availability separately
Details differ—shown as “Details do not match” in the checker—means recognizable supplied fields disagree with the matched metadata. Abbreviated journals, complex surnames and unusual reference layouts can also limit reliable comparison. Review what was actually compared, not only the verdict label. Ambiguous match means more than one plausible publication identity remains; provide a DOI rather than applying a correction from whichever title result happened to appear first.
Retracted means a retraction is recorded for the matched work. Follow the notice to understand what it concerns; a correction or the retraction notice itself is not automatically the retracted original article. Check unavailable means a provider failed or evidence could not be obtained. Retry later and retain the uncertainty. These states should not be collapsed into either “fake” or “safe,” because publication identity, record agreement and scientific reliability are different questions.
Providers, coverage and your review record
Bibliographic evidence comes from Crossref, OpenAlex and PubMed. Retraction evidence is inspected on Crossref records, including Retraction Watch and publisher updates when present. Read the checked-source states on each result because providers can return different coverage or be unavailable. No AI determines whether a reference is real. The tool uses metadata lookup and comparisons, and it cannot establish whether a reference was written by an AI system.
Upload extraction happens in the browser; reference text and identifiers are sent through the server for metadata checks. Review any result or proposed change against the source before using it in a manuscript. Keep the original reference and evidence together when making corrections, especially when the provider record omits details present in your citation. The first 20 completed reference checks are free per browser, shared with Retraction Checker.
Read each result with its evidence
- Found: a publication record matches the checked fields; this is not verification of the paper’s claims.
- Not found: no confident publication match. This does not prove fabrication.
- Retracted: a recorded retraction for the matched work. Read the linked notice.
- Details differ / Details do not match: review recognizable supplied fields against the matched record.
- Ambiguous match: more than one plausible work remains. Supply the original DOI.
- Check unavailable: a provider failed or evidence is unavailable. Retry without treating this as a negative finding.
Frequently asked questions
What can I paste or upload?
Paste complete references, DOIs or publication titles. Reference Checker also accepts text-based PDF, TXT and Markdown papers for reference-section extraction, within the stated limits. Review the extracted list before checking; scanned documents and papers without a detectable bibliography may require you to paste the references yourself.
Does Not found mean a reference was fabricated?
No. It means the checked databases did not supply a confident match. Coverage, metadata quality and parsing vary. Search the publisher, library catalog or original source before drawing a conclusion. This tool cannot determine whether a person or AI wrote the reference.
What should I do with Details differ?
Inspect the particular field and matched publication identity. The year example changes 2019 to 2020 only after the DOI identifies the intended NumPy paper. Compare the original publication before applying a correction, and retain valid volume, page and author details even when a provider summary omits them.
Why does an ambiguous title need a DOI?
Several records can share a title or look very similar. A confident identity is necessary before comparing fields or attributing a retraction. Supply the DOI of the publication you used instead of accepting a guessed correction or treating an ambiguous result as confirmation.
Which services supply retraction evidence?
The checker inspects Crossref metadata for publisher and Retraction Watch retraction updates and links to a notice when recorded. Crossref, OpenAlex and PubMed provide bibliographic evidence. Read each source state; an unavailable check or absent update is not proof that the work is reliable.
Does this validate the claims in my paper?
No. A Found result concerns publication metadata, not whether a source supports your claim or whether its findings are valid. Read the work and any relevant notices. The reporting checklist and literature review tools address other research tasks, but their outputs also require evidence review.
What is free, and where does my input go?
The first 20 completed checks are free without login, shared with Retraction Checker in this browser. Upload extraction is local, while submitted reference text and identifiers go through the server to metadata providers. Download or copy results before continuing; opening AnswerThis does not transfer your paper or reference list.