Science advances
one verified claim at a time.
Every published paper becomes a permanent node in the scientific record. What it claims, and what it cites to support those claims, shapes what the field believes — and builds upon — for decades.
Founded by a scientist.
Built from a problem
the field refuses to name.
Submedit was founded by an active scientist — a corresponding and first author with publications in journals above impact factor 40, and a researcher with postdoctoral experience at Harvard Medical School and the Johns Hopkins School of Medicine.
The founding insight did not come from a market analysis. It came from years of reading the primary literature and sitting on both sides of peer review — and noticing how consistently strong manuscripts were undermined by failures that had nothing to do with their science. A value in the text that no longer matched the figure it came from. An argument whose logic broke between two sections. A claim positioned so cautiously that its significance never registered. And, most often of all, a citation that — when retrieved and read against the specific assertion it was meant to support — said something meaningfully different from what the author believed.
Not occasionally. Consistently. Across fields. Across journals. Across four decades of documented research. The scientific community has known about this problem since at least 1984. It has not improved.1,2,3
"A single citation error rarely ends a review. But when a careful reviewer flags one in the Methods — then finds another inconsistency in the Discussion, and a third mismatch between the text and a figure — the cumulative effect on their confidence in the manuscript is not recoverable. The direction of the assessment is decided before the final paragraph is written."
Do-Hyeon Kim, Ph.D.·Founder, SUBMEDIT"The most frustrating rejections — those I witnessed as both author and reviewer — were not failures of science. They were failures of communication. The reviewer did not recognise what the manuscript was claiming because it had not been delivered with enough clarity and precision. That is a preventable loss to science. It is what we exist to prevent."
Do-Hyeon Kim, Ph.D.·Founder, SUBMEDITHow a single flawed citation
corrupts an entire field.
Of every integrity failure a manuscript can carry, one does damage beyond the paper it appears in. Science advances by building on what came before — each paper extending a chain of cited evidence. Independent meta-analyses across 32,074 references now establish that approximately 1 in 6 citations contains a significant error, a rate that has remained unchanged for over 40 years since first documented in 1984.1,3 Critically, 38% of those errors cite conclusions the source paper never actually demonstrates — not misquotation, but systematic misrepresentation of the evidence.5 When a flawed claim passes undetected through peer review, each paper that cites it carries the error forward. This is the snowball effect: a single unverified citation cascading through the literature across decades.
Drug X showed mild efficacy in 12% of mice models.
Drug X has proven efficacy in mammalian subjects [1].
Drug X is a highly effective treatment [2].
Given Drug X's established cure rate [3], our new compound…
Why peer review cannot solve this alone.
Peer reviewers are among the most capable scientists in their fields. But the structure of peer review does not permit exhaustive citation verification. At leading journals, editors report approaching 8–10 researchers before one agrees to review. Those who accept evaluate complex manuscripts in hours — not the days required to retrieve and read every cited source against the claim it supports, reconcile every number against every figure and supplementary table, and trace the logic of the argument across all sections.
The result is systematic: the dimensions of a manuscript that are most mechanically verifiable — citation validity, numerical consistency, figure–text agreement — are among the least rigorously checked at the point that matters most. This is not a failure of peer reviewers. It is a structural limitation of the process. The verification that peer review cannot reliably provide must happen before submission.
The global scientific community collectively invests over 100 million hours annually in peer review.7 Every hour spent checking citation mismatches or figure inconsistencies that should have been resolved before submission is an hour not spent on the scientific evaluation that peer review exists to perform.
Enabling peer review to do
what only it can do.
Peer review exists for one purpose: to evaluate whether a study's claims are scientifically valid, whether the findings are meaningful, and whether they contribute sufficiently to warrant publication. That evaluation — the assessment of whether the science itself holds — is work that only a qualified scientific expert can perform. It cannot be delegated. It cannot be automated.
