[{"data":1,"prerenderedAt":460},["ShallowReactive",2],{"blog-post-/blog/ai-in-peer-review":3},{"id":4,"title":5,"author":6,"body":7,"canonicalUrl":448,"category":449,"coverImage":450,"date":451,"description":452,"extension":453,"meta":454,"navigation":455,"path":456,"published":455,"seo":457,"stem":458,"__hash__":459},"blog/blog/ai-in-peer-review.md","AI in Peer Review: What's Allowed and What's Coming (2026)","Thesify Team",{"type":8,"value":9,"toc":431},"minimark",[10,18,21,26,29,35,52,107,110,114,117,125,128,134,140,143,147,159,165,168,222,225,229,232,239,273,276,279,283,286,289,321,324,333,348,351,355,358,364,370,376,382,386,391,394,398,405,409,412,416,419],[11,12,13,17],"p",{},[14,15,16],"strong",{},"AI in peer review"," means two different things. Reviewers using AI on someone else's manuscript is restricted or prohibited by most major publishers, chiefly on confidentiality grounds. Authors running an AI review on their own draft before submission involves no reviewer-confidentiality breach, and it is a fast-growing legitimate use of the technology in publishing.",[11,19,20],{},"Most of the confusion around this topic comes from collapsing those two situations into one. This post separates them, maps the policy landscape as it stands in mid-2026, and explains what authors can do today without breaking any rules.",[22,23,25],"h2",{"id":24},"ai-peer-review-means-two-very-different-things","\"AI peer review\" means two very different things",[11,27,28],{},"When someone asks whether AI peer review is allowed, the first question back should be: whose manuscript, and who is running the AI?",[11,30,31,34],{},[14,32,33],{},"Situation one: a reviewer uses AI on a manuscript they were asked to evaluate."," The manuscript is confidential. It was shared with the reviewer under an explicit or implicit agreement that it goes no further. Pasting it into a public chatbot shares unpublished work with a third-party system the authors never consented to. That is why publishers treat this as a confidentiality problem first and a quality problem second. There is also an accountability problem: the journal invited a specific expert's judgment, not a language model's.",[11,36,37,40,41,45,46,51],{},[14,38,39],{},"Situation two: an author runs an AI review on their own manuscript before submitting."," There is no reviewer-confidentiality breach, because the work is theirs and nothing has been shared with a journal yet, though you (and your co-authors) should still check how any tool handles your data before uploading an unpublished draft. The relevant policy question for authors is a different one entirely: whether and how to disclose AI assistance in ",[42,43,44],"em",{},"writing"," the manuscript, which most journals now address in their authorship policies. Using an ",[47,48,50],"a",{"href":49},"/ai-peer-review","AI peer review tool"," to critique your own draft is not the same policy area as using AI to write it, and it is not the same policy area as reviewer conduct.",[53,54,55,70],"table",{},[56,57,58],"thead",{},[59,60,61,64,67],"tr",{},[62,63],"th",{},[62,65,66],{},"Reviewer-side AI use",[62,68,69],{},"Author-side AI review",[71,72,73,85,96],"tbody",{},[59,74,75,79,82],{},[76,77,78],"td",{},"Whose manuscript",[76,80,81],{},"Someone else's, held in confidence",[76,83,84],{},"Your own",[59,86,87,90,93],{},[76,88,89],{},"Core policy issue",[76,91,92],{},"Confidentiality and reviewer accountability",[76,94,95],{},"Disclosure of AI use in writing (a separate policy area)",[59,97,98,101,104],{},[76,99,100],{},"Typical stance, mid-2026",[76,102,103],{},"Prohibited or heavily restricted by major publishers",[76,105,106],{},"No reviewer-confidentiality issue; nothing is shared with any journal",[11,108,109],{},"Keep this distinction in mind for the rest of the post. Everything that follows sorts into one column or the other.",[22,111,113],{"id":112},"whats-actually-happening-ai-text-is-showing-up-in-review-reports","What's actually happening: AI text is showing up in review reports",[11,115,116],{},"The policy crackdown did not come out of nowhere. It followed evidence that reviewers were already using generative AI, quietly and at scale.",[11,118,119,120,124],{},"Researchers who analyzed review reports at major AI and machine-learning conferences estimated that between 6.5% and 16.9% of the review text at ICLR 2024, NeurIPS 2023, CoRL 2023, and EMNLP 2023 could have been substantially modified by large language models (Liang et al., ICML 2024). Authors have reported the pattern independently: reviews that read as fluent but generic, praise the paper's \"comprehensive approach\" without naming a method, or raise concerns that do not match anything in the manuscript. If you have read enough real reviewer feedback, the difference is recognizable; our guide to ",[47,121,123],{"href":122},"/blog/reviewer-comments-examples","reviewer comments examples"," shows what substantive human feedback patterns look like.",[11,126,127],{},"Publishers responded on two fronts.",[11,129,130,133],{},[14,131,132],{},"Restriction."," Elsevier's reviewer guidance instructs reviewers not to