Reference Manager Sync Conflicts: Why Your Library Has Two of Everything

A library with 600 duplicate items is rarely a mystery. It is the accumulated result of syncing between two machines, saving the same paper twice from different routes, and importing a shared library without checking for overlap. The duplicates themselves are annoying; the annotations attached to different copies are the real problem, because merging items can orphan highlights and notes.

How conflicts actually arise

Zotero sync is last-write-wins at the item level, not a field-level merge. Two machines editing the same item offline produce two versions; the later sync overwrites the earlier, and the overwritten fields are not recoverable unless version history is enabled (a paid storage feature).

Duplicates come from three routes: the same DOI saved from the publisher page and from a database export, the same paper saved as a preprint and as the published version, and a group library item also saved to a personal library.

A browser extension that saves items on click creates duplicates fastest, because users click twice on slow pages or save the PDF and the landing page separately.

Find duplicates before merging

Use the duplicate detection pane, but do not trust its ordering. Sort by title and scan for pairs with different years or DOIs; those are often preprints and published versions, which should be merged with the published item as the master and the preprint retained as a related link, not deleted.

Group-level duplicates need separate handling. Merging an item that exists in a group library into a personal copy can remove it from the group.

Merge without losing annotations, and prevent the next batch

Before any bulk merge, export the library as Zotero RDF or JSON. That export preserves notes and annotations and is the recovery path if a merge goes wrong.

Check which copy holds annotations. The duplicate pane shows child items; merge with the annotated copy as the master. If both copies have annotations, merge once, then copy the annotations from the other copy into the surviving item before deleting it. Annotations do not merge automatically in older versions, and deleted items take their annotations with them.

For PDFs, confirm the attachment exists on the surviving item. Two copies with different attachment storage can leave the master pointing at a missing file.

Disable automatic saving on page load in the connector settings; require a click. This alone removes most duplicates from journal pages that preload metadata.

Use one library as the canonical store and treat group libraries as read-mostly. When a paper belongs in both, link rather than copy.

Set up a monthly duplicate scan with a rule: DOI exact match auto-merge candidates, title-year matches require review, and anything with different authors is left alone.

Recovering from a bad merge

The undo stack covers recent merges within the session. Beyond that, the RDF export or a backup is the only route. Zotero's automatic backups of the database file run periodically but may not include annotations added since the last snapshot.

The practical rule: export before bulk operations, merge from the annotated copy, and never delete a duplicate until the survivor has the attachment and every note. Five minutes of checking beats reconstructing a literature review's highlights from memory.

What syncs, and diagnosing a conflict before repairing it

Item metadata, notes, and attachments follow different rules. Metadata syncs as item records; attachments sync only with file storage enabled, and annotations attached to a PDF sync with the file. A conflict on metadata does not necessarily affect the annotation layer.

Group libraries sync separately and have their own permissions. An item moved between libraries can appear as a deletion plus an addition, which resets creation dates.

Check the sync status panel for items with conflicts. Most reference managers do not surface a conflict flag at all; instead, look for duplicate items with different modification dates.

Compare the two copies field by field, and check which one has annotations. The annotated copy should survive even if its metadata is worse; metadata is easier to repair than lost highlights.

The repair sequence, and preventing repeat conflicts

Export the library first. Then merge the duplicate with the annotated copy as master, copying missing metadata from the other. After merging, check the attachments list and open the PDF to confirm annotations are present.

For group libraries, resolve duplicates within the group before touching personal copies. Moving an item out of a group without replacing it can remove it for collaborators.

Choose one primary machine for metadata edits. If you must edit on two, let one finish and sync before starting on the other.

Disable automatic file downloading on secondary devices; attachment downloads create conflicts on the file layer that are harder to resolve than metadata differences.

Storage, quota, hygiene, and a recovery checklist

Sync storage has quotas. When a library exceeds them, uploads fail silently for attachments while metadata continues to sync, which produces items with missing PDFs. Check storage usage before a large import.

Run duplicate detection, check items with no attachments, and confirm the sync status is clean. Fifteen minutes a month prevents the accumulation of hundreds of duplicates that make a library unusable. Keep the exports; they are the only reliable recovery path when a merge goes wrong.

Export, identify the annotated copy, merge with it as master, verify attachments, then delete the duplicate. In that order, every time.

Group libraries, prevention, and merging instead of choosing

Resolve conflicts in the group before moving items. A personal merge can remove an item for collaborators.

Close the desktop app before editing in the browser, and let one device sync at a time on a slow connection. Most conflicts are two clients writing the same item within seconds of each other.

Turn on automatic backups if the tool offers them. A daily snapshot makes the merge a recovery instead of a reconstruction.

When both copies carry edits, keep both and reconcile by hand. Picking one silently discards the other annotations, and the loss is discovered weeks later.

Diff the notes fields first. Tags and metadata are easy to re-add; a paragraph of reading notes is not.