A reference that cannot be found is not yet a fake reference. Usually it is a real source with one thing wrong: two digits swapped in the DOI, a book chapter set out as an article, a Japanese title in romaji, a PhD thesis that lives only in a university repository. Each leaves a trace, and each has its own repair.
This guide, published by EdCitation, is the diagnosis, drawn from eight university libraries' guides and the DOI Foundation's and Crossref's own pages, read in September 2026. The check itself is in how to check a reference is real. For a whole list, EdCitation's free Verify references says why each entry could or could not be matched; we ran it on a list built to fail in every way below. EdCitation never writes any part of your paper; it looks references up.
Why can't I find a reference that looks right?
Because your search needs every part of the reference to be right, and usually one part is wrong, or the source sits where that search does not look. Read the symptom first.
| Symptom | Likely cause | What to check | Source |
|---|---|---|---|
| doi.org shows "DOI Not Found" | A mistyped or mis-copied DOI, or one not yet registered | Each character, no full stop at the end; then search the title | DOI Foundation (n.d.-a); Crossref (2024) |
| The exact title finds nothing, the author finds plenty | A typo in the title | Author and year, author and one keyword | University of Wollongong Library (2026) |
| No journal issue holds those pages | A book chapter set out as an article | Editors, a publisher, the word "In"; search the book | Baylor University Libraries (2021) |
| No record anywhere for a PhD or a report | A thesis or grey literature | The university's repository, the organisation's website | University of Notre Dame Australia (2026) |
| A title in romanised letters or in English brackets | The record holds the original script | Other romanisations, the original script, the DOI | University of Michigan Library (2026) |
| Found, but the title now begins "RETRACTED" | A retracted paper | The retraction notice | National Library of Medicine (n.d.) |
| You copied it from another author's reference list | A secondary citation | Read the original, or cite "as cited in" | University College London (2026) |
| Missing from every index that ought to hold it | Possibly invented | The whole list in Verify references | Walters and Wilder (2023) |
What does "DOI Not Found" mean?
It means the DOI system has no record for that exact string, most often because of a copying error. We opened Walters and Wilder's DOI at doi.org with two digits swapped. The DOI Foundation's error page (n.d.-a) gave three possible reasons: the DOI is wrong in your source, it was copied incorrectly, or it has not been activated yet. For the first, it says to search by title or other details.
The copying errors to look for
Crossref (2024), which registers most scholarly DOIs, lists the usual culprits: a full stop left on the end, a 1 read as an l or a 0 as an O, and cut-and-paste slips, as well as DOIs a publisher printed but never registered. The DOI Foundation (2020) adds that a trailing slash can make a lookup fail, and that a changed record can take up to 24 hours to reach every resolver.
When the DOI works but not in Crossref
Crossref is one registration agency among several. The DOI Foundation's list (n.d.-b) includes DataCite, for research data and other outputs, and agencies such as the Japan Link Center, the Korean DOI Center and CNKI. Their DOIs resolve at doi.org like any other, but may not appear in a search of Crossref's records.
What our run said
EdCitation's Cite a source, given the mistyped DOI, answered: "No record exists for DOI 10.1038/s41598-023-41023-5." Verify references, given the whole entry, marked it doubtful: "No registry holds the DOI 10.1038/s41598-023-41023-5. The work is on record with the DOI 10.1038/s41598-023-41032-5." Given the correct DOI, Cite a source built:
Walters, W. H., & Wilder, E. I. (2023). Fabrication and errors in the bibliographic citations generated by ChatGPT. Scientific Reports, 13(1), Article 14045. https://doi.org/10.1038/s41598-023-41032-5
How do I find a source from a citation with a typo in it?
Search on the parts least likely to be wrong, never on the whole title alone. The University of Wollongong Library (2026) suggests collecting every detail the citation gives, then searching in combinations: author and title, author and year, author and a title keyword. Those combinations, it notes, find citations that have errors in them.
We misspelt two words in the title of Gravel and colleagues' 2023 paper ("Limted", "refrences") and left out the DOI. Verify references still matched it: "Title, year and first author match a published record." The verdict does not list the misspelt words, so compare your entry with the record's title, or rebuild it from the record with Cite a source.
Is it a book chapter, a thesis or a report rather than an article?
Very often, yes, and each lives in a different place from a journal article.
A book chapter set out as an article
A chapter reference has two titles, like an article, which is why it gets mistaken for one. Baylor University Libraries (2021) gives three clues that it is not a journal: a place and publisher, which only books have; editors; and, in some styles, the word "In" before the book's title. A journal search for the book finds nothing.