upload a submitted manuscript, or any part of it, into an AI tool; since a June 2026 policy update, AI may be used only in a supportive capacity (for example to improve the language and structure of the review report), with confidentiality maintained and the use disclosed. Springer Nature asks peer reviewers not to upload manuscripts into generative AI tools. In 2023, the NIH prohibited the use of generative AI in its grant peer review process (notice NOT-OD-23-149), treating it as a breach of confidentiality. COPE has also published discussion and guidance on AI in peer review that many society journals lean on.",[11,135,136,139],{},[14,137,138],{},"Adoption: on the publisher's own terms."," At the same time, publishers and venues have been piloting AI-assisted integrity and completeness checks that run inside their own infrastructure rather than in a public chatbot: image-integrity screening (Proofig at the Science family of journals, Springer Nature's SnappShot), AI-generated-text detection (Springer Nature's Geppetto), and cross-publisher screening through the STM Integrity Hub. Major machine-learning conferences have gone further: ICLR 2025 piloted an AI feedback agent inside its review workflow: one that critiques reviewer-written reviews rather than writing them, with every decision staying human.",[11,141,142],{},"The pattern is consistent: publishers are not anti-AI. They are against uncontrolled AI use on confidential material, while building controlled AI checks of their own.",[22,144,146],{"id":145},"can-reviewers-use-ai-the-policy-map-in-mid-2026","Can reviewers use AI? The policy map in mid-2026",[11,148,149,150,153,154,158],{},"If you review manuscripts yourself, the safe summary is: ",[14,151,152],{},"assume you may not put any part of a confidential manuscript into a generative AI tool unless the journal explicitly says otherwise."," Some policies distinguish between uploading manuscript content (prohibited) and using AI to polish the language of your own review text without pasting in manuscript material: Elsevier and Wiley tolerate this with disclosure, while the NIH and some conferences prohibit AI use in review entirely. When in doubt, ask the editor. Our guide on ",[47,155,157],{"href":156},"/blog/how-to-peer-review-a-paper","how to peer review a paper"," covers the underlying obligations that these AI policies are protecting.",[11,160,161,162,164],{},"Authors have a mirror-image set of obligations: most journals now require disclosure of generative AI use in drafting the manuscript (per the ICMJE recommendations), and virtually all prohibit listing an AI tool as an author. Note again that these rules govern AI as a ",[42,163,44],{}," tool. Running an AI critique of your own finished draft, then revising it yourself, generally sits outside both of these policy buckets, though disclosure wording varies by journal, so check how your target journal defines AI assistance.",[11,166,167],{},"Here is the landscape by publisher type. Treat this as orientation, not legal advice: policies are moving, and the current policy page always wins.",[53,169,170,180],{},[56,171,172],{},[59,173,174,177],{},[62,175,176],{},"Venue type",[62,178,179],{},"Typical stance on reviewers using generative AI (mid-2026)",[71,181,182,190,198,206,214],{},[59,183,184,187],{},[76,185,186],{},"Large commercial publishers (e.g., Elsevier, Springer Nature, Wiley)",[76,188,189],{},"Prohibit uploading manuscripts to generative AI tools; reviewer remains fully accountable for the review",[59,191,192,195],{},[76,193,194],{},"Funders",[76,196,197],{},"Generative AI prohibited in grant peer review at the NIH; other funders vary; check individually",[59,199,200,203],{},[76,201,202],{},"AI/ML conferences",[76,204,205],{},"Mixed: some prohibit LLM use in review writing entirely (e.g., CVPR), some run sanctioned AI-assistance pilots inside the review platform (e.g., ICLR, NeurIPS)",[59,207,208,211],{},[76,209,210],{},"Society journals",[76,212,213],{},"Often follow COPE guidance; specifics vary widely by journal",[59,215,216,219],{},[76,217,218],{},"Preprint and open-review platforms",[76,220,221],{},"More experimentation; confidentiality concerns are weaker for public preprints, but platform rules vary",[11,223,224],{},"Two practical takeaways. If you are a reviewer: the burden is on you to check the venue's policy before any AI touches the manuscript. If you are an author: you cannot control whether your reviewer quietly uses a chatbot, but you can control whether your paper survives a generic, surface-level reading, which is an argument for making your contribution impossible to miss.",[22,226,228],{"id":227},"why-peer-review-is-strained-and-where-ai-legitimately-helps","Why peer review is strained, and where AI legitimately helps",[11,230,231],{},"Global submission volumes have grown for years while the pool of willing reviewers has not kept pace. Editors commonly report sending more invitations to secure the same number of reviews, and review turnaround has stretched at many journals. Reviewing is unpaid, largely uncredited work performed by the same overcommitted