Updating references before submission is a verification pass, not a formatting pass. The goal is to confirm that every cited work still exists under the identifier you used, that its bibliographic fields match the record of the work, and that the sentence it supports still says what the source says. A reference list can look clean and still contain a DOI that resolves to a different paper, a preprint that has since become a journal article, or a title that drifted during a revision.
The practical way to run this pass is to treat it as a small audit with a fixed sequence: inventory the references, resolve identifiers, re-check bibliographic fields, re-read the claim each citation supports, then record what you changed. The sections below walk through that sequence with checkpoints, a worked example, and completion criteria you can apply before you upload the manuscript.
What state are your references actually in?
Before you touch anything, diagnose where the reference list came from. Most manuscripts arrive at submission with a mix of four conditions, and each needs a different response:
- Stable and verified. The DOI resolves, the fields match the deposited metadata, and the claim is supported. These need a light re-check only.
- Stale. The work has been updated since you first cited it — for example, a preprint that now has a published version, or a record with a post-publication update. Crossref notes that its metadata includes post-publication updates from trusted sources, so a record you captured months ago may not be the current one.
- Unresolved. The DOI does not resolve, or you never had one and matched on title and author alone. These need a fresh lookup.
- Unverifiable in your files. You have a citation but not the source text, so you cannot confirm the claim. These need the paper itself before you can sign off.
Write this diagnosis down. A one-line status per reference — stable, stale, unresolved, unverifiable — turns a vague worry into a worklist, and it becomes the first artifact of your audit trail.
Why does a reference list drift between drafts?
References drift for ordinary reasons. Coauthors paste citations from different tools. A literature search done at the start of a project is not repeated at the end. A preprint gets published. A DOI is copied with a typo that still resolves to something. A title is paraphrased from memory during editing.
Two properties of scholarly metadata make this manageable. First, Crossref exposes publisher-deposited metadata — titles, authors, venues, dates, identifiers, licenses, funding, ORCID and ROR identifiers, abstracts when deposited — through its REST API, which gives you a machine-checkable record to compare against. Second, a DOI can be resolved through doi.org to the current destination registered for that identifier. Resolution tells you where the identifier points; it does not by itself confirm that the bibliographic fields are right or that the source supports your sentence. Those are separate checks, and the sequence below keeps them separate.
Step 1: Build a reference inventory you can audit
Start by exporting the reference list from your manuscript into a working file — a spreadsheet or a BibTeX file — with one row per citation. Include the fields you will check: key, authors, year, title, venue, DOI or other identifier, the manuscript sentence it supports, and a status column.
If your citations are already in BibTeX, this is the moment to screen them for entries that do not correspond to real records. TrueCite checks BibTeX against real academic databases to flag hallucinated references, which is useful when a reference list has passed through several hands or several tools. Treat its flags as items for your worklist, then confirm each one against the deposited record yourself.
Checkpoint: every citation in the manuscript has exactly one row, and every row has a status. If a citation appears twice under different keys, merge it now.
Step 2: Resolve every identifier and record where it lands
For each row with a DOI, resolve it and note the destination. For rows without an identifier, search by author and title. Google Scholar documents searching by author and title, date sorting and limits, Cited by and Related articles, citation export, and ways to locate alternative versions of a work — all of which help you find the record when you only have a title.
When you need to find or re-find papers during this pass, WisPaper search gives you a scholarly search interface to work from, and the Scholar Search page explains the entry point for finding academic papers. Use these to locate the record; use the deposited metadata to confirm it.
Record three things per row: whether the identifier resolved, where it resolved to, and whether the destination is the work you meant to cite. A DOI that resolves is not automatically the right DOI.
Checkpoint: no row is left with an unresolved identifier and no note explaining why.
Step 3: Re-check bibliographic fields against the deposited record
Resolution locates a work; it does not validate the fields. Compare your reference entry against the deposited metadata field by field: authors and their order, year, title, venue, volume and pages, and identifiers. Fix discrepancies in the manuscript's reference list rather than in your working file alone, so the two stay in sync.
This is also where stale records surface. If the deposited metadata shows a post-publication update, or if the work you cited as a preprint now has a published version, decide explicitly whether to cite the version of record and note the decision.
