Look at a typical student reference list. A few entries carry a DOI. Most don't, because the style guide didn't ask for one, the journal is older than DOIs, or the student copied the citation from a PDF that never printed it.
A DOI makes checking easy: resolve it, read the record, compare. Without one you're back to the question a careful supervisor would ask. Does this paper exist, and is it the paper this citation describes?
You can answer that by hand. It takes a few minutes per entry. Here is how it works, and how Folio's reference checker automates the same steps.
Search by the title, not the whole citation
The first mistake is pasting the entire reference into a search box. Author initials, journal name, volume and page numbers all add noise, and the paper you want can rank below an unrelated one.
Pull out the title instead. In a reference like this one:
Roediger, H. L., & Karpicke, J. D. (2006). Test-enhanced learning: Taking memory tests improves long-term retention. Psychological Science, 17(3), 249–255.
the title is "Test-enhanced learning: Taking memory tests improves long-term retention". Search that in Crossref or OpenAlex, both free and public. The checker does this for you: it splits the entry, drops the author list and the journal/volume/pages segment, and searches the title first. If that finds nothing, it falls back to searching the full string, which catches formats it couldn't parse.
A title match isn't enough
Here is where hand-checking goes wrong. You find a paper with the same title and tick the box.
But an invented citation can borrow a real paper's title and attach invented authors. A title match alone would pass it. So the checker holds a found record to three tests:
- Title: it has to match closely. A shared opening isn't enough. "Attention is all you need" shouldn't match a longer, different paper that starts with those words.
- Authors: if your citation names authors, at least one surname must appear on the record.
- Year: if your citation gives a year, the record must be within a year of it. The one-year slack covers online-first versus issue dates.
Only when everything your citation gives agrees does the entry come back Verified.
What each result means
Verified. A real record matched your title, and the authors and year agree. The checker also looks at that record's retraction status and tells you if the paper has been retracted or carries an expression of concern. You also get the reference formatted in APA 7 from the record, so a wrong year or a missing DOI can be fixed by copying instead of retyping.
Uncertain match. A real record exists, but your year, authors or DOI disagree with it. This is often a typo or a preprint-versus-published mismatch. Open the record and decide. The checker won't call it fine and won't call it fake.
Not found. Nothing matched confidently. The checker only says this for entries that look like journal articles (it sees volume and page numbers or a DOI). So treat it as "look at this one closely", not as proof of invention. Very recent papers, non-English titles, heavily abbreviated references and grey literature can all land here.
Could not check. This is not a pass, and it isn't a fail. It means the checker has no verdict. There are two causes. One of the indexes didn't answer in time, in which case you can run it again. Or the entry is a book or a web page, which Crossref and OpenAlex don't cover. Books get a look in Open Library and show "Book found" on a match; a book that doesn't match, and any web page, is reported as could-not-check rather than guessed at.
If your list comes back with several "Could not check" entries, don't read "nothing flagged" as "all clear". Those entries are the ones you still have to confirm yourself.
Why it asks more than one index
No single index is complete or always right. Famous papers can carry the wrong year in one database, and a database can be slow or refuse requests when it's busy. So the checker doesn't give up after one index. When OpenAlex finds nothing or a record whose year or authors disagree, it asks other indexes (arXiv for preprint-first fields, Crossref when OpenAlex didn't answer), and a confirmation there has to clear the same title, author and year bar. An index that doesn't answer is recorded as a failed lookup, never as "not found".
A short routine
- Paste the whole list into the reference checker. It takes up to 25 references per run, so do longer bibliographies in batches.
- Fix every Uncertain match from the record it shows.
- Look at every Not found by hand. Search the title yourself. If you can't find the paper anywhere, take it out.
- Confirm every Could not check yourself: books in a library catalogue, web pages by opening them.
- If you have a DOI for an entry, add it. Entries with a DOI resolve directly, and a DOI that points to a different paper gets flagged.
If the list came from an AI tool, step 3 matters most. Fabricated references are a design property of these tools, not an occasional glitch. Once the list is clean, the APA 7 guide covers formatting it.
The checker is free and needs no account. Your references are used to look up matches and are not stored unless you ask for a shareable link.