The Core Requirement: Why Disclosure Matters Now More Than Ever
The question of whether and how to disclose AI use in manuscripts has moved from a niche ethical discussion to a mainstream editorial requirement across scientific publishing. By September 2026, the majority of major publishers including Elsevier, Springer Nature, Wiley, and the American Psychological Association have formalized policies that explicitly address generative AI tools, and the expectations continue to tighten. A Nature survey conducted in the preceding years found that researchers remain split on whether AI should be permitted to write sections of journal papers, reflecting genuine disagreement within the community rather than a settled consensus. The CDC and UNC Health have both published guidance noting that transparency about AI assistance protects research integrity and helps readers evaluate the provenance of scientific claims. Failure to disclose AI involvement can trigger retractions, editorial expressions of concern, or civil investigative actions, as the FTC's inquiry into OpenAI demonstrated that regulatory bodies are paying close attention to how AI-generated content enters public knowledge streams. The fundamental principle is straightforward: if a human reader cannot reasonably distinguish between human-authored and machine-generated text, the author has an obligation to say so.
Also worth reading: How Should Authors Disclose AI Assistance in Books and Manuscripts? · How to audit AI book manuscripts for quality, bias, and publication readiness? · Why Is AI Publishing Consultant storywriter.pro the Answer Publishers Aren't Ready For?
How Disclosure Policies Have Evolved Since 2023
Disclosure expectations have shifted dramatically since early 2023, when many journals lacked explicit AI policies and authors operated in a gray zone. The Register reported in March 2023 on a leaked unpublished manuscript challenging Google's AlphaChip claims, illustrating how undisclosed AI use in research workflows can distort competitive claims and peer review processes. By mid-2023, journals including those published by the Association for Psychological Science had begun issuing editorials weighing the risks and benefits of AI-assisted writing, and by 2024 the Committee on Publication Ethics and the International Committee of Medical Journal Editors had released updated guidance. The scientist.com documented cases where AI policies failed to reduce undisclosed use, revealing that formal rules alone do not solve the problem without enforcement mechanisms. The Conversation summarized what various academic guidelines say about AI use, noting wide variation in specificity and enforceability. In 2025 and 2026, publishers have moved toward mandatory disclosure fields in submission systems, requiring authors to identify which sections of a manuscript were generated or substantially edited by AI and which tool was used. This evolution reflects a recognition that the replication crisis is compounded by opaque AI use, since incomplete disclosure of methods and tools undermines reproducibility and auditing.
Practical Steps for Complying with Journal AI Disclosure Rules
Authors preparing manuscripts for submission should begin by checking the specific AI policy of each target journal, because requirements vary even within the same publisher group. Most journals now require authors to state in a dedicated section or cover letter whether generative AI tools were used during writing, editing, figure generation, or data analysis, and to name the tool and version. The UNC Health research highlights that best practices include keeping a log of all AI interactions, saving prompts and outputs, and noting which portions of the text were drafted by machine versus human authors. Authors should never list an AI tool as a co-author, as this contradicts the standards set by most publishers and research institutions, though some journals require acknowledging AI assistance in the acknowledgments section. When AI is used for literature review or data extraction, the methodology section should describe the tool and the validation steps taken to verify accuracy. The CDC guidance emphasizes that disclosure should extend to supplementary materials and preprints, not just the final published version, because readers may encounter earlier drafts. A practical workflow is to draft normally, then run a disclosure check before submission, flagging every passage that contains AI-generated phrasing or structure and deciding whether to rewrite, cite, or disclose it.
