# AI for Preprint & Manuscript Fact-Checking — Biomedical Citation Verification > **Reviewed:** 2026-05-29 > **Canonical HTML:** https://bioskepsis.ai/use-cases/ai-for-preprint-fact-checking > **Publisher:** BioSkepsis (EFEVRE TECH LTD, Larnaca, Cyprus) ## Goal Systematically verify the claims, citations, effect sizes, methods, and novelty assertions of an uploaded preprint, manuscript, or research proposal against the published literature — to identify what is well-supported, what is overstated, and what is missing. ## Built for peer reviewers, editors, PIs, and authors - **Peer reviewers & editors:** triage a submission's claims, citations, and methods fast, with PMID-grounded evidence behind every verdict. - **PIs & authors:** pressure-test your own draft before submission — catch overstated effect sizes and weak novelty claims before a reviewer does. - **Research-integrity & due-diligence teams:** spot selective citation, inflated statistics, and unsupported attributions. ## The 9-step workflow — upload to verified verdict ### Step 1 — Upload the document and request a fact-check of its core mechanistic claims Attach a PDF (preprint, manuscript, or proposal) and ask BioSkepsis to verify its central claims against the published literature, providing relevant citations for each validated or challenged assertion. **Example query** > "Fact check this preprint article and give me all relevant citations" ### Step 2 — Review the initial fact-check synthesis BioSkepsis returns a structured verification organised by the document's main mechanistic claims. Each claim is tagged as validated, partially supported, or not reported (NR) with PMID-grounded evidence. Check the **unverified citations panel** — this is where the system catches mismatches between what the document asserts and what the cited papers actually show. ### Step 3 — Request a deeper verification of the core claim Ask BioSkepsis to re-verify the central mechanistic chain with more granularity. This second pass often surfaces additional supporting or contradicting evidence that the first pass did not fully elaborate. **Example query** > "Verify the core claim" ### Step 4 — Verify cited effect sizes and quantitative claims Ask BioSkepsis to check the specific numbers, statistics, and quantitative results the document cites. This catches inflated effect sizes, misquoted p-values, or figures that do not match the source papers. **Example query** > "Check cited effect sizes" ### Step 5 — Identify contradicting or qualifying evidence the authors did not cite Ask what published studies contradict or qualify the document's main conclusions that were not included in its reference list. This exposes selective citation. **Example query** > "What published studies contradict or qualify the main conclusions of this preprint that we did not cite?" ### Step 6 — Assess whether the proposed mechanism is established, contested, or speculative Ask BioSkepsis to evaluate each link in the proposed mechanistic chain and classify it as well-established, contested, or speculative based on the broader literature. **Example query** > "The preprint proposes a specific biological mechanism. What does the broader literature show about this mechanism — is it well-established, contested, or speculative?" ### Step 7 — Check for selective citation bias Ask whether the document's references are representative of the field or whether the authors have selectively cited papers that support their argument while ignoring those that do not. **Example query** > "Are the references in this preprint representative of the field, or has the author selectively cited papers that support their argument while ignoring those that don't?" ### Step 8 — Evaluate the reliability of the experimental methods Ask what the literature says about the reliability, limitations, and known pitfalls of the specific experimental methods or analytical approaches used in the document. **Example query** > "The preprint uses a specific experimental method or analytical approach. What does the literature say about the reliability, limitations, and known pitfalls of this method?" ### Step 9 — Verify sample sizes, gene/protein attributions, missing citations, and novelty claims Close the fact-check by probing four remaining dimensions: whether the sample sizes and power claims are adequate, whether the document's attribution of function to specific genes or proteins is confirmed by the literature, what the authors should have cited but did not, and whether the novelty claim holds or has been undermined by prior work. **Example query — sample size** > "Verify sample size and power claims" **Example query — attribution** > "The preprint attributes a specific function or disease association to a gene, protein, or variant. Does the published literature confirm this attribution, or is the evidence weaker than presented?" **Example query — missing citations** > "Identify what the authors should have cited but didn't" **Example query — novelty** > "The preprint claims this finding is novel or first-of-its-kind. Has anything similar been published before that would undermine that claim?" ## What you walk away with - **Claim-by-claim verdict** — each main claim tagged validated, partially supported, or not reported, every verdict PMID-grounded - **Effect-size & stats check** — inflated effect sizes, misquoted p-values, mismatched figures, and weak sample-size or power claims caught - **Selective-citation audit** — contradicting or qualifying evidence the authors omitted, plus what they should have cited but did not - **Methods & novelty assessment** — reliability and pitfalls of the techniques used, and whether the "first-of-its-kind" claim actually holds ## Try the workflow Upload a preprint, manuscript, or proposal PDF and run the fact-check against the published literature: **https://app.bioskepsis.ai/signup** ## Honest limits - **BioSkepsis verifies against what is published and indexed.** It cannot check raw data, rerun analyses, or detect fabrication that the literature does not contradict. - **A "not reported" (NR) tag means the corpus does not confirm a claim — not that it is false.** Absence of supporting evidence is not proof of error. - **It assists peer review; it does not replace it.** The reviewer's judgement and the journal's process remain the decision-makers. - **Verify each flagged citation against the source** before acting on it in a review, rebuttal, or revision. ## FAQ **What document types can I upload to fact-check?** Any biomedical preprint, manuscript, or research proposal as a PDF. BioSkepsis verifies the document's claims against its 40M+ paper biomedical corpus (PubMed, bioRxiv, medRxiv) and returns PMID-grounded evidence. **What do the validated, partially supported, and not reported (NR) tags mean?** Each claim is tagged by how well the published literature supports it: validated (well-supported), partially supported (some support, with caveats), or not reported / NR (the corpus does not confirm it). NR means no supporting evidence was found — not that the claim is necessarily false. **Can it catch selective citation and missing references?** Yes. Steps 5 and 7 ask which published studies contradict or qualify the conclusions that were not cited, and whether the reference list is representative or skewed. Step 9 also asks what the authors should have cited but did not. **Does it check effect sizes and sample sizes?** Yes. Step 4 verifies cited effect sizes and quantitative claims against the source papers; Step 9 separately probes whether sample sizes and statistical-power claims are adequate. **Is this for authors or for reviewers?** Both. Authors pressure-test their own draft before submission; peer reviewers and editors triage a submission's claims, citations, and methods. It assists the review — it does not replace the reviewer's judgement or the journal's process. ## Related - [Use Case: AI for Literature Gap Analysis](/use-cases/ai-for-literature-gap-analysis) - [Use Case: AI for Biomedical Thesis Writing](/use-cases/ai-for-thesis-writing) - [Use Case: AI for Precision Medicine](/use-cases/ai-for-precision-medicine) - [Feature: AI Reference Finder](/features/ai-reference-finder) --- BioSkepsis is a product of EFEVRE TECH LTD (Larnaca, Cyprus). BioSkepsis is a verification aid for peer review and authoring; it does not replace editorial judgement or the journal's peer-review process.