Contents
Two similar titles are a reason to look closer. They are not enough to prove that one record should disappear. A source library can contain a preprint, a published article, a revised edition and several genuinely duplicated records. I want a cleanup to preserve those distinctions instead of making the list merely shorter.
The example here is invented: a library contains two records for a paper about reading-room use. One came from a publisher page; another was imported with a PDF. I will compare them without inventing the paper's findings or assuming the imported metadata is complete.
Decide whether the works are actually the same
I would begin with authors, title, year, publication details and stable identifiers where available. A matching title alone can hide a different edition or version. A slight title difference can also occur between records for the same work. I would inspect the source itself when the metadata does not settle the question.
In the demonstration, suppose the PDF record is a preprint and the publisher record is the later version. That is not simply a spelling inconsistency. The two files may differ in content and pagination. I would retain the distinction unless I have a clear, documented reason to combine the records in a way that preserves it.
Use the duplicate view as a shortlist
Zotero's Duplicate Items feature identifies possible matches within a library. Its documentation explains reviewing and merging genuine duplicate records. I would treat that view as a shortlist requiring judgment, not as an instruction to accept every proposed match. A program can identify similarities without knowing why I saved a version.
I would also keep the scope in mind. The documented feature does not compare everything across separate group libraries. A clean-looking duplicate view is therefore not proof that an entire research setup contains no repeated material. The tool answers a narrower question, which can still be useful.
Inspect the useful material around the record
Before merging genuine duplicates, I would look at notes, attachments and collection membership. The record with a worse title might hold the annotation I need. The one with the clean citation might belong to an active project. A cleanup should preserve those useful relationships rather than rank records by appearance.
I would check whether the files themselves differ. Two attachments can have similar names while one is an annotated copy or a corrected version. If I cannot establish identity, I would pause. Storage savings are not a good reason to discard the only copy of work that has not yet been understood.
Make the comparison explicit
This is the comparison note I would keep for the fictional pair. It records what I need to establish rather than supplying made-up bibliographic answers. A brief note is particularly useful if another person will finish the review later.
The decision “not merged yet” is a legitimate outcome. It gives uncertainty a visible state. I would rather have two labeled records for a while than one deceptively tidy record that mixes a preprint's page reference with a published version's title and publication details.
Candidate pair: publisher import + PDF import Same authors and title? Check against the source. Same version or edition? Not established. Attachments: Inspect both files before removing either. Notes and memberships: Record what must be retained. Current action: Resolve version identity; do not merge yet.
Merge genuine duplicates using documented behavior
For records that really are duplicates, Zotero documents merging rather than simply deleting one. It preserves collection and tag memberships through the merge and allows choices between differing field values. I would inspect those choices against the source instead of assuming the first displayed value is authoritative.
The application behavior does not replace an appropriate backup. Before a substantial cleanup, I would follow the software's current backup guidance and work in manageable batches. A small batch gives me a chance to notice an unexpected result while the affected records are still easy to identify.
Stop when the library is clear enough for the task
After a batch, I would try the actual use: find the source, open the intended version and understand the citation. If that works, the cleanup has provided a practical benefit. A perfectly short list is not the objective, especially when keeping distinct versions supports the research.
I would write down a recurring import problem if I noticed one. Perhaps a particular source supplies incomplete metadata or a workflow imports the same record twice. Improving that step may help more than repeatedly cleaning the same confusion. The record should become easier to trust, not just easier to count.
Compare a concrete pair before deciding to merge
I would open the invented publisher record and PDF import side by side and compare identity first. The title, authors, year and publication details form a starting point, but none should be trusted simply because one record looks more polished. I would inspect the source and identifiers available for the actual work. The question is whether these records describe the same item and version, not which one has more fields filled in. Completeness and correctness are related but different properties.
Suppose the PDF is labeled as an earlier manuscript while the publisher page describes the later article. I would preserve that distinction while checking what the project needs. A quotation or page reference made against the manuscript may not transfer cleanly to the later file. The title may be nearly identical while the content or pagination changes. I would not merge away the version difference merely to remove a duplicate-looking row from the library. The record should support accurate retrieval and citation, including identifying which text was actually read.
Now consider a genuinely duplicated pair imported from two locations for the same published item. One may have a better title while the other holds useful notes or belongs to a project collection. I would inspect the full record before choosing field values in the documented merge process. The goal is to retain correct metadata and useful relationships together. Deleting the visually weaker record would be a different action and could discard context that was not apparent in the main item list.
I would finish the comparison with a one-line decision: same item and version, distinct versions, or unresolved. The last outcome is legitimate. A review state can preserve uncertainty without making the library unusable. I would give the unresolved pair a specific next check, such as establishing which version an attachment contains, instead of treating every possible duplicate as an urgent obligation. A careful library can contain an explained pair longer than a tidy library contains an unexplained mistake.
Inspect notes and attachments as separate information
The bibliographic record is not the whole object I may need to preserve. Notes can hold the reason the source was saved, and attachments can include an annotated copy or supplementary material. I would inspect those before removing anything. Similar filenames do not establish identical contents. A file called article.pdf may be a clean publisher copy in one record and a locally annotated version in another. Their relationship needs to be understood, not guessed from the name.
I would compare a manageable set rather than open every attachment indiscriminately. For a proposed merge, I need to know what each relevant file is and whether the project relies on it. If the pair contains several unfamiliar files, I would pause the merge and make a short inventory. That is a different task from correcting capitalization in a title. Recognizing the larger scope helps prevent an apparently quick cleanup from becoming an uncontrolled removal of materials that have not been identified.