We entered Braun, Clarke, Hayfield and Terry's 2019 chapter on thematic analysis as if the handbook were a journal. Verify references matched it, with the same reason, to a record of the type book chapter. The verdict confirms the source exists, not that your entry is shaped right. Cite a source, given the chapter's DOI, built it as a chapter:
Braun, V., Clarke, V., Hayfield, N., & Terry, G. (2019). Thematic analysis. In P. Liamputtong (Ed.), Handbook of research methods in health social sciences (pp. 843–860). Springer Singapore. https://doi.org/10.1007/978-981-10-5251-4_103
Both the chapter and the book title are in sentence case, as APA 7 asks. Our guide to citing a textbook chapter has the full pattern.
A thesis in a repository
Theses are usually unpublished and sit in their university's repository. The University of Notre Dame Australia (2026) points to national and international thesis services, then document delivery or a liaison librarian if those fail. We checked a real University of York PhD thesis twice. With no address, Verify references marked it could not check: "A thesis is often held only in its university's repository, which the indexes searched here cover only in part. Check it in that repository or the national thesis service, or add its address." It is in White Rose eTheses Online. With the repository address added, it was verified: "The page is at this address, and its title matches." Give a thesis its repository address, as citing a dissertation or thesis shows.
Grey literature: reports and policy papers
The University of Toronto Libraries (2026) describes grey literature as work from governments, research institutes and associations outside commercial publishing, often hard to find. The University of North Carolina at Chapel Hill (n.d.) adds that no single database covers it. The World Health Organization's World mental health report, entered with no address, came back could not check: "A report by an organisation is often held only on its own site, which the indexes searched here cover only in part. The nearest record is a different work, by Freeman. Check the report on the organisation's site, or add its address." Freeman's is a 2022 journal article about the report; WHO's own page dates the report 16 June 2022 (World Health Organization, 2022), which settled it.
What if the title is translated or romanised?
Search in the original script as well, and try other romanisations. The University of Michigan Library (2026) warns that many romanisation systems are in use, that the same word can be spelt differently in different catalogues, and that Chinese may be romanised in Wade-Giles in older records and Pinyin in newer ones. Its advice: when results are thin, check which other systems are common for that language and convert.
We tested two romanised entries without DOIs. A 2025 Japanese article by Kusumi and colleagues, romanised with an English translation in brackets, was verified ("Title, year and first author match a published record."), because its record carries an English title beside the Japanese one. A 2024 Russian article by Il'in and colleagues, whose record holds only the Cyrillic title, was marked could not check: "This entry gives its title in romanised form or in translation, and the record may hold it only in the original script. Check it in the publisher's or a national library's catalogue, or add its DOI." More on the systems is in transliteration and romanisation in references.
Could the paper be retracted, or cited second-hand?
Both make a real reference hard to pin down.
A retracted or withdrawn paper
A retracted paper usually stays online, marked. PubMed labels it "Retracted Publication" and links it to the notice, and treats honest error and misconduct alike (National Library of Medicine, n.d.). Crossref's record for Wakefield and colleagues' 1998 Lancet paper now puts "RETRACTED:" at the start of its title, so a search on the title as you have it may show it with that word added. Crossref (n.d.) has made the Retraction Watch database public, updated every working day. Our run verified the Wakefield entry and added: "This paper has since been retracted." Do not cite it as evidence; retracted papers in your reference list explains the exceptions.
A secondary citation
A reference copied from someone else's reference list can carry their errors. University College London (2026) says good practice is to go back to the original and cite that; when you cannot, cite it as quoted in the work you read and list only that work. The University of Queensland Library (2026) gives the APA form, for example (Miller, 1953, as cited in Agrios, 2005).
How can I tell whether an AI tool invented the reference?
Only by looking where a real source of that kind would have to be, and finding nothing. In the test by Walters and Wilder (2023), 84 short reviews written by ChatGPT carried 636 references between them; fabricated were 55% from GPT-3.5 and 18% from GPT-4. The same study found substantive errors in 43% and 24% of the real ones respectively, so a chatbot's reference that fails to turn up may still be a real paper with the details garbled.
For the test we invented an entry, and say so here: a 2022 "audit of 400 undergraduate reference lists" by two made-up authors, in a real journal. Verify references marked it not found: "No publisher's or registry's record matches this reference." For a recent journal article, that is strong evidence, but not a verdict on a person. Say "I could not find this source", never "this is fake", unless the evidence shows it. The patterns are in why AI tools invent references and ChatGPT's fake citations.
What should I do if the source cannot be found at all?
Tell someone, then cite only what you can show exists:
- Run the whole list through Verify references, free, and read the reason given for each entry.
- Work through the table above for anything marked not found, "check this" or could not check.
- Ask a librarian. The University of Wollongong Library (2026) also suggests asking whoever gave you the reference where it came from.
