Questions, answered plainly.

Both. Paste the trailing reference list on its own, or paste an entire manuscript draft that ends with a References, Bibliography or 参考文献 section. In draft mode CitePure isolates the bibliography, verifies its entries, and additionally compares in-text markers such as [1] against the list to flag out-of-range numbers and entries that are never cited.

No. CitePure never reads prose style and has no AI-text detector. It checks whether bibliographic entries correspond to real, findable records and whether their fields agree. A fully human-written paper can contain a fabricated reference, and a fully AI-written paper can have a perfect one.

No. Nothing is compared against a manuscript database, no similarity score is produced, and no text is added to any repository. CitePure is also not a citation formatter or a reference manager.

It means no comparable record was found in Crossref or OpenAlex during that run. Books, theses, conference abstracts, national-language journals and very recent preprints may not be indexed there. It is a retrieval result, not a fabrication verdict, and the exported report states this explicitly.

Crossref and OpenAlex, both public and keyless, queried directly from your browser. A DOI is resolved in Crossref and retried in OpenAlex if missing; title searches run in both. Google Scholar, PubMed, the arXiv API, CiNii / J-STAGE / JaLC, SciELO, CNKI and Wanfang are not connected in this version. See the methodology page for why.

No. All parsing happens in your browser with local JavaScript. The only outbound requests carry a DOI or a title to the two public metadata APIs. There is no account, no database, no analytics script, no cookie and no server-side copy of your text.

APA, MLA and Vancouver-style entries, numbered GB/T 7714 entries with [J] markers, and plain one-per-line lists, with or without a DOI. Chinese and other CJK titles are tokenised as character bigrams for comparison. Unusual formats can still be mis-parsed; the process log shows exactly which title rule fired.

There is no entry cap and there are no credits. Requests are throttled by 180 ms between entries to stay within the public APIs' polite-use limits, so a very long list simply takes longer. A built-in offline self-test verifies the logic without making any network calls.

Yes. The checker generates a bilingual (Chinese / English) Methods statement with the date, the number of entries, the three verdict counts and an explicit statement of limits. It says the authors read each source themselves, because the tool does not judge whether a source supports a claim.

A chimera citation combines real elements — a real author, a real journal, sometimes a real DOI — with a title or paper that does not exist. The most deceptive form attaches a valid DOI to a different paper. CitePure resolves the DOI first and, when the resolved title does not match the submitted title, reports "Real DOI resolves to a different paper" as its own Mismatch reason.

The site currently ships in English. CJK titles are compared using character bigrams, and 参考文献 headings are recognised, so Chinese reference lists parse. Local indexes such as CiNii, J-STAGE, JaLC, SciELO, Dialnet, CNKI and Wanfang are not yet connected, so coverage for non-English material is weaker. Japanese, Spanish and Portuguese editions are planned for a later phase; connecting their local catalogues is the more important part of that work.

Not in this version. The verification logic is plain client-side JavaScript in assets/app.43ddfd7e.js. A documented API / MCP tier is a future capability listed on the pricing page, not something that exists today.