Citation verification is the step where you confirm that every reference in a manuscript points to a real, correctly described work and that the cited source actually supports the sentence it is attached to. It is not a single pass through a reference list. It is a short, repeatable audit that produces a record you can hand to a coauthor, a reviewer, or your future self. This guide gives you a diagnostic framework, a set of decisions, and a way to know the audit is finished.
What state is your citation trail in right now?
Before you open a reference manager, name the situation you are actually in. The checklist you need depends on where you are entering the work.
| Your situation | What is at risk | Where to start |
|---|---|---|
| Revising a manuscript before resubmission | References added under deadline pressure; claims edited but citations left unchanged | Full audit of changed sections first |
| Closing a project or lab handoff | Knowledge that lives only in one person's memory | Build a durable trail, then audit |
| Responding to peer review | Reviewer disputes a citation or asks for support | Targeted audit of the disputed claims |
| Starting a new review | Inconsistent search and screening records | Set up the trail before collecting |
| Reproducing a published method | Unclear which source supplied which detail | Trace each method step to its source |
A useful diagnostic question: if a coauthor asked you tomorrow why a specific sentence cites a specific paper, could you answer without re-reading everything? If not, the trail is incomplete regardless of how clean the reference list looks.
The four layers of a citation audit
Treat verification as four layers, each answering a different question. Skipping a layer is the most common reason an audit feels finished when it is not.
- Existence — does the identifier resolve to a real record?
- Identity — do the bibliographic fields match the work you mean?
- Support — does the source text actually back the claim?
- Trail — can you show how you checked, and can someone else repeat it?
Each layer has its own evidence. A DOI that resolves proves existence, not identity or support. Crossref metadata helps with identity, but the source text still has to be read for support. NIH guidance on rigor and reproducibility treats transparent methods, validation, and reproducibility as core research practice, which is the same logic applied to your references: the method of checking should be visible and repeatable.
Layer 1: Does every identifier resolve?
Start with the mechanical pass. A DOI can be resolved through doi.org to the current destination registered for that identifier. Resolution tells you the identifier is live and where it currently points.
Practical steps:
- List every DOI, PMID, or other identifier in the manuscript.
- Resolve each one and note whether it lands on the expected work.
- Flag any reference with no identifier for a manual identity check.
- Record the date of resolution, since destinations can change.
Resolution is a locating step. It does not confirm that the authors, year, or title in your manuscript match the record, and it says nothing about whether the source supports your sentence. Those are the next two layers.
Layer 2: Do the bibliographic fields match the work you mean?
This is where most silent errors live: a right-sounding title attached to the wrong author list, a preprint cited as the published version, a year off by one.
Crossref exposes publisher-deposited scholarly metadata through its REST API, including titles, authors, venues, dates, identifiers, licenses, funding, ORCID and ROR identifiers, abstracts when deposited, and post-publication updates from trusted sources. That makes it a strong reference point for field-level comparison.
A field-by-field check:
| Field | What to confirm |
|---|---|
| Title | Matches the deposited record, including subtitle |
| Authors | Order and spelling match; ORCID where available |
| Venue | Journal or conference name is the one you intend |
| Date | Publication year matches the version you read |
| Identifiers | DOI or PMID points to this exact record |
| Version | Preprint vs. published version is stated correctly |
When you are locating a work and comparing versions, Google Scholar documents searching by author and title, date sorting and date limits, Cited by and Related articles, citation export, library access links, email alerts, and ways to locate alternative versions of a work. Those features help you find the version you actually read so the fields match it.
If your reference list is stored as BibTeX, run it through TrueCite, which checks BibTeX against real academic databases to flag hallucinated references. That catches entries that never existed, leaving you to handle the subtler mismatches by hand.
Layer 3: Does the source actually support the claim?
This is the layer that reference managers cannot do for you. A citation can be perfectly formatted and still be attached to a sentence the source does not support.
Work sentence by sentence on the claims that matter most: your central argument, any quantitative statement, and anything a reviewer is likely to challenge.