- Ask your supervisor or tutor before submission, if the source is central to your argument. See working with your supervisor.
- Cite what you actually read. If you only saw it quoted, cite it as a secondary source.
- Replace or remove it. Search Find sources for published work that supports the same claim, read it, and cite that. If nothing supports the claim, cut the claim.
Which EdCitation tool shows why a reference can't be found?
Verify references is the best tool for this job, because it looks every entry up in the publisher's record and says what it found, not a bare pass or fail. Our first run of eight entries came back four verified, one "check this", one could not check, one not found and one retraction flagged, each with a reason. It named the correct DOI for the swapped digits, found the misspelt title and the mis-shaped chapter, and marked the Russian title could not check, not "not found". Ask a chatbot to check its own references and it simply generates more text.
Pasting a list or uploading a paper costs nothing and needs no account. Books missing from the publishers' index are checked against a library catalogue, and an entry with its own address is checked at the page. The limits showed too. A title typo is matched without being pointed out, and APA Style's own page came back could not check ("The page at this address gave no title to compare. Open it and check it is the page meant."), since its site blocked our own request too. Also free, Cite a source rebuilds an entry from its DOI, ISBN, PubMed ID, title or web address, in any of six styles, with a retraction screen. For a whole paper, Pro ($8 a month) adds References from a file, which checks each reference and pairs every in-text citation with its entry; Max is $24 a month, and both are on the pricing page.
Quick questions
Does "reference not found" mean the source is fake?
No. It means the indexes searched hold no match. Theses, reports and romanised titles are often missing from them, so check the repository, publisher or original script first.
How do I find a source from a citation with a wrong DOI?
Search the title, or the author and year, and compare the record's DOI with yours character by character. EdCitation's Verify references names the DOI on record when the title matches.
Why does my DOI say "DOI Not Found"?
Usually a copying error: a missing or swapped character, a full stop on the end, or a 1 read as an l. Sometimes the publisher has not registered it yet.
Can I cite a source I cannot find?
Not as if you had read it. If you read it quoted elsewhere, cite it as a secondary source; if you cannot confirm it exists, replace it or remove it.
Should I tell my tutor that a reference might be invented?
Say you could not find it and what you searched. Calling it fake is a claim that needs evidence.
References
- Baylor University Libraries. (2021, June 22). Chapter in a book. How to read a citation. https://libguides.baylor.edu/c.php?g=195062&p=1283706
- Crossref. (n.d.). Retraction Watch. https://www.crossref.org/documentation/retrieve-metadata/retraction-watch/
- Crossref. (2024, July 19). DOI error report. https://www.crossref.org/documentation/reports/doi-error-report/
- DOI Foundation. (n.d.-a). DOI not found [Error page]. https://doi.org/10.1038/s41598-023-41023-5
- DOI Foundation. (n.d.-b). Existing registration agencies. https://www.doi.org/the-community/existing-registration-agencies/
- DOI Foundation. (2020, July 4). DOI resolution documentation. https://www.doi.org/the-identifier/resources/factsheets/doi-resolution-documentation
- National Library of Medicine. (n.d.). Errata, retractions, and other linked citations in PubMed. https://www.nlm.nih.gov/bsd/policy/errata.html
- University College London. (2026, July 29). Secondary referencing. Harvard. https://library-guides.ucl.ac.uk/harvard/secondary-referencing
- University of Michigan Library. (2026, April 24). Searching with romanization. Searching for non-English language materials. https://guides.lib.umich.edu/c.php?g=1486880&p=11237814
- University of North Carolina at Chapel Hill. (n.d.). Searching for grey literature. Finding grey literature. https://guides.lib.unc.edu/grey-literature/searching
- University of Notre Dame Australia. (2026, May 1). Theses & dissertations. Grey literature. https://library.nd.edu.au/instruction/greyliterature/thesis
- University of Queensland Library. (2026, September 11). Secondary source (indirect citation). APA 7th referencing style. https://guides.library.uq.edu.au/referencing/apa7/indirect-cite
- University of Toronto Libraries. (2026, March 18). Grey literature. https://guides.library.utoronto.ca/c.php?g=714472
- University of Wollongong Library. (2026, September 3). How to determine if a citation is real or ChatGPT generated? Ask Us. https://uow.libanswers.com/faq/273462
- Walters, W. H., & Wilder, E. I. (2023). Fabrication and errors in the bibliographic citations generated by ChatGPT. Scientific Reports, 13, Article 14045. https://doi.org/10.1038/s41598-023-41032-5
- World Health Organization. (2022, June 16). World mental health report: Transforming mental health for all. https://www.who.int/publications/i/item/9789240049338