A free resource · Yours either way
These are the questions we ask of every reference list, in a form you can run against your own manuscript. No tool required beyond a browser. Set aside an unhurried block of time, open your reference list, and work through it once before you submit.
Whether or not we ever work together, a reference list that has been through this is stronger than one that has not.
Check at the registry, never from memory. Yours or an AI’s.
Resolve every DOI. Paste each one into doi.org. Does it load, and is it the paper you meant to cite?
Check every PubMed ID at PubMed. Title, authors, journal, year, pages: does every field match your reference entry?
Watch for the near miss. The most common defect is not a fake paper; it is a real paper with a wrong identifier. Correct title and authors, wrong digits, pointing at a different work or at nothing.
Treat “could not resolve” as unverified, not as fine. A reference you cannot confirm is a question, and a reviewer may ask it.
Existence is the floor. Support is the point.
Reread the source behind every load-bearing claim. The abstract at minimum. Does it support your sentence as written, not as remembered?
Check the scope. Same population, same outcome, same direction? A study that measured a related construct is adjacent evidence, not support.
Find the reference doing the work of two papers. Are you citing one paper for findings that actually belong to another, often by the same group? Split it and re-point each citation.
Check for hedging drift. If your abstract hedges and your conclusion asserts, harmonize toward the hedge your limitations section already commits to.
The small things a tired reviewer notices first.
Verify the author list on each reference. Every name correct, and actually on that paper.
Include the references inside tables and figures. Not just the bibliography. Partial checks report as complete and are not.
Confirm every referenced exhibit is present. A table cited with a full caption but missing from the draft leaves a reviewer unable to evaluate the claim it carries.
Check the journal’s live author instructions. Abstract format, reference style, length limits: against the journal’s own current page, not a template you saved years ago.
The rule that makes it work
Do not let the tool that wrote a citation be the tool that checks it. A general AI assistant cannot catch its own fabricated references. If AI touched your reference list, verify it somewhere the AI cannot: at the registry itself.
This checklist is the first pass of the same discipline our review runs at depth, across every reference and every claim, with an independent check on each finding and a person releasing the record by hand. If you would rather have that done for you, request a review. And if you run this list and your manuscript comes through clean, that is the best outcome there is.