Research & PolicyFree
Reference and Citation Check for Manuscripts
Send the manuscript and the sources it cites, get every in-text citation checked against the passage that is supposed to support it.
River takes each in-text citation and finds the passage in the cited source that is supposed to support it. Every citation then gets one of five verdicts: supported, narrower than claimed, attributed onward to a third paper, not located, or pointing at the wrong item. It also checks each reference's current status for a retraction, a published correction or a version since superseded. Both registers come back with the quoted passage sitting next to your own sentence, so the edit to make is visible rather than described.
Every result for this query is a formatting checker. They validate a reference list against a style, resolve a DOI, and stop. The status check has a documented hole too. The Retraction Watch Database's own user guide states that it does not aim for comprehensiveness in expressions of concern or corrections, and that figures drawn from those categories will be skewed. A corrected paper is the one whose number your sentence quotes.
Written for the author doing a final pass before submission, the co-author who inherited a reference list, and the reviewer asked whether a paper's citations hold. It sits after a deduplicated citation library and before submission. Where a cited item is a preprint, reconcile it against the published version first. Where the claims came out of a review, an extraction table preserving the original wording is where the passages already live. More research tools sit alongside it.
Twenty-four of 68 citations needed an edit
A 7,000-word manuscript with 68 in-text citations drawn from 41 unique references. Forty-four check out: the passage is there and it says what the sentence claims. Eleven support a narrower version of the claim, usually because the sentence dropped a population or a condition. Six are attributed onward, meaning the cited paper is itself citing someone else for the point. Four cannot be located in the cited source at all. Three point at the wrong item by the right author, which is the classic same-surname slip.
Those five verdicts sum to 68, and 24 of them need an edit before submission, which is 35.3 per cent. Seven are hard errors that a referee would treat as carelessness: the four that are not in the source and the three that name the wrong paper. The eleven narrowed claims are the more interesting group, because each one is a sentence that was true when it was drafted and got broader during revision.
The status pass runs separately over the 41 references. One has been retracted. Three carry published corrections. Two are preprints that now have a journal version reporting different numbers. Six of 41, 14.6 per cent, changed status after they were cited, and none of it shows up in the manuscript. A reference manager will validate all 41 formats and check none of the 68 claims, which is the division of labour this run exists to correct.
How it works
Send both halves
The manuscript, and the sources it cites as PDFs or identifiers a reader can resolve.
Locate each passage
For every in-text citation, the sentences in the cited source that bear on the claim.
Return a verdict
Supported, narrowed, attributed onward, not located, or the wrong item, with the passage attached.
Check the status
Every reference tested for a retraction, a correction or a version that has moved on.
What you get
- One verdict per in-text citation, with the supporting passage quoted next to your own sentence
- Claims the source supports only in a narrower form, separated from claims it does not support
- Citations the cited paper attributes onward, so you can cite the study rather than the review
- A status pass over every reference for retractions, corrections and superseded versions
- The missing references, and the in-text citations with no entry in the list
- Counts by verdict, so you know how much work the revision actually is
Common questions
How is this different from what my reference manager does?
A reference manager works on the reference list. It resolves identifiers, applies a style and finds duplicates, and it never opens the cited paper to see whether your sentence is true. Both jobs are necessary. This one reads the passage and returns a verdict on the claim, which is the error that costs a paper its credibility rather than its tidiness.
What does attributed onward mean?
That the paper you cited is itself citing someone else for the point. It happens legitimately in reviews and it becomes a problem when the original source turns out to say something narrower, or to be twenty years old, or not to exist. The register gives you the upstream reference so you can cite the study instead of the summary.
Can it check a citation behind a paywall?
Only where you supply the file. The run will not assert a verdict from an abstract, a database record or a different one of the seven named article versions, and it marks those citations as unchecked with the reason rather than guessing. An unchecked count next to a checked count is honest. A verdict inferred from a title is not.
Why do corrections matter as much as retractions?
Because a correction usually changes a number rather than withdrawing a paper, and a number is what your sentence quotes. Retraction coverage is good and correction coverage is not: the largest retraction database says so about itself. That asymmetry means a corrected reference is the likeliest thing in your list to be quietly wrong, and the hardest to find by searching.
Will it rewrite the sentences it flags?
No. It gives you the passage, the verdict and the narrowest claim the source will actually carry, and you write the sentence. Where a citation cannot be saved, it says whether the claim needs a different source or needs cutting. The byline is yours and so is the wording.
What do I get back?
A Sheet with one row per in-text citation: the sentence, the source, the located passage with its page, the verdict and the fix. A second tab holding the reference status pass. A Doc listing the citations that do not support their claim, the missing references and the counts by verdict.
Reference and Citation Check for Manuscripts
Fill in the form and your workspace opens with the work already underway.