- Read the cited passage, not just the abstract.
- Note the page or section where the support appears.
- Distinguish what the source found from what it merely discusses.
- Check that you have not strengthened a hedged finding.
- Confirm the claim is attributed to the right source when several are cited together.
Record the location of the supporting passage next to the claim. That single habit turns a private judgment into an auditable one.
How do you keep the audit trail reproducible?
A verification pass is only as good as the record it leaves. NIH guidance frames careful assessment of prior knowledge, strengths, weaknesses, knowledge gaps, transparent methods, validation, and reproducibility as core to rigorous practice. Apply that standard to your own audit.
Your trail should let a coauthor answer: which sources were checked, against what, on what date, and with what outcome. Keep it lightweight enough to maintain:
- A status column per reference: resolved, identity-confirmed, support-confirmed.
- The date of each check.
- The location of the supporting passage for key claims.
- A short note on any version ambiguity.
For a lab handoff or project close, this record is the deliverable as much as the manuscript is. Store it where the next person will look.
Where does WisPaper fit into the workflow?
WisPaper is an AI research assistant for finding, screening, organizing, and working with academic papers, and its tools map onto specific steps in this checklist rather than replacing the audit itself.
Finding and screening sources. When you need to locate papers for a new section or a review, use WisPaper search to find academic papers through the scholarly search interface. For readers who want a public entry point first, WisPaper Scholar Search explains and opens the paper-finding workflow.
Organizing what you verify. Save or upload the papers you are checking into My Library so the audit has a stable set of sources rather than scattered PDFs.
Interrogating your own collection. Once papers are in My Library or added from search results, Library QA lets you ask questions based on those papers. This is useful for the support layer: you can query your saved set to locate which paper addresses a claim before you confirm the passage yourself.
Longer synthesis tasks. For literature review or reproduction-oriented work that spans many sources, Scholar Agent supports longer research tasks in its literature review, inspiration, or reproduction use cases.
Project-level organization. When the audit belongs to a larger effort, Research Projects gives you a place to create and organize the work.
Each of these supports a step. None of them removes the need to read the supporting passage and record where it is.
Decision table: how deep should each reference be checked?
Not every reference deserves the same effort. Allocate depth by risk.
| Reference type | Minimum check | Add support check? |
|---|---|---|
| Background citation in the introduction | Existence and identity | Only if it carries a specific claim |
| Source of a quantitative result | Existence, identity, support | Yes, always |
| Method you are reproducing | Existence, identity, support | Yes, with page-level location |
| Citation a reviewer disputed | All four layers | Yes, with a written note |
| Self-citation | Existence and identity | Yes if it supports a claim |
| Reference added late in revision | All four layers | Yes |
Late additions and disputed citations are the highest-risk entries because they bypassed your earlier checks. Audit them first.
Routes for different entry points
Revising a manuscript. Audit only the sections you changed, then run a full existence-and-identity pass over the whole list. Prioritize any reference added after the last full check.
Responding to peer review. Start with the disputed citation. Confirm identity, then read the passage and write a short note you can paste into your response. Then check whether the same source is used elsewhere with a different claim.
Closing a project or lab handoff. Build the trail before you audit, so the record survives the handoff. Store the status table with the project materials.
Starting a new review. Set up the trail first: a status column, a date column, and a place for supporting-passage locations. Collecting sources before the trail exists is what makes later verification expensive.
Reproducing a method. Trace each method step to its source and record the page or section. Where a step is unclear, note the gap rather than filling it from memory.
How do you know the audit is ready?
Use these completion signals rather than a feeling that the list looks clean:
- Every identifier resolves to the expected work.
- Every bibliographic field matches the version you read.
- Every key claim has a recorded supporting location.
- Every late addition and disputed citation has all four layers checked.
- The status table has dates and can be read by someone who was not part of the audit.
If any signal is missing, the audit is incomplete. If all are present, you can hand the manuscript and the trail to a coauthor or reviewer with confidence about what was checked and how.




