How to Verify That a Citation Is Real and Correct
A reference can look perfect and still be wrong or invented. This guide gives a repeatable check, a list of red flags for fabricated or AI-generated references, and steps to take when nothing is found.
On this page
In brief
- Verify against the source record, not against a search engine snippet alone.
- A DOI should resolve to the paper that the reference describes, not just to something.
- A missing result is a reason to investigate, not proof of fabrication.
- Text generated by AI tools can contain plausible but non-existent references, so check every one.
Why verification matters
Readers trust a reference list. A wrong year, a swapped author or a non-existent paper weakens your argument and can mislead others who copy your list. Errors also spread: metadata copied from one search tool into many papers keeps the original mistake alive.
Verification has two goals. First, confirm the work exists. Second, confirm that every detail you cite matches it. The diagram above shows the flow; the steps below explain each stage.
Verification takes minutes per reference once you have a routine. It is quickest when you do it as you write rather than at the end. Do it again before submission, because links break and records are corrected. If you review other people's work, apply the same steps to a sample of its references, and to every reference that carries a surprising claim.
The verification steps
- Search the exact title in quotation marks in Google Scholar and a second tool such as OpenAlex. Google Scholar documents quoted-title search.
- Open the source record. Go to the publisher page, repository page or proceedings volume rather than relying on a summary.
- Resolve the DOI by entering it after https://doi.org/. Check that the landing page shows the same title and authors.
- Compare registry metadata (see the next section).
- Match authors, venue and year against the paper's own title page.
- Check pages or article number against the volume's table of contents.
- Confirm the version you cite: published paper or preprint.
Each step can fail for innocent reasons. Record what you found at each step so you can see where the problem begins.
Checking Crossref and DBLP metadata
Crossref is a registry of DOIs and the descriptive metadata that publishers deposit. Its public REST API returns JSON for one DOI at the address https://api.crossref.org/works/ followed by the DOI, and needs no sign-up. Compare its title, author list, container title and year with the reference.
For computer science, look the paper up in DBLP. Open the venue and volume page to confirm that the paper appears in that proceedings volume. Remember that DBLP is a bibliography, so a missing record does not by itself mean the paper is false.
Differences to expect:
- Author names may be spelled or abbreviated differently.
- The registry year may be the online publication year, while the volume prints another.
- Titles may differ in capitalisation or punctuation. Treat different wording as a warning.
DOIs also exist outside Crossref. If a DOI does not appear there, other registration agencies exist, and the identifiers guide explains the difference.
Red flags of fabricated or AI-generated references
Generative AI tools can produce references that look right but name papers that do not exist. They may combine a real author, a real venue and an invented title. None of the signs below proves fabrication alone. Several together deserve a full check.
- No trace of the title. An exact-title search in several tools returns nothing relevant.
- DOI that fails or points elsewhere. It does not resolve, or it lands on a different paper.
- Real authors, wrong topic. The author exists, but their publication list has no such paper.
- Venue mismatch. The proceedings volume exists, but its table of contents lacks the paper.
- Impossible details. The year precedes the venue's first edition, or the pages exceed the volume's length.
- Generic or too-fitting title. The title restates your claim almost word for word.
- Uniform formatting with no identifiers. Many entries are polished but none has a verifiable URL or DOI.
- Mixed-up details. Elements from two real papers are blended into one entry.
Note that sloppy but genuine references share some of these signs. A typo in a title is more common than fabrication, so verify before you accuse anyone.
What to do when nothing is found
- Loosen the search. Drop the subtitle, try a key phrase, search by author plus year, or translate a non-English title.
- Check the venue. Open the volume or series page and read the table of contents.
- Look at the author's pages. Check their institutional page, ORCID record or DBLP profile.
- Ask the source. If someone else gave you the reference, ask for the PDF or link. If it is your own draft that came from an AI tool, remove the entry until you find the real paper.
- Ask a librarian. Library staff can search databases you cannot reach.
- Replace, do not guess. Never fill gaps with invented details. If you cannot verify a source, do not cite it.
If you suspect fabrication in someone else's work, report it to the editor or the venue's organisers with your evidence, and follow the venue's integrity process, such as the guidance of the Committee on Publication Ethics (COPE).
Habits that prevent errors
- Take metadata from the paper or the volume page, not from a search snippet.
- Keep the DOI or stable URL in your reference manager and re-test it before submission.
- Run through the 12-point citation checklist at the end.
- Avoid asking AI tools for references and pasting them in unchecked.
The ten most frequent slips are listed in common citation mistakes.
Keep a short log for any reference that needed more than one attempt: what you searched, where you looked and what you found. The log helps when a co-author, supervisor or editor asks how you confirmed a source. It also stops you from repeating the same failed search a week later.
Frequently asked questions
Can AI tools invent references?
Yes. Language models can produce plausible-looking references to papers that do not exist. Verify every reference they give by finding the actual paper.
What if the DOI resolves but the title is different?
Treat the reference as wrong until proven otherwise. The DOI may have been mistyped, or the reference may mix details from different papers. Use the metadata of the paper the DOI really points to.
Is a paper fake if it is not in Google Scholar?
Not necessarily. Coverage is incomplete and new records can lag. Check the publisher or volume page, DBLP and other tools before deciding.
How do I check a DOI quickly?
Open https://doi.org/ followed by the DOI. Confirm the landing page shows the same title, authors and venue as the reference.
What should I do with a reference I cannot verify?
Remove it or replace it with a source you can read. Do not cite work you cannot confirm exists.
Sources and further reading
- Crossref — Crossref REST API documentation (2024)
- Google — Google Scholar Help (2024)
- International DOI Foundation — DOI Foundation (2024)
- dblp team, Schloss Dagstuhl — dblp: scope of dblp (FAQ) (2024)
- OurResearch / OpenAlex — OpenAlex Help Center (2024)
- COPE — Committee on Publication Ethics (COPE) (2024)
Related topics
- How to Find Workshop and Conference Proceedings PapersPractical search strategy for workshop and conference papers: Google Scholar, DBLP, OpenAlex, publisher libraries, repositories and citation chaining.
- Proceedings Citation Checklist: 12 Checks Before You SubmitTwelve checks to run on a reference to a workshop or conference paper before you submit: authors, year, title, volume, editors, pages, DOI, URL and style.
- DOI, ISSN, ISBN, URN and ORCID: Identifiers for ProceedingsWhat a DOI, ISSN, ISBN, URN and ORCID iD each identify, how to check them, how Crossref and DataCite differ, and what to do when a paper has no DOI.