Questions around AI disclosure in publications and reports have plagued academics since the release of GenAI in November 2022. Publisher guidelines were very different, causing uncertainty with many authors as to when, and how, AI use must be disclosed. This uncertainty was exacerbated by the increasing feeling of many authors that the guidelines are out of touch with the realities of knowledge production in a world where authors interact with a variety of AI tools while producing publications.
The past twelve months brought many new perspectives on the matter of AI disclosure, as well as new guidelines. Some of the large publishing houses issued new updates of the guidelines, whereas others implemented new arrangements and enforcing mechanisms for dealing with manuscripts, from their expectations from the author through the reviewing and editing process.
The various guidelines are sometimes specific to a journal, with differences even amongst journals from the same publishing house (such as Elsevier) under a broader umbrella of guidelines. Authors submitting manuscripts should always ensure that they consult the latest guidelines of the specific journal to which they make a submission.
Publications in the course of 2026 clearly indicated that there is a growing gap between researchers’ use of AI writing tools and the actual declaration or disclosure of the use of AI. It stands to reason that publishing houses are taking steps to move towards more clarity for researchers on when and how the use of AI tools should disclosed.
The typical question from authors might be: Do the new guidelines amount to a softening or a hardening of the rules regarding the use of AI-tools and disclosure? This binary choice is unfortunately not appropriate to cover the nature of the changes or the (sometimes silent) confirmation of earlier positions from the various publishers. The changes, or the implicit confirmations, rather require authors to be more cautious about the inclusion of AI-generated content/text into the final manuscripts and to be more explicit and structured in the nature of the disclosure of the use of AI tools. This can also be described as more permissive in allowing the use of AI tools yet also being stricter regarding the when and how of disclosure.
The five major publishing houses, which cover more than half of the high-quality scholarly journals, are now taking positions that give expression to the following:
- AI can (still) not be acknowledged and credited as an author. This has implications not only in the formal list of authors, but also for other statements below.
- In most cases, there is a clear recognition and acceptance of the fact that AI tools are increasingly used by authors in executing their research and in preparing their manuscripts. In fact, in its latest guidance, Elsevier shifted to a positive evaluation of the contribution AI tools can make to the research process for journal articles.
- All use of GenAI tools in the development of a manuscript, except when used for language correction and polishing and for style change, must be disclosed. This includes using GenAI for ideation, for developing syntheses of literature, and for generating research hypotheses and AI drafts. To enhance transparency, the disclosure must be within the text and for some journals also in a separate document (mostly in a paragraph preceding the list of references).
- All content developed by GenAI should be meticulously checked for accuracy and relevance by the author. Even so, content generated by GenAI should be disclosed as per relevant section or by way of a general statement. The disclosure relates to the origin of the text, not to whether the text was evaluated and edited by the author in a subsequent stage (together with the required statement that the author takes full responsibility for the text). References to publications should be checked for existence, for correctness, or as to whether they do in fact support the GenAI statements based on them.
- Authors might have to ask different questions in relation to the matter of disclosure. The spontaneous question of the author might be ‘how much of an AI-generated draft can survive, after checking for accuracy, into the final text?’ This is an invalid question in view of the new (and confirmed) regulations. The core question relating to disclosure is: ‘What information about the origin of the text, or the approach in producing the text, should the reader or reviewer have in order to understand how the research was conducted, or to assess the reliability or validity of the published work?’ If the default in reading a journal article is that a position or approach originates from the author, any other source for a position or approach must be indicated; if it is not a published source (to be cited), but produced by GenAI, the reader should know about it.
- Disclosure must be according to formats described by the journal. Mostly, the following is required: Name of the tool (in the case of chatbots: brand and model); date of use; and purpose of use.
The new focus on disclosure might seem to be an unnecessary procedural and accountability burden on authors. However, it is important to understand that it necessary to assure the integrity of research across the whole value chain, from the initial actions of the author through the task of the reviewer up to the reader.
Walter Claassen
SFA Lead: Digital Transformation
7 August 2026