Comparison Table: Disclosure Approaches Across Major Publishers
| Publisher / Policy | Disclosure Required | AI Listed as Co-Author | Specific Tool Naming | Enforcement Mechanism |
|---|---|---|---|---|
| Elsevier (2024-2026) | Mandatory in cover letter | No | Required | Editorial screening, retraction |
| Springer Nature | Mandatory in methods | No | Recommended | Plagiarism + AI detection tools |
| Wiley | Mandatory in acknowledgments | No | Required | Peer review flagging |
| APA Journals | Mandatory in methods | No | Required | Editor judgment, correction |
| Nature Portfolio | Mandatory in methods | No | Required | Editorial screening, editorial expression of concern |
The most frequent error is assuming that minor AI assistance does not need disclosure, when in fact even light editing or phrasing suggestions can cross the line into undisclosed use if the resulting text is substantially machine-generated. Another mistake is vague language such as 'AI tools were used' without specifying which tool, which version, and what it was used for, leaving readers and reviewers unable to assess the contribution. Some authors attempt to circumvent detection by paraphrasing AI output extensively, but this does not satisfy disclosure requirements and may still constitute undisclosed use if the structure and argumentation originate from the machine. The Frontiers study on student AI disclosure practices found that stigma around AI use leads many researchers to underreport, a pattern that persists in professional settings and creates a false baseline of 'purely human' writing. A related error is failing to disclose AI use in data analysis or figure creation, focusing only on text generation when the policy covers all manuscript components. Finally, authors sometimes disclose AI use only after being questioned, rather than proactively, which damages trust and can trigger investigation even if the use was technically permissible.
When to Disclose and When Disclosure Is Not Needed
Authors should disclose any use of generative AI for text generation, editing, summarization, code writing, figure creation, or data interpretation, regardless of the extent, because the default assumption in 2026 is that undisclosed use is a violation. Disclosure is not required for basic spell-checking, grammar correction tools that do not generate new text, or reference management software, as these are considered standard research tools rather than generative AI. However, the boundary is blurring: some AI-powered reference checkers now generate summary sentences about papers, which may cross into disclosure territory. The Times Higher Education editorial argued that the community must set clear rules for AI use in writing and peer review, distinguishing between tools that assist and those that substitute for human judgment. If an author is unsure whether a particular use case requires disclosure, the safest course is to disclose it, because over-disclosure carries less risk than under-disclosure in the current enforcement environment. The Nature survey showing researchers split on AI writing suggests that the community has not yet reached consensus on what constitutes acceptable assistance, making individual judgment and institutional policy review essential.
Cost, Pricing, and Access Considerations for AI Disclosure Tools
Most journal submission systems do not charge extra for AI disclosure fields, but authors may need to invest in plagiarism and AI-detection screening tools to verify their own compliance before submission. Turnitin, iThenticate, and GPTZero offer institutional licenses that range from free for university-affiliated researchers to several hundred dollars per year for independent scholars. Some publishers provide free AI-detection checks as part of the submission process, while others rely on manual editorial review. The cost of non-compliance is far higher: retractions, damaged reputation, and potential funding consequences can exceed thousands of dollars in lost opportunities. For authors using AI tools like ChatGPT, Claude, or Gemini, the subscription costs range from zero for limited free tiers to twenty dollars per month for premium access, but these are separate from disclosure obligations. The key financial consideration is that disclosure itself is free, but the time required to document AI use, rewrite flagged sections, and respond to editorial queries can add days or weeks to the publication timeline.
Looking Ahead: What Will Change by 2027
The trajectory of AI disclosure policy suggests that by 2027, most journals will require not just disclosure of AI use but also documentation of the specific prompts, model versions, and human editing steps involved in manuscript preparation. The CONSORTIA initiative and similar standards bodies are working on frameworks for responsible research assessment that include generative AI use and disclosure as core components, alongside research security and Indigenous data governance. The replication crisis literature highlights that incomplete disclosure of training details and methodology undermines reproducibility, and AI disclosure is becoming part of that broader transparency agenda. Authors who build disclosure habits now will be ahead of the curve as policies tighten, while those who treat disclosure as optional risk being caught in enforcement sweeps that several publishers have signaled are coming. The critical takeaway is that AI manuscript disclosure is not a one-time checkbox but an ongoing practice that must be integrated into every stage of the research and writing workflow.