Why it is so hard to find peer reviewers

Finding peer reviewers is now the hardest part of an editor’s job, and the usual explanation — reviewers are busier — is only part of it. The structural problem is concentration: invitations cluster on a small, visible, senior group who are already over-asked, while a much larger pool of capable reviewers is never invited at all.

Editors say so themselves. In Publons’ 2016 editor survey, 75% named “finding reviewers and getting them to accept review invitations” as the hardest part of the job — ahead of every other editorial task.

The ratio is measured, not anecdotal. Publons’ Global State of Peer Review (2018), drawing on ScholarOne submission data, found that an editor in 2017 needed an average of 2.4 invitations to obtain one completed review, up from 1.9 in 2013, and projected 3.6 invitations per review by 2025 if the trend continued. Over the same window, review invitations grew 9.8% year-on-year while articles accepted for publication grew 4.9% annually — the gap, not the absolute volume, is the thing that is breaking.

That framing matters because it points at levers an editor actually controls. You cannot make the community less busy. You can change who gets asked, how often, and how much reaches review in the first place.

What the reviewer shortage actually measures

The reviewer shortage is a shortage of invited reviewers, not of qualified ones, and two independent published attempts to quantify it agree on the shape.

Kovanis, Porcher, Ravaud and Trinquart modelled the whole biomedical system in PLOS ONE (2016;11(11):e0166387) and estimated 63.4 million hours devoted to peer review in 2015, of which 18.9 million hours — close to 30% — came from the top 5% of contributing reviewers. Their central finding is the distributional one: “20% of the researchers performed 69% to 94% of the reviews,” and among researchers who reviewed at all, 70% gave 1% or less of their research work-time to it while 5% gave 13% or more. The system is not short of hours. It is short of hours from the people it never asks.

Publons reaches the same shape from a different direction with its Review Distribution Index, a Gini coefficient over reviews rather than income: 0 means every researcher reviews equally, 1 means one person does everything. The 2018 report puts it at 0.599 for established regions and 0.642 for emerging ones, and notes that within Publons “~10% of reviewers are responsible for ~50% of the peer review records.”

Both figures carry a caveat worth stating rather than eliding. Publons records are self-registered, so its population is people who chose to log reviews; Kovanis and colleagues modelled from publication counts and per-review time estimates rather than from any platform’s users. They are worth citing together precisely because their biases do not overlap, and they converge on a Pareto distribution either way.

Why the same names keep coming up

Invitation lists are built from what is easy to see: corresponding authors of recent related papers, previous reviewers for the journal, editorial board members, and names the handling editor already knows. Every one of those selects for visibility, and visibility correlates with seniority.

The mechanism is mundane and it is structural. The corresponding author is usually the only author whose email address survives into the citation record, so the reachable name attached to a relevant paper tends, by construction, to be the most senior or most administratively central author on it. Editorial systems inherit that constraint from the metadata, not from any editorial judgement about who reads best.

The consequence is that a minority of researchers carry a disproportionate share of review, get invited most often, decline most often because they are saturated, and are replaced by invitations to the same kind of person.

Meanwhile postdocs and newly independent researchers — who are closest to the methods, often the most careful readers, and demonstrably willing — are rarely on the list because they are rarely corresponding authors. That exclusion is not a quality trade. Callaham and Tercier tested the assumption directly in PLoS Medicine (2007;4(1):e40), scoring 2,856 reviews of 1,484 manuscripts by 306 reviewers at Annals of Emergency Medicine between 2002 and 2005 against editor ratings. In the multivariable model, “relative youth (under ten years of experience after finishing training)” was one of only two characteristics that predicted higher-quality reviews; formal critical-appraisal training, academic rank and principal-investigator status all failed to. No marker predicted performance usefully — the odds ratios sat below 2 and the area under the curve was 0.52, barely better than chance. Seniority is not a quality signal. It is a findability signal.

The decline data points the same way. Publons’ 2018 Global Reviewer Survey of 11,838 researchers, run May to July 2018, asked why reviewers turn invitations down and let respondents pick up to two of twelve options. The top answer was not workload: 70.6% chose “article was outside my area of expertise,” against 42% for “too busy with my own research.” The most common reason an invitation fails is that it went to the wrong person — which is a targeting defect an editor can fix, not a capacity ceiling the community imposes.

