Generative AI is entering academic publishing faster than journals can settle on consistent rules. Publishers are trying to protect trust in research while deciding how authors may use these tools—and how to identify misuse.
A chatbot sentence raises questions
In August 2022, the journal Resources Policy published a study about e-commerce and fossil fuels containing a sentence associated with AI chatbots: “Please note that as an AI language model, I am unable to generate specific tables or conduct tests, so the actual results should be included in the table”. The sentence alerted researchers to possible undisclosed AI use. Elsevier is investigating the incident.
The episode highlights a challenge for editors: AI assistance may leave clues, but many papers will not contain such an obvious sign. Detecting AI-written text is extremely difficult, and no foolproof method has been developed.
Rules are taking shape
Major publishers and journals, including Science, Nature, and Elsevier, have moved to introduce AI policies amid concerns about credibility. These policies typically require authors to disclose AI use and prohibit naming AI systems as authors.
Disclosure rules provide a way to make AI assistance visible, but they do not settle every question. Journals still face the practical problem of assessing whether research is reliable and whether an author has accurately described the role a tool played.
Useful assistance, real risks
AI tools may help researchers make academic writing clearer. Experts say they could also benefit non-native English speakers by improving the quality of their writing and their chances of publication. In one recent experiment, researchers used ChatGPT to produce a passable paper in just an hour.
Those benefits come with significant limitations. Generative AI can fabricate facts and references, repeat biased data, and help disguise plagiarism. A polished manuscript therefore cannot, on its own, establish that its claims or sources are sound.
The risk also extends beyond individual submissions. AI could help “paper mills” sell low-quality AI-assisted research to academics facing “publish or perish” pressure. If such work spreads through journals, it could pollute the research literature and divert attention from legitimate studies.
Images and multimodal tools add complexity
AI-generated images create another concern for research publishers. Tools such as Stable Diffusion and generative features in Photoshop could be used to manipulate or fabricate images. Nature has banned these outright.
Multimodal systems may make review harder as they handle more than text. The source points to Google Deepmind's Gemini as an example on the horizon, with rumors that it may analyze graphs and tables. If those capabilities materialize, AI could process full papers and supplementary material more easily.
Publishers may also begin integrating AI tools into their own workflows. That could change how papers are assessed, but it does not remove the need for careful standards. The challenge is to support legitimate writing assistance while protecting the reliability of published research. Finding that balance will take further trial and error.