A pre-publication check on a manuscript's reference list — run per submission, as part of your normal review process. Every citation gets checked against 7 academic databases and comes back Verified, Mismatch, or Hallucinated, with a real replacement suggested when something doesn't check out.
A single database lookup has a specific failure mode: a fabricated or duplicate DOI can happily resolve inside one provider's records even when the citation is wrong, because that provider indexed the bad entry too. Trusting whichever database answers first means trusting whichever database is wrong.
AccuraCite queries OpenAlex, Crossref, Semantic Scholar, PubMed, DBLP, arXiv, and CORE for every reference and looks for independent agreement across them, rather than accepting the first hit. A citation is marked Verified only when multiple independent sources agree on it — which is specifically what makes the check resistant to a single bad or duplicate DOI entry sitting in any one database.
That handles whether a paper exists. It doesn't handle whether it says what the manuscript claims it says — so for every candidate match, AccuraCite also pulls the source's abstract and checks that it actually supports the citation, not just that the title and author string line up. A real paper cited for the wrong claim is still flagged.
Below is an illustrative excerpt — 9 of 32 references from a hypothetical submitted manuscript — showing the mix of results a real reference list typically produces. Most citations verify cleanly; a small number get flagged, and each flag comes with the reason.
Manuscript #2026-0114 — Reference List
Showing 9 of 32 citations (illustrative example)
Ferreira, T., & Nakamura, H. (2021). Attention bottlenecks in multimodal transformers. Advances in Neural Information Processing Systems, 34, 8891–8903.
VerifiedMatched on OpenAlex, Crossref, and Semantic Scholar; abstract confirms the claim it's cited for.
Patel, S. (2019). Longitudinal effects of sleep fragmentation on working memory. Journal of Cognitive Neuroscience, 31(4), 512–529.
VerifiedMatched on Crossref and PubMed.
Kowalski, M., Lindqvist, A., & Osei, B. (2020). Federated learning under non-IID data: A survey. IEEE Transactions on Neural Networks and Learning Systems.
MismatchTitle and authors match a real paper — but it published in 2022, not 2020, and the DOI in the manuscript points to an unrelated article. Suggested replacement: the correctly dated 2022 record.
Chen, L. (2018). Microbiome diversity and host immune response in zebrafish. Cell Reports, 24(3), 601–612.
VerifiedMatched on Crossref, PubMed, and OpenAlex.
Alvarez, R., & Kim, J. (2023). Scaling laws for instruction-tuned language models. arXiv:2301.04567.
VerifiedMatched on arXiv and Semantic Scholar.
Whitfield, D. (2017). A unified theory of catalytic asymmetric synthesis. Journal of the American Chemical Society.
HallucinatedNo record across any of the 7 databases — this author, title, and venue combination doesn't correspond to any indexed publication. A closest-match real paper is suggested for the authors to confirm or correct.
Brandt, S., & Okafor, C. (2022). Carbon sequestration potential of urban green infrastructure. Environmental Science & Policy, 128, 45–56.
VerifiedMatched on Crossref and OpenAlex; abstract confirms the sequestration claim cited.
Liu, Y. (2020). Deep reinforcement learning for traffic signal control. Transportation Research Part C.
MismatchThe DOI listed in the manuscript resolves to a different, unrelated 2015 conference paper — a duplicate/fabricated DOI, the exact case cross-provider consensus is built to catch. The title exists in the record, but under a different DOI.
Suzuki, N., Mehta, R., & Dubois, P. (2021). Placebo response rates in randomized trials for chronic pain: A meta-analysis. Pain Medicine, 22(6), 1301–1314.
VerifiedMatched on PubMed, Crossref, and Semantic Scholar.
This is built to run per manuscript, when it needs to — not as a one-time audit of everything a journal has ever published.
No. It checks whether a citation exists, matches its stated metadata, and is topically supported by its abstract. It doesn't evaluate methodology, novelty, or whether the manuscript's argument holds up — that's still on you and your reviewers.
PDF files aren't permanently stored — they're processed for the verification job and discarded once it completes. To check each reference, we send the citation metadata (titles, authors, DOIs) to public academic databases like Crossref and PubMed, the same information already visible in the citation itself. See the Privacy Policy for full detail.
The Bulk Verification plan scans up to 30 PDFs per job — sized for a batch of the week's incoming submissions, not a full-journal audit. It's built to run per-manuscript as part of your normal rolling review, not as a one-time sweep of everything you've ever published.
There's no dedicated institutional plan yet. Individual editors can use the Bulk Verification plan directly, and if you want to talk through how it would fit your review process, contact us.
Tell us how your review process works and we'll tell you honestly whether this fits.
Want the mechanics in more detail first? Read how AccuraCite works.