What changes the numbers

Invite from middle authorships, not just corresponding authors. The person who did the experiments is frequently the best-qualified reader of a paper using that method, and is almost never asked. Two artefacts now make that person addressable. The CRediT taxonomy, standardised as ANSI/NISO Z39.104-2022 with 14 contributor roles, defines Investigation as “Conducting a research and investigation process, specifically performing the experiments, or data/evidence collection” — journals that print CRediT statements are publishing a list of exactly who ran the assay you need assessed. ORCID identifiers on the byline give that person a stable, resolvable handle when their email is not printed. A practical heuristic: for a manuscript whose central claim rests on one technique, read the methods section, find the paper it cites for that technique, and invite the Investigation-role author of that paper rather than its last author. See postdocs and PhD students for how the same people describe being on the wrong side of this.

Ask senior reviewers to name a co-reviewer. Formal co-review with a trainee, credited by name, expands the pool and trains it. Many journals now support this explicitly; those that do not are leaving capacity unused. The practice is already happening whether or not the journal sanctions it — see the next section for what the surveys found and what the policies actually say.

Make the invitation answerable in thirty seconds. Title, abstract, deadline, and scope of what you want assessed. Invitations requiring a login to see what is being asked convert worse. The reason is the 70.6% figure above: an invitation that does not show the abstract in the body of the email forces a scope judgement the recipient cannot make, and the safe answer to a question you cannot evaluate is no.

Ask for less, explicitly. A targeted request — assess the statistics, or the clinical relevance, not the whole paper — is easier to accept and produces a more useful review than a general one. Two focused reviews often beat one exhaustive one. A named partial role also lets you reach specialists who would decline a general invitation on scope grounds: a biostatistician asked to assess the survival analysis and nothing else is being asked a question they can answer in ninety minutes rather than an open-ended one. Statistical review capacity covers the specialist shortage inside the general one, and reviewer objections about statistics shows what those reviews contain when they arrive.

Stop inviting people who declined twice recently. Repeated invitations to a saturated reviewer produce declines and resentment. Most editorial systems record a decline reason code; a cooldown keyed to “too busy” — say, no further invitation for 120 days — is one configuration change and preserves a reviewer you will need later. A decline coded “outside my expertise” is a different signal and should update the targeting logic, not the calendar.

How to run co-review without ghostwriting it

Co-review is already the majority practice among early-career researchers, and the failure mode is not the co-reviewing — it is that the journal is never told.

McDowell and colleagues surveyed 498 researchers in September 2018 and reported in eLife (2019;8:e48425) that 73% of respondents had co-reviewed a manuscript, that 44% had ghostwritten a review — done the work without their name reaching the editor — and that 79% of postdocs had co-reviewed on an invitation addressed to their own principal investigator. The ethical position is not in dispute among the same respondents: 81% disagreed that ghostwriting a review is an ethically sound practice, and 82% agreed it would be valuable to have their own name added to a peer review report. The gap between those numbers and the 44% is a policy gap, not a values gap.

The governing text is permissive and specific. The ICMJE Recommendations, in the section on the responsibilities of peer reviewers, state that “Reviewers who seek assistance from a trainee or colleague in the performance of a review should acknowledge these individuals’ contributions in the written comments submitted to the editor,” alongside the standing requirement that “Reviewers therefore should keep manuscripts and the information they contain strictly confidential.” PLOS ONE’s reviewer guidelines go further and encourage it: “Co-reviewing is a great way to gain peer review experience under the mentorship of an experienced reviewer and we encourage this collaboration,” coupled with a hard requirement — “If you had help completing the review you must share your collaborator’s name with the journal when you submit the review,” entered in the confidential comments to the editor rather than in the review text.

What an editor can do with that, concretely: put a named, structured field for co-reviewers in the review form rather than leaving the name buried in an unstructured confidential comments box; say in the invitation email that co-review with a trainee is welcome and how to declare it; and add the declared co-reviewer to the reviewer database with the subject terms of the manuscript they reviewed. That last step is the one that converts a training practice into recruitment, because it is the only point at which the invisible reviewer becomes findable. What a peer reviewer does sets out the role a trainee is being credited for.

A worked invitation

An invitation that gets answered states the paper, the ask, the scope and the deadline in the body of the email, before any link. Compare the two forms.

