CV Google Scholar MSU

Part of AI and Research Practice.

Journal AI policies

What journals actually say about generative AI, for authors and for reviewers, in their own words, with the page each quote came from and the date it was read.

Policies as read on —
Snapshots held for every source
Journal missing? Suggest one →

These quotes are not current, they are dated. Each one is what the cited page said on the date shown beside it. Publishers rewrite these pages without notice and without version numbers. Three moved domain during collection. Follow the live link to see what the page says today, and check it against the archived copy linked beside it. A local snapshot and a SHA-256 hash is held for every source and is available on request, so a reader who wants to verify that the quoted text is what was read can do so without taking it on trust.

Every quote carries whether it was checked. Each one is matched word by word against the stored copy of the page it came from. Not found does not mean the quote is wrong, it means no source held here contains it, which so far has almost always meant the page was never captured or the recorded text is a fragment rather than a full sentence. As of

Two kinds of permission, and one kind of silence. A journal permits an activity explicitly when a governing text names it as allowed, and implicitly when it is allowed without being named, either through a general permission whose stated scope covers it or through a rule that its use be disclosed. No mention means no governing text addresses the activity at all.

All journals are coded against the same scheme: fourteen author activities, eight reviewer tasks, and whether a reviewer may upload the manuscript. Recoded in October 2026 under one set of written rules. Every cell quotes the sentence it rests on, from the journal's own pages, its publisher's policy, guidance pages and FAQs, its society, or a cross-journal policy it adopted. A general permission or ban counts only for the activities inside its stated scope. Where documents disagree, a prohibition anywhere wins. Every reading is recorded once in a shared table and applied everywhere, and a blind re-code of a stratified sample of 60 cells agrees on 88 percent.