Direct comparison
Reviewer AI Confidentiality: 4 Publishers + NIH
Elsevier, Wiley, Springer, and NIH (NOT-OD-23-149) bar reviewer AI manuscript uploads — see where each policy diverges on disclosure.
Ask about Reviewer AI Confidentiality: 4 Publishers + NIH
Answers are drawn from this comparison and the rest of the CASRAI corpus, with a link to every source.
Answers are AI-generated from CASRAI’s own published pages and can be wrong, so check the linked sources before relying on one; your question is logged without personal data — never sold, never used to train a third-party model — to show us what CASRAI is missing, so please do not type personal or confidential details. How we use this
How do Elsevier, Springer Nature, Wiley, Taylor & Francis, NIH (grant peer review) compare side by side?
The table below compares Elsevier, Springer Nature, Wiley, Taylor & Francis, NIH (grant peer review) across 7 procurement-relevant dimensions, from uploading the manuscript/application to an ai tool through source / policy last checked.
Side-by-side comparison
| Dimension | Elsevier | Springer Nature | Wiley | Taylor & Francis | NIH (grant peer review) |
|---|---|---|---|---|---|
| Uploading the manuscript/application to an AI tool | Prohibited — manuscript or any part of it may not be uploaded to a generative AI tool; risk to confidentiality/proprietary rights and, where PII is present, data-privacy law. | Prohibited — reviewers are asked not to upload manuscripts into generative AI tools. | Prohibited — no uploading manuscripts or parts (incl. figures/tables) into tools such as ChatGPT; risk to confidentiality, privacy, and copyright. | Prohibited — must not upload unpublished manuscripts or project proposals, including associated files, images, or information. | Prohibited — uploading or sharing application/proposal/critique content to online generative AI tools violates NIH peer-review confidentiality and integrity requirements (NOT-OD-23-149). |
| Using AI to analyze, summarize, or draft the substantive review | Not permitted — critical thinking and original assessment are described as outside the scope of the technology; the reviewer remains accountable for the review's content. | Not permitted — reviewers must independently evaluate; policy cites AI's outdated knowledge and risk of biased or false output. | Not permitted — human judgment is essential; evaluation cannot be outsourced to AI. | Not permitted — reviewers 'should not be using Generative AI for analysis or to summarise submitted articles.' | Banned outright for 'analyzing and formulating peer review critiques' — no supportive-use carve-out, only an accessibility-technology exception. |
| Permitted supportive AI use | Improving the language/structure of the review report, or a background literature search — only with a private, non-data-retaining AI tool and human oversight (per the June 2026 policy update). | Not detailed in the public policy beyond the general caution; any AI support of the evaluation must be declared. | Polishing the wording of the reviewer's own already-written report. | Improving review language. | None — assistive technology for accessibility is the only stated exception. |
| Disclosure to the editor/journal required? | Yes — declare the tool and purpose in the review report (suggested template wording provided); routine spelling/grammar checks are exempted. | Yes — declare the use of such tools transparently in the peer review report whenever AI supported the evaluation. | Yes — disclose to the handling editor if AI was used to polish the report. | Not specified in Taylor & Francis's public reviewer policy — contrast with the other three publishers, each of which names an explicit disclosure step. | N/A — use itself is prohibited; reviewers instead sign a Security, Confidentiality and Nondisclosure Agreement that names the prohibition. |
| Stated rationale | Confidentiality and proprietary rights of the manuscript; data-privacy risk where personally identifiable information is present. | Confidentiality of unpublished material, plus AI's own reliability limitations. | Confidentiality, privacy, and copyright of unpublished work. | Confidentiality, proprietary rights, and data protection. | Peer-review confidentiality and integrity; no guarantee of where uploaded data are sent, saved, viewed, or used. |
| What it governs | Journal manuscript peer review | Journal manuscript peer review | Journal manuscript peer review | Journal manuscript peer review | Grant application and R&D contract proposal peer review (not journal manuscripts) |
| Source / policy last checked | elsevier.com generative-AI review-process and journals policy pages; explicit June 2026 update. Verified 2026-07-09. | springernature.com / springer.com editorial-policies AI pages (undated). Verified 2026-07-09. | authors.wiley.com reviewer confidentiality and AI guidelines pages (undated). Verified 2026-07-09. | taylorandfrancis.com AI policy, corroborated via publisher newsroom announcement (undated policy text; announcement dated). Verified 2026-07-09. | NOT-OD-23-149 (issued 23 June 2023), grants.nih.gov / osp.od.nih.gov. Verified 2026-07-09. |
Common questions
Common questions about Elsevier vs Springer Nature vs Wiley vs Taylor & Francis vs NIH (grant peer review)
Can a peer reviewer paste a manuscript into ChatGPT to help write their review?
+
No, not at any of the organisations compared here. Elsevier, Springer Nature, Wiley, and Taylor & Francis all explicitly prohibit uploading a manuscript, or any part of one, into a generative AI tool, citing confidentiality risk and, in Elsevier's case, data-privacy risk where the manuscript contains personally identifiable information. NIH applies the same prohibition to grant application and contract-proposal peer review.
Is any AI use by reviewers actually allowed?
+
At Elsevier, Springer Nature, and Wiley, yes, but narrowly: reviewers may use AI to polish the wording of a review they have already written, and Elsevier's June 2026 policy also permits a background literature search, using a private, non-data-retaining AI tool. All three require that use to be disclosed. Taylor & Francis permits similar language-polishing use but does not state a disclosure requirement in its public policy. NIH permits no AI use in peer review at all, aside from assistive technology used for accessibility.
Do reviewers have to disclose AI use even if they never uploaded the manuscript?
+
At Elsevier, Springer Nature, and Wiley, yes: any AI assistance used in preparing the review, beyond basic spelling or grammar checking (which Elsevier exempts), must be declared in the review report or to the handling editor. This disclosure obligation is separate from, and in addition to, the manuscript-upload ban.
Can editors reliably detect undisclosed AI use in a review?
+
Not reliably, based on current evidence. A 2026 preprint (Saha et al., accepted at ICML 2026) tested existing AI-text detectors against LLM-polished peer reviews and found they misclassify a meaningful share of genuinely human-and-AI-collaborative reviews as fully AI-generated. That means the disclosure-based model these publishers rely on isn't currently backed by dependable automated detection — a live research problem, not a solved one.
Does this apply to grant reviewers as well as journal peer reviewers?
+
The rules differ by context. The publisher policies compared here (Elsevier, Springer Nature, Wiley, Taylor & Francis) govern journal manuscript peer review. NIH's policy (NOT-OD-23-149) is separate and stricter: it bans generative AI use entirely for NIH scientific peer review of grant applications and R&D contract proposals, and reviewers sign a modified confidentiality agreement naming the prohibition explicitly.
How is this different from CASRAI's guidance on authors using AI to write a manuscript?
+
It covers a different actor and a different question. This page is about what a REVIEWER may do with AI while evaluating someone else's confidential, unpublished manuscript. CASRAI's author-facing generative AI disclosure guidance covers what an AUTHOR may do with AI while drafting their own manuscript for submission — a different confidentiality posture, and a different disclosure mechanism (a declaration published in the article itself, rather than a note to an editor).
