A hallucinated citation checker exists for one specific failure in AI-generated reference lists: entries that look exactly right and point at nothing. Real-sounding journal, plausible authors, a DOI in the correct format, and no such paper. This is a procedure for finding those before a supervisor, editor or reviewer does.
The useful part is that it is checkable rather than a judgement. An identifier either resolves against a live database or it does not. This page is about reading those results correctly, because the four verdicts you get back do not carry equal weight and treating them as if they do is the main way people misuse this.
Paste prose containing the references, not the bibliography on its own.
1. Paste the passage, with its citations in place. Up to 200 words a check on the free tier, thirty words minimum. Including the sentences matters: a reference that exists but is attached to the wrong claim is only visible when the claim is there to read.
2. Read the identifiers first. Five kinds are resolved against a live source: a DOI against CrossRef, an arXiv ID against arXiv, a PubMed ID against PubMed, an ISBN against Open Library, and any URL by fetching it. These are the findings that mean something definite.
3. Then read the prose flags. Titles, journal names and author names with no identifier attached are assessed by a model rather than looked up. Useful, and a different class of evidence.
4. Replace, do not repair. If an entry is fabricated, find the real source for the claim and cite that. Editing a fabricated reference until it looks plausible is how a fake citation survives into a final draft.
We gave it a sentence citing a study with a DOI of 10.1234/jwb.2019.887. The result came back fabricated, with the explanation that CrossRef has no record of that DOI, and the finding was tagged as coming from the citation verifier rather than from the model. That is the shape of a result you can act on immediately.
The single most important thing to get right, because one word covers two very different findings.
Verifiable. A live database confirmed the record. Strong, and the result names which database answered.
Fabricated. This arrives by two routes under one label. Either a lookup came back empty, which is close to conclusive, or a model judged the entry invented, which is a prompt to check. Look at whether the finding is tagged as a verifier result or a model assessment before you delete a reference.
Unverified. Deliberately not the same as fabricated. An older monograph missing from Open Library, a conference paper with no DOI, a dataset or a departmental working paper will all land here. Reading unverified as fake is the most common misuse of this kind of tool, and reading it as fine is the other one.
Suspicious. Always a model judgement. It picks up false precision and confident claims with nothing behind them, which is useful for deciding what to open first and is not a finding.
That nothing was caught in that passage. Not that the reference list is sound. A genuine DOI attached to a claim the paper does not support passes every check available here, because every signal says the reference is real. That is the error most likely to reach a reviewer, and only reading the source closes it.
Two different tools, and it is worth knowing where one stops.
Once a source is confirmed real, the formatting is a separate job. Our citation generator builds APA, MLA, Chicago, Harvard, IEEE and Vancouver entries, it is free with no account and no daily limit, and it runs entirely in your browser: it makes no network request, so nothing you type into it is sent anywhere.
Its honest limit is worth stating, because it decides whether it is the right tool for your list. It takes author, title, source, year, URL and access date, which is the correct shape for web pages and books. It has no fields for volume, issue, page range or DOI, so it cannot produce a correct journal-article reference in any of those six styles. For journal articles, take the publisher's own export or build the entry by hand against your style guide.
So the two steps divide cleanly. Checking asks whether the source exists, and for DOIs and arXiv IDs that is close to definitive. Formatting asks how to write it down, and for journal articles that part is still manual here.
The daily count gates scans and rewrites; citation checking is counted separately, and the generator is ungated.
| Free | Starter | Pro | |
|---|---|---|---|
| Price | $0 | $9.99/mo · $7.49 yearly | $19.99/mo · $14.99 yearly |
| Actions a day (shared) | 3 | 20 | Unlimited |
| Words per scan | 1,500 | 5,000 | 7,500 |
| Words per humanization | 260 | 5,000 | 10,000 |
| Detection a month | No cap | No cap | No cap |
| Humanizer a month | No cap | 40,000 words | 80,000 words |
One action is one scan, one file upload or one rewrite, from a single shared daily count. That daily number is the gate that matters: there is no monthly ceiling on any plan, so what separates the tiers is how often you can work and how long a single piece can be, not how much you get through in a month.
Read the free row together rather than line by line. No monthly cap sounds unlimited and is not: three actions a day at up to 260 words a rewrite is roughly 780 rewritten words in a day. It is a real allowance for checking and fixing specific passages, and it is not a plan for processing a backlog.
Current prices, the Business and Enterprise tiers and regional pricing are on the plans page, which is the page we keep up to date.
The hallucination checker is not one of those shared actions. It runs on its own allowance: 3 checks a day with no account, 5 on a free account, 25 on Starter and no daily limit on Pro. So checking a passage here does not spend a scan or a rewrite, and the row above does not cap it.
Paste the passage with its citations in place, and read the identifier results first. A DOI is resolved against CrossRef, an arXiv ID against arXiv, a PubMed ID against PubMed, an ISBN against Open Library, and a URL by fetching it. Those five are lookups rather than judgements, so a DOI that CrossRef has no record of is close to conclusive. Titles and journal names with no identifier are assessed by a model instead, which is a weaker class of finding.
That the entry could not be checked, which is deliberately a different result. An older monograph with no Open Library record, a conference paper without a DOI, a dataset or a departmental working paper all come back unverified. Treating unverified as fake is the most common misuse of a citation checker, and treating it as fine is the other one. It means go and look, by hand.
Because fabricated arrives by two routes: a lookup that returned nothing, which is close to conclusive, or a model judging the entry invented, which is a prompt rather than proof. The result tags which one it is, as a verifier finding or a model assessment, so check that tag before deleting a reference on the strength of the word alone.
No, and that is the gap worth understanding. Confirming that a reference exists is not confirming it says what your sentence says. A genuine DOI attached to an overstated claim passes every check available, because every signal reports the reference as real. That is the error most likely to reach a reviewer, and the only thing that closes it is opening the source and reading it.
Replace it. Find the real source for the claim and cite that, or cut the claim if no source supports it. Editing a fabricated reference until it looks plausible is how a fake citation survives into a final draft, because each small correction makes it harder to spot and none of them make the paper exist.
Not by volume: it is free, needs no account, has no daily limit, and runs entirely in your browser, so nothing you type into it is sent anywhere. Its limit is coverage. It takes author, title, source, year, URL and access date, which is right for web pages and books, and it has no fields for volume, issue, page range or DOI, so it cannot produce a correct journal-article reference. For those, use the publisher export or build the entry by hand.
Where the verification is real, and the gap it cannot close.
Read the guide →Six styles, no account, runs in your browser.
Open the tool →Three checks a day, no account.
Check a passage →Six ways to catch a fabrication, ranked by what is verified.
See the comparison →Three checks a day, no account. A DOI either exists in CrossRef or it does not, and that answer takes seconds.