researchers who are writing their own papers. That mismatch, more than any fascination with the technology, is what pulls AI into peer review.",[11,233,234,235,238],{},"So where does AI genuinely help without violating anyone's trust? The pattern across current pilots is that AI performs best on ",[14,236,237],{},"checkable, mechanical dimensions"," of a manuscript, and worst on judgment calls:",[240,241,242,249,255,261,267],"ul",{},[243,244,245,248],"li",{},[14,246,247],{},"Reporting completeness."," Does the methods section contain what a checklist such as CONSORT or ARRIVE expects for this study type? SciScore, deployed at American Heart Association journals, runs exactly this kind of check.",[243,250,251,254],{},[14,252,253],{},"Statistics sanity checks."," Are test statistics, degrees of freedom, and p-values internally consistent? statcheck, run at triage by the Journal of Experimental Social Psychology, automates this.",[243,256,257,260],{},[14,258,259],{},"Image integrity."," Duplicated or manipulated figure panels, which specialized screening tools such as Proofig flag for human follow-up.",[243,262,263,266],{},[14,264,265],{},"Reference and citation hygiene."," Retracted or nonexistent references.",[243,268,269,272],{},[14,270,271],{},"Structure and clarity."," Whether a reader can locate the contribution, the design, and the limitations where they expect to find them.",[11,274,275],{},"What stays human: judging novelty, significance, and whether the interpretation is warranted by the evidence. No major publisher or funder policy hands those calls to a machine.",[11,277,278],{},"Notice something about that list of mechanical checks. Every item on it is also a common reason manuscripts get negative reviews. Which brings us to the use of AI that is unambiguously allowed.",[22,280,282],{"id":281},"the-author-side-playbook-ai-review-of-your-own-manuscript","The author-side playbook: AI review of your own manuscript",[11,284,285],{},"Running an AI review on your own draft before submission is the compliant version of AI peer review. No reviewer obligation is triggered, nothing is shared with any journal, and you make all revision decisions yourself. Done well, it compresses a cycle that often costs months of review time: find the weaknesses a reviewer would find, but find them while you can still fix them quietly.",[11,287,288],{},"A good pre-submission AI review is not a grammar check. It should interrogate the same dimensions a human reviewer will:",[240,290,291,297,303,309,315],{},[243,292,293,296],{},[14,294,295],{},"Contribution clarity."," Can a reader state, after the introduction, what is new and why it matters? Vague contribution framing is one of the most commonly cited reasons for rejection at selective journals.",[243,298,299,302],{},[14,300,301],{},"Methods reporting."," Are sample sizes, inclusion criteria, statistical procedures, and materials described completely enough to evaluate, and to reproduce?",[243,304,305,308],{},[14,306,307],{},"Unsupported claims."," Does every conclusion trace back to a result actually shown? Reviewers are ruthless about claims that outrun the data.",[243,310,311,314],{},[14,312,313],{},"Structure and argument flow."," Do sections deliver what their headings promise? Does the discussion answer the questions the introduction raised?",[243,316,317,320],{},[14,318,319],{},"Presentation issues."," Figure–text mismatches, undefined abbreviations, inconsistent terminology: small flags that erode reviewer confidence in everything else.",[11,322,323],{},"Here is a composite example, based on common feedback patterns, of the kind of flag a substantive pre-submission review should raise:",[325,326,327],"blockquote",{},[11,328,329,332],{},[14,330,331],{},"Unsupported claim (Discussion, para 2)."," The manuscript states the intervention \"reduces symptom burden in the broader patient population,\" but the study enrolled a single-center sample with narrow inclusion criteria and reports no subgroup or generalizability analysis. A reviewer is likely to ask you to restrict this claim to the studied population or justify the generalization.",[11,334,335,336,342,343,347],{},"That is the level of specificity to demand from any tool you use. A pre-submission review with ",[47,337,341],{"href":338,"rel":339},"https://reviewer.thesify.ai",[340],"nofollow","Thesify Reviewer"," applies this reviewer's-eye reading to your manuscript (contribution framing, methodological gaps, unsupported claims, structure) before an editor ever sees it. For a comparison of the options in this category and what separates a real manuscript review from a summarizer with opinions, see our roundup of the ",[47,344,346],{"href":345},"/blog/best-ai-manuscript-review-tools","best AI manuscript review tools",".",[11,349,350],{},"One discipline note for biomedical authors: your field has the most developed reporting standards and the most active integrity screening. That cuts both ways. Journals are more likely to run automated checks on your submission, and a pre-submission review that covers reporting completeness pays off most in exactly these