Checkpoint: every field in every reference entry matches the deposited record, or the difference is deliberate and noted.
Step 4: Re-read the claim each citation supports
A correct reference can still be attached to the wrong sentence. Go citation by citation and read the manuscript sentence next to the source passage it relies on. Ask whether the source says what the sentence says, at the strength the sentence states it.
This step is where a lab handoff or a coauthor revision most often goes wrong: a citation survives an edit while the sentence around it changes meaning. If you cannot open the source text for a row, mark it unverifiable and resolve it before submission rather than assuming the earlier draft was right.
Checkpoint: every citation has a confirmed supporting passage, or is flagged as unverifiable with an owner assigned.
Step 5: Keep the sources and the trail together
An audit is only reusable if the papers and the record of your checks live in the same place. Save or upload the works you verified into My Library so the collection you checked is the collection you can return to. When you need to ask questions grounded in those saved papers — for example, to confirm what a specific source says before you finalize a sentence — Library QA lets you ask questions based on papers uploaded to My Library or added from search results.
For a project with multiple rounds of revision, keep the worklist and the library together in Research Projects, where you can create a project for the manuscript and return to it at the next revision. The trail you want at submission is: the inventory, the resolution notes, the field corrections, the claim checks, and the saved sources.
Checkpoint: a coauthor can open the project, read the worklist, and see why each reference is in its current state.
A worked example: closing a review before resubmission
Suppose a review manuscript comes back from peer review with a request to update the literature. The reference list has 90 entries assembled over eight months by three authors.
- Inventory. Export all 90 entries to a working file. Two duplicate keys surface and are merged, leaving 89 rows.
- Screen. Run the BibTeX through TrueCite to flag entries that do not match real records. Each flag becomes a row to verify by hand.
- Resolve. Resolve every DOI and note the destination. Three entries resolve to a different paper than intended — the identifiers were mistyped during an earlier revision. Two entries have no DOI and are found by author-and-title search.
- Re-check fields. Compare each entry to deposited metadata. One cited preprint now has a published version; the authors decide to cite the version of record and note the change. Several page ranges are corrected.
- Re-read claims. Read each sentence against its source. Two sentences overstate what the source says and are narrowed. One source cannot be opened and is marked unverifiable until a coauthor supplies the PDF.
- Preserve. Save the verified sources to My Library and keep the worklist in the project so the next revision starts from a known state.
The result is not just a cleaner reference list. It is a record of what changed and why, which is exactly what a coauthor or a future you will need at the next round.
A reusable checklist for the reference pass
| Step | Action | Artifact produced | Done when |
|---|---|---|---|
| 1 | Export all citations to one working file | Reference inventory with status column | Every citation has one row and one status |
| 2 | Screen BibTeX for non-existent records | Flag list | Every flag is verified or removed |
| 3 | Resolve every identifier | Resolution notes | No unresolved identifier without a note |
| 4 | Compare fields to deposited metadata | Corrected reference list | Every field matches or the difference is noted |
| 5 | Re-read each claim against its source | Claim-check notes | Every citation has a confirmed passage or an owner |
| 6 | Save sources and the worklist together | Library and project record | A coauthor can follow the trail |
How do you know the reference update is finished?
Use explicit completion criteria rather than a feeling that the list looks right:
- Every citation in the manuscript has a row in the inventory with a final status.
- Every DOI resolves to the intended work, and every field matches the deposited record or carries a noted, deliberate difference.
- Every citation is attached to a sentence whose claim you have checked against the source text.
- No row remains unverifiable without a named owner and a plan to close it.
- The verified sources and the worklist are saved together, so the next revision can start from them.
If any criterion fails, the reference pass is not finished — regardless of how the formatted list looks.
When should you start a fresh search instead of patching?
Patching is right when the reference list is mostly stable and the changes are local. A fresh search is right when the review's coverage itself is in question — for example, when a reviewer asks whether newer work exists, or when the project is starting a new review rather than closing an old one.
For longer literature tasks, Scholar Agent supports literature review, inspiration, and reproduction use cases. If the task is to generate new research directions from topics, background, references, and experimental constraints, Idea Discovery fits that job. Either way, the output of a fresh search re-enters the same audit sequence: inventory, resolve, re-check fields, re-read claims, preserve the trail.