Notes deserve the same attention. An early note may contain a mistaken interpretation, but it can still explain a sentence in an old draft. I would correct the interpretation where appropriate and preserve context when the change matters. A merge should not silently turn two different reading histories into an apparently single unchanged record. The amount of detail needed depends on the project, but a brief explanation can protect the relationship between a source, the version read and the argument that used it.
After a genuine duplicate merge, I would inspect the surviving record and a representative project connection. Can I find the intended file? Are the source details correct? Is the relevant note still understandable? The application may preserve certain relationships as documented, but I would still check the material that matters to the active task. Successful completion of a command is useful evidence about the operation; it is not by itself proof that the record now means what the project needs it to mean.
Distinguish a repeated import from a related work
Two records by the same authors may be different works on a shared subject. I would not merge them because a duplicate view or a quick visual scan makes them seem similar. An article, a correction, a chapter and a later expanded treatment can have overlapping wording while serving different purposes. I would inspect their actual identities and preserve a relationship if that helps retrieval. Relatedness is a reason to connect records, not automatically a reason to combine them.
A preprint and published article provide an especially useful test. The appropriate handling depends on the library's purpose and citation needs, but the distinction should remain visible. I would identify the version used in my note and follow the relevant publication or course requirements for citing it. This article does not establish one universal policy for every discipline. The practical point is to make a deliberate choice based on the actual material rather than allow an automated similarity signal to decide.
Different editions of a book raise a comparable issue. A page reference may depend on the edition, and an older note may discuss wording changed later. I would not keep only the newest-looking record without checking those dependencies. If the current project uses one edition and an archive preserves another, both can have a legitimate place. A shorter list is not necessarily a better research collection if it removes the distinctions needed to understand earlier work.
I would use a comparison note when the distinction is likely to confuse another contributor. It can say why the records remain separate and which one the current draft cites. That is often enough to prevent the next person from repeating the same investigation. The note should not invent a technical relationship the sources do not establish. It should describe the reason for the library decision in ordinary language, with a route back to the records and relevant publication details.
Make cleanup a controlled maintenance task
Before a substantial cleanup, I would follow the application's current backup guidance and decide how much I intend to review. The batch should be small enough that I can inspect the result while remembering the decisions. I would not use a number of removed records as the success target. A review can be productive when it identifies that several apparent duplicates should remain separate, or when it clarifies a recurring import problem without deleting anything.
I would choose the easiest well-understood pair first, then include one awkward case. That gives me evidence about the workflow and about my own decision criteria. If I discover that I cannot reliably distinguish versions or locate attachments, the next task is improving the review method, not accelerating it. The library's size does not create permission to lower the standard for preserving information. The maintenance plan should fit the value and complexity of the material.
A short decision log can help during a large project: which batch was reviewed, what kind of merge was performed and which cases remain open. I would avoid recording sensitive document contents unnecessarily. The log's purpose is operational continuity. If someone notices a missing relationship later, it gives them a place to start investigating. It also prevents the appearance that a whole library was verified when only a small set of possible duplicates received attention.
Finally, I would inspect the import route that produced the duplicates. Perhaps two team members saved the same source without checking the shared collection, or a workflow routinely imports a record and a separate PDF. I would address that specific cause with a small instruction or check. Preventing a repeated ambiguity can matter more than another round of cosmetic cleanup. The outcome I want is a library whose records and versions are easier to interpret, not a continually shrinking count.
Explain one retained pair to the next researcher
For an unresolved or deliberately separate pair, I would write the reason in ordinary language: these records may concern different versions, this attachment was used for the cited passage, or the publication relationship still needs checking. The explanation should identify the actual uncertainty rather than merely label the pair “not duplicate.” That makes the review useful beyond the person who performed it.
I would include the next source of clarification if one is known, such as publication details or the actual files. If the distinction is already understood, the note can simply preserve it. The goal is not to create a new recurring task for every pair. It is to prevent the next person from repeating the same guess or merging records because their titles look similar in a list.
A cleaned library should make those differences easier to retrieve and explain. I would inspect one representative citation and attachment after maintenance, then stop the batch when its decisions remain understandable. The meaningful result is not the removal count. It is a record that preserves the work, the version read and the context that the current project needs, with any unresolved boundary left visible rather than erased by a convenient-looking command.
Frequently Asked Questions
Q: Are identical titles enough to merge?
A: No. Check authors, version or edition, publication details and identifiers where available. Different works or versions can share a title, and genuine duplicates can have small metadata differences.
Q: Should I delete the record with fewer fields?
A: Not without review. It may contain useful notes, attachments or memberships. For genuine duplicates in Zotero, consult the documented merge workflow rather than simply deleting the less complete-looking record.
Q: Can I merge a preprint and a published version?
A: That requires a deliberate library decision that preserves the distinction and your citation needs. This guide does not recommend treating them as automatically identical. Inspect the files and the source details first.
Q: Does a duplicate view check every group library?
A: Zotero documents duplicate detection within a library, not across all separate group libraries. Check the current feature scope rather than treating one view as a global inventory.
Q: What should I do if I am unsure?
A: Leave the candidates separate, label the unresolved question and keep both files. You can make a later informed decision more easily than reconstruct information that was prematurely removed.
One thing to take forward
I would call the job complete when I can identify the work and the version I intend to use. A library with a few clearly explained pairs is more useful than one where a fast cleanup has erased the reason those pairs existed.
Sources & further reading
Something needs correcting? Write to the editors with the article link.