fields.",[22,352,354],{"id":353},"what-to-expect-next-our-read-on-where-this-goes","What to expect next: our read on where this goes",[11,356,357],{},"What follows is analysis, not established fact. But the trajectory through mid-2026 points in a consistent direction.",[11,359,360,363],{},[14,361,362],{},"Disclosure norms will standardize."," Right now, AI-use disclosure requirements differ journal by journal. Expect convergence toward a standard disclosure statement covering both authors and reviewers, the way conflict-of-interest statements standardized over the past two decades.",[11,365,366,369],{},[14,367,368],{},"Journal-side AI screening will become routine."," Plagiarism checks went from novelty to default within a decade. Integrity and completeness screening looks set to follow the same path: every submission passing through automated checks before a human editor reads it. Authors should expect their manuscript to be read by software first.",[11,371,372,375],{},[14,373,374],{},"Hybrid human-plus-AI review models will be piloted openly."," Instead of reviewers covertly pasting manuscripts into chatbots, expect journals to offer sanctioned AI assistance inside their own review platforms: pre-drafted completeness reports that the human reviewer verifies, corrects, and takes responsibility for. The accountability stays human; the mechanical labor does not.",[11,377,378,381],{},[14,379,380],{},"The author-side shift may matter most."," If authors routinely run rigorous AI review before submission, the average submitted manuscript gets better, desk rejections fall, and human reviewers spend their scarce attention on science rather than on catching missing sample sizes. Peer review's strain is a supply-and-demand problem, and pre-submission review is one of the few levers that reduces demand.",[22,383,385],{"id":384},"faq","FAQ",[387,388,390],"h3",{"id":389},"can-peer-reviewers-use-chatgpt-to-review-a-paper","Can peer reviewers use ChatGPT to review a paper?",[11,392,393],{},"Generally no. Major publishers, including Elsevier and Springer Nature, direct reviewers not to upload manuscripts into generative AI tools, because doing so breaches confidentiality. Some venues tolerate AI help with the reviewer's own prose if no manuscript content is shared, but the reviewer remains fully accountable. Check the specific journal's policy before using any AI tool in a review.",[387,395,397],{"id":396},"is-it-allowed-to-use-ai-to-review-my-own-paper-before-submission","Is it allowed to use AI to review my own paper before submission?",[11,399,400,401,404],{},"Yes. Reviewing your own manuscript with an AI tool raises no confidentiality issue, because the work is yours and nothing is shared with a journal. The separate question is disclosure: if AI helped ",[42,402,403],{},"write"," the manuscript, most journals require you to say so. Using AI to critique a draft you then revise yourself is not covered by reviewer-conduct rules.",[387,406,408],{"id":407},"do-journals-use-ai-to-screen-submissions","Do journals use AI to screen submissions?",[11,410,411],{},"Increasingly, yes. Many publishers run automated checks at submission: plagiarism detection is long established, and deployments now extend to image integrity, statistical consistency, and reporting completeness. These tools flag issues for human editors; they do not make accept/reject decisions on their own.",[387,413,415],{"id":414},"will-ai-replace-peer-review","Will AI replace peer review?",[11,417,418],{},"Not in any near-term scenario we find credible. Judgments about novelty, significance, and whether conclusions are warranted remain human, and current policies keep them that way. What AI is absorbing is the mechanical layer: completeness, consistency, and integrity checks. The likely future is fewer avoidable flaws reaching reviewers, not fewer reviewers.",[325,420,421],{},[11,422,423,426,427],{},[14,424,425],{},"Before you submit, see your paper the way reviewers will.","\nThesify Reviewer runs a pre-submission review of your manuscript. It flags the issues peer reviewers are most likely to raise, from unclear contributions to methodological gaps, before a journal ever sees it.\n",[47,428,430],{"href":338,"rel":429},[340],"Try Thesify Reviewer",{"title":432,"searchDepth":433,"depth":433,"links":434},"",2,[435,436,437,438,439,440,441],{"id":24,"depth":433,"text":25},{"id":112,"depth":433,"text":113},{"id":145,"depth":433,"text":146},{"id":227,"depth":433,"text":228},{"id":281,"depth":433,"text":282},{"id":353,"depth":433,"text":354},{"id":384,"depth":433,"text":385,"children":442},[443,445,446,447],{"id":389,"depth":444,"text":390},3,{"id":396,"depth":444,"text":397},{"id":407,"depth":444,"text":408},{"id":414,"depth":444,"text":415},null,"AI Use Policies","/blog/ai-in-peer-review/cover.jpg","2026-07-28","AI in peer review means two very different things. What reviewers can and can't do in 2026, how publishers are responding, and the one compliant use.","md",{},true,"/blog/ai-in-peer-review",{"title":5,"description":452},"blog/ai-in-peer-review","zIfbzSa3H2VAyygATO_00WIazJktcGDbEvk_GTLNZVA",1787302047654]