The version that converts badly: “You have been invited to review a manuscript for Journal X. Please log in to the editorial system to view the submission and respond by 12 September.” The recipient can evaluate nothing without a login, and the only information given — the journal’s name — does not answer the scope question that 70.6% of surveyed reviewers named as a reason for declining.

The version that converts: the manuscript title, the 250-word abstract pasted in full, one sentence naming why this reviewer specifically (“you published the 2024 validation of this assay”), the specific scope requested (“we would like your assessment of the flow cytometry gating and the immunophenotyping claims; a second reviewer is covering the clinical interpretation”), the expected time commitment, the return date, and only then the accept and decline links. Publons put the median time to complete a review at 16.4 days after acceptance and the median writing time at 5 hours per review in 2016; an invitation that names a scope narrower than the whole paper is asking for a defensible fraction of those 5 hours, and can say so.

Add one field to the decline path: “if you are declining, can you name someone who could do this?” A decline that returns a name is a recruitment event rather than a loss, and it reaches exactly the middle-author population that the corresponding-author metadata hides.

The upstream lever

The most effective way to reduce reviewer load is to send fewer papers to review.

Every manuscript that reaches a reviewer with an incomplete methods section, an unreported ethics statement, missing data availability, or a claim the design cannot support consumes reviewer time on problems no reviewer is needed to identify. Reviewers report this as the most demoralising part of the task: spending an evening of their own writing out what the authors could have been told before submission.

The arithmetic is worth doing on your own journal. Publons found an average of 2.7 completed reviews per submission across research areas once desk rejections were removed from the sample, and a median of 5 hours spent writing each review. A submission that reaches review therefore consumes on the order of 13 reviewer-hours. Every manuscript sent out that a checklist pass would have caught spends those hours on items that are enumerable in advance: CONSORT for randomised trials, STROBE for observational designs, ARRIVE for animal work, and the data availability and accession statements that are present or absent with no judgement required.

Sharpening desk screening does more for reviewer capacity than any recruitment strategy, because it removes demand rather than chasing supply. Desk screening at scale covers what can be checked before a reviewer is invited. The boundary matters in both directions: a screen that rejects on reporting completeness alone will discard sound work that was written carelessly, which is the complaint behind desk rejected without review. Screening should decide what is reviewable, not what is true.

When the obvious remedy is unavailable

Broadening the pool fails in specific, recognisable situations, and each has a different second-best.

The field is genuinely small. In a subfield with thirty active groups, conflict-of-interest exclusions can remove most of them. The workable move is to split the review by competence rather than by paper: invite a methods reviewer from outside the subfield who can assess the design and the statistics without knowing the literature, and pair them with one subject expert who is asked only about novelty and interpretation. Two narrow reviews from two shallow pools beat one impossible search for a reviewer who is expert in both and conflicted in neither.

The authors have excluded everyone plausible. A non-preferred-reviewer list that removes the entire competent population is itself editorial information. Journals differ on whether such requests are binding, and an editor is entitled to say in writing that the exclusion cannot be honoured without abandoning review.

Everyone declined. After a second failed round, the diagnostic question is which decline code dominates. Expertise declines mean the targeting is wrong and a new search strategy is needed. Workload declines mean the timing is wrong and a longer deadline, or a narrower ask, will land. Silence — no response at all — usually means the invitation was not read, and is a subject-line and sender-address problem before it is a supply problem.

The manuscript needs a competence the journal cannot recruit. Statistical review is the standard case, and the honest options are a standing statistical board, a commissioned external statistician, or a documented decision to publish without statistical review. Pretending the third did not happen is what produces the corrections.

Transfer is available and underused. Cascading review — moving a manuscript with its completed reports to another journal in the same portfolio — recovers reviewer hours already spent. Its constraint is consent: the reviewers agreed to review for one journal, and the transfer policy has to say what happens to their reports and their names.

What a declining reviewer is telling you

Decline messages carry operational information that is usually thrown away, and reading them by category changes the next invitation rather than the next reviewer.

“This is outside my area — I work on the imaging, not the model” says the paper was matched on a keyword rather than on a method, and the same search will fail again. “I would be glad to but not before March” is a scheduling answer being read as a refusal; a deferred invitation, logged with a date, converts. “I have already reviewed this for another journal” is a signal about the manuscript’s history that belongs in the editor’s file. “My postdoc did the last one with me and would do this” is a recruitment offer that most systems have nowhere to record. “I have declined three invitations from you this quarter” is a load complaint about your journal specifically, not about peer review.

