Upload a PDF. Citation Monster extracts every reference with the page and claim it is cited for, checks each one on the web, and flags citations that look fabricated or misattributed. Then ask questions and get answers cited to the page.
Free to start. No credit card needed.
Reference check
Attention is all you need
VerifiedVaswani et al., 2017. Cited on page 3
For the Transformer encoder that BERT's architecture is based on.
Improving language understanding with unsupervised learning
Metadata differsRadford et al., 2018. Cited on page 1
Published as “Improving Language Understanding by Generative Pre-Training”.
Quora question pairs
UnverifiableChen et al., 2018. Cited on page 14
A dataset release, not a paper, so there is nothing to match.
Four views of the same paper, all built from one upload.
Claude reads the whole paper and builds a clean reference list with titles, authors, years and venues. It also notes the page where each reference is cited and the claim it is cited for, so the reference check can test that claim.
One click searches for every cited work and compares it with the citation. Fabricated references show up as not found; wrong titles, authors or years show the corrected details.
Ask about methods, results or a single reference. Answers point to the page they come from, and the assistant can read the other papers in your workspace to compare them.
A knowledge graph links the paper to every work it cites and shows which of those cite each other, so you can spot the core works and the outliers.
Drop in a paper, preprint or thesis chapter. Claude reads every page and pulls out the title, authors and full reference list, usually within a few minutes.
Look over the extracted list and approve it. Approval builds the knowledge graph, and one click on the references page checks every reference on the web.
See which references need attention, ask questions about the paper with answers cited to the page, and export the list to CSV, BibTeX or RIS.
Start without a credit card. Paid plans are billed per workspace through Stripe. Cancel renewal any time from the account menu; you keep access until the end of the period.
How it works, what it checks and what happens to your papers.
You upload a paper as a PDF. Claude reads it and lists every reference it cites, with the page and the claim each one supports. Each reference is then checked on the web, so you can see which ones are verified, which have the wrong title, authors or year, and which could not be found at all. You can also ask questions about the paper and explore its references as a graph.
Choose Check references on a paper's references page. For each reference, Claude searches the web for the cited work and compares what it finds with the citation: verified means the work exists as cited, metadata differs means the title, authors, year or venue don't match (the corrected details are shown), and not found means no matching work turned up, which is often a sign of a fabricated citation. It also checks whether the source supports the claim the paper makes with it. Treat the results as a first pass and open the linked source before you rely on it.
Reference extraction and verification run on Claude, Anthropic's model family. In paper chat you can pick the model yourself with the model picker next to the message box, including GPT models from OpenAI.
Yes. Chat answers cite the page they come from, and the assistant can read the other papers in your workspace when you ask it to compare or connect them. Invite colleagues to a workspace to share papers.
Extraction works on any PDF you upload, including papers you have access to through a subscription. Verification uses publicly available information, so a paywalled reference can still be confirmed from its publisher page or index entry, but its full text may not be checked.
CSV, JSON, BibTeX, RIS and a plain list of links. The knowledge graph exports as PNG, JPEG or PDF.
Papers you upload are stored in your workspace so you can come back to them, and only members of that workspace can open them. To extract, verify and answer questions, the text of a paper is sent to the AI providers that run those steps.
Uploads must be PDF files up to 50 MB. Scanned PDFs without a text layer may come out empty; try a version with selectable text. If it still fails, the upload screen shows the reason and lets you try again.
Upload a PDF and see which citations hold up.
Citation Monster