How to Check a Zotero Library for Duplicate References

Why duplicate detection is harder than it looks

Two records can represent the same work while differing in punctuation, author order, transliteration, journal abbreviations, page ranges, or DOI formatting. Conversely, two versions of a work can share a title and author but should not be merged without review.

Zotero's duplicate view is a useful starting point, but a large library still needs a consistent decision process for near-duplicates and version differences.

A safe review sequence

Group likely duplicates using DOI when it is present, then compare normalized title, first author, year, venue, and pages. Inspect preprint, conference, accepted-manuscript, and published versions separately before merging.

Keep one record as the metadata anchor, copy any unique attachments or notes, and record why a merge was made. Never use a text similarity score as the only merge criterion.

• Start with exact DOI matches, including normalized DOI values.
• Review title and first-author agreement for records without DOI.
• Treat publication versions as candidates, not automatic duplicates.
• Preserve attachments, notes, tags, and collection membership.
• Run a second check after bulk cleanup to catch newly exposed duplicates.

Common failure modes

A reference manager may import the same work from different databases with slightly different metadata. Merging too early can discard a better page range, an alternate title, or an important attachment. Failing to normalize identifiers can also leave obvious duplicates hidden.

A review-first workflow is slower than blind bulk merging, but it is much cheaper than repairing a corrupted library before a thesis or systematic review deadline.

When a separate verification pass helps

Duplicate detection answers whether records look alike. Identity verification answers whether they resolve to the intended scholarly work. Use both when a library contains incomplete imports, older references, or records generated by multiple tools.