What does not belong there is everything else: verifying whether citations say what the authors claim they say, reconciling numerical values across figures and supplementary files, identifying where the logical argument breaks down between sections. A diligent reviewer who flags a citation inconsistency in the Methods section will note it, and then look more carefully. If they find another in the Discussion, and a third inconsistency between the text and a figure, the manuscript's evidentiary credibility has been placed in doubt — not because the science is wrong, but because the presentation of the evidence is unreliable. That accumulated doubt often shapes the final decision.
The consequence is a systematic misallocation of effort. Important findings fail because the manuscript did not deliver the central claim with enough precision for the reviewer to recognise its significance. The reviewer did not grasp the breakthrough — not because the breakthrough wasn't there, but because the manuscript didn't make it unmissable.
Submedit exists at this precise gap. Our role is not to evaluate whether the science is correct — that is the function of peer review, and it should remain so. Our role is to ensure that when a manuscript enters peer review, the reviewer encounters nothing but the science.
Every citation verified against its source so the reviewer need not check it. Every numerical value cross-validated so no inconsistency remains to trigger doubt. Every argument structured so that what the authors are claiming is unmistakable. When a manuscript arrives in this condition, reviewers can do what they are uniquely positioned to do: assess the science.
At the scale of the scientific literature, this effect reaches beyond any individual paper. When manuscripts enter the record with their evidence verified, the chain of citations through which science advances carries less accumulated error. The scientific record becomes more reliable.
That is what Submedit is built to contribute. Not to edit a manuscript. To improve the conditions under which science advances — one verified paper at a time.
Two kinds of intelligence.
One editorial system.
Traditional manuscript editing is constrained by a structural limitation that quality control alone cannot resolve. A single editor reads a complex paper sequentially — and as the manuscript grows longer, cognitive fatigue sets in. Language receives scrutiny while citations are skimmed. A senior editor reviewing the result compounds the limitation further: they see the work from a summary level, and the errors that passed a close reading rarely surface from an overview.
This is not a problem of diligence. It is a problem of architecture.
Submedit’s answer is Orchestrated Intelligence™ — a review architecture in which machine intelligence and expert human intelligence are not alternatives, but a single collaborating system. A team of specialised AI agents processes the full manuscript before any editor opens it: every citation retrieved and tested against the claim it supports, every value reconciled against every figure and supplementary table, every argument traced across sections, and the current literature searched for what has changed since the draft was written. The AI is not fatigued. It reads page 38 with the exact same absolute precision as page 1. The output is not a summary — it is a complete map of every target that requires human expert judgment.
That map is then distributed — not to one generalist, but to the specific experts each layer demands: a language editor, a sub-field scientist for structure and argument, a reviewer-level specialist for strategy and hedging, and a QA editor who verifies the work before it returns to you. Neither intelligence reaches this standard alone. The AI cannot judge what a finding means to your field. The expert cannot hold thirty-eight pages at uniform attention. Orchestrated together, each covers precisely what the other cannot.
A team of specialised AI agents processes the full manuscript before any human editor opens the file. Every citation is retrieved and tested against the claim it supports. Every numerical value is mapped against every downstream instance. Figures, tables, and supplementary files are reconciled against the body text. The current literature is searched for work published since the draft was written. Nothing is skimmed.
Editors receive a structured, prioritised report — the precise location and nature of every issue identified. They apply domain expertise to a verified target set, not to an unsorted sequential reading of a long document.
The same framework drives expert assignment. It analyses the sub-field specificity, citation landscape, methodological context, and target journal of each manuscript — then matches it to available researchers actively publishing at that exact intersection. Identifying who is genuinely qualified to edit a given manuscript is itself a problem conventional editing services cannot solve.
Each layer is handled by the specialist it requires — language, structure and argument, reviewer-level strategy — and a QA editor verifies the complete work before delivery. The result is the review neither intelligence achieves alone: exhaustive machine coverage combined with the judgment of scientists who understand what your findings mean and what your reviewers will ask.
Trusted by researchers across
every major scientific discipline.
Upload your manuscript.
We review it to the standard of the scientists who will.
Every citation verified. Every numerical value cross-checked. Every argument assessed by an active researcher in your field.
Upload Your Manuscript →