The one that should stop an editor is: “I read the first three pages and this is not ready for review.” That is not a reviewer problem. It is a screening problem arriving late and landing on a volunteer, and it is the complaint that predicts the reviewer’s next decline.

What AI can and cannot do here

AI cannot replace peer review. Judgement about significance, novelty and whether a field should believe a claim is the reviewer’s contribution and does not transfer.

What AI can do is remove the mechanical layer before a human sees the paper: whether the reporting guideline items are present, whether numbers reconcile across tables, whether the statistical approach matches the design, whether claims in the abstract are supported by the figures. That is the part reviewers most resent spending time on and the part that is most reliably checkable.

The distinction is not a matter of taste, and it maps onto the decline data. A reviewer who accepts a targeted invitation and then spends the first two of their five hours listing absent checklist items has given up an evening to clerical work, and will decline the next invitation. Removing that layer does not shorten the review; it changes what the five hours are spent on.

The confidentiality constraint is separate and non-negotiable — see can reviewers use AI on manuscripts, which is a different question from what a journal itself may run on a submission it holds. A journal’s own position on running checks belongs in its published policy rather than in an editor’s practice; AI peer review journal policies sets out what such a statement has to cover.

For editors evaluating a tool

PerfectPaper produces structured, anchored findings on submitted manuscripts, with the review contract and output format fixed by the application rather than by the prompt. Editorial questions — including what a journal-level arrangement looks like — are handled directly rather than through a sales process.

PerfectPaper applies the same review contract to every submission, so the findings on two manuscripts are comparable to each other rather than to how a prompt happened to be phrased that day. That property is what makes the output usable as a screening input at all: an editor deciding whether a paper is reviewable needs the check to be identical across the queue, and a check that varies with wording is a check that cannot be audited.

More in this cluster: the peer review capacity problem.

Related

The peer review capacity problem · Desk screening at scale · Statistical review capacity · Can reviewers use AI on manuscripts · What a peer reviewer does

Talk to us about journal use

Frequently asked questions

How do I find peer reviewers for a manuscript?

Search on the method rather than the topic, then invite the person who performed it: the Investigation-role author under a CRediT statement, or the middle author of the paper the methods section cites for its central technique. Send the abstract in the invitation email, name the specific scope you want assessed, and ask anyone who declines to suggest a replacement.

Why is it so hard to find peer reviewers?

Because invitation lists are built from visible senior authors, the same reviewers are asked repeatedly and saturate, while a large pool of capable early-career reviewers is never approached. Publons recorded review invitations growing 9.8% year-on-year against 4.9% annual growth in accepted articles, and 75% of editors in its 2016 survey called finding reviewers the hardest part of the job.

How many invitations does it take to find one reviewer?

An average of 2.4 in 2017, up from 1.9 in 2013, according to Publons’ Global State of Peer Review (2018) using ScholarOne data across all fields; the same report projected 3.6 by 2025 on the existing trend. The ratio differs by field and journal, and targeting and invitation quality move it more than most editors expect.

Where should an editor look for reviewer names?

Beyond the corresponding authors and the board list: CRediT contributor statements, ORCID records on the byline, the methods sections of the manuscript’s own citations, declined-invitation suggestions, and declared co-reviewers already in your database. Every one of those reaches people the corresponding-author metadata hides, which is where the unused capacity actually sits.

How do I find early-career peer reviewers?

Invite from middle authorships rather than only corresponding authors, support formal credited co-review with trainees, and make the first invitation small and specific. Callaham and Tercier found relative youth — under ten years after finishing training — was one of only two characteristics predicting higher review quality across 2,856 reviews at a single journal.

How do I find reviewers when everyone I invite declines?

Read the decline codes before sending more invitations. Expertise declines mean the search matched on keywords rather than methods and needs rebuilding; workload declines mean a longer deadline or a narrower ask will land; no response at all is usually a subject-line problem. Publons found 70.6% of surveyed reviewers named “article was outside my area of expertise” among their reasons for declining, against 42% naming workload.

How do I find a reviewer in a very small field?

Split the review by competence rather than by paper. Invite a methods reviewer from outside the subfield who can assess the design and the statistics without knowing the literature, and pair them with one subject expert asked only about novelty and interpretation. Two narrow reviews from two shallow pools beat one impossible search for someone expert in both and conflicted in neither.