Generative AI has quickly become part of the academic publishing landscape. For us, this creates a difficult challenge: how should journals respond when authors use AI to support manuscript preparation?
A recent article by Brendal Aformeziem on the LSE Impact Blog argues that we should be careful not to assume bad intentions when AI is detected in a submission. Instead, growing evidence suggests that many researchers are turning to AI because they face structural disadvantages that the publishing system has not yet resolved.
The researchers most likely to use AI are often those who need the most support. Studies cited in the article show that adoption is highest among authors from non-English-speaking countries, early-career researchers and scholars at less well-resourced institutions. For these authors, AI tools can help overcome the additional burden of writing for journals that operate primarily in English.
Importantly, the evidence suggests that AI is not removing the barriers faced by non-native English-speaking researchers. Research analysing nearly 80,000 peer reviews at ICLR found that biases linked to language and author background persisted after the arrival of ChatGPT. In some cases, reviewers have simply shifted their attention to language patterns they associate with AI-generated text, creating new forms of suspicion rather than reducing existing prejudice.
This has important implications for editorial teams. The presence of AI-generated phrasing should not automatically be treated as evidence of misconduct or an attempt to deceive. For many authors, AI is functioning as an accessibility and language-support tool rather than a shortcut to avoid scholarly work.
While many journals now require authors to declare AI use, disclosure rates remain much lower than estimated usage rates. Researchers may be reluctant to disclose because they fear stigma, negative judgement or what some scholars describe as “AI shaming”. These concerns may be particularly acute for authors working in under-resourced environments who lack access to institutional guidance and support.
For Managing Editors, this points to a need for nuance. Editorial policies should distinguish between acceptable language assistance and inappropriate uses of AI. Equally important, editorial staff should avoid treating AI detection as a proxy for author quality, integrity or competence. AI use alone tells us very little about the rigour, originality or validity of the underlying research.
As journals develop policies and workflows around AI, empathy and context will be essential. Rather than asking, “Why did this author use AI?”, editorial teams may be better served by asking, “What challenge was this author trying to overcome?” By approaching AI use as a potential signal of need rather than automatically viewing it as a signal of misconduct, journals can support both research integrity and greater inclusion in scholarly publishing. While journal AI policies are often set at an organisational level and may be outside the control of editorial staff, editorial assistants and managing editors can still play an important role by approaching suspected AI use with an open mind, recognising that for many authors these tools may represent a response to linguistic, institutional or resource-related barriers rather than an attempt to circumvent scholarly standards.
References
Aformeziem, B. (2026). AI promised to democratise academic publishing – the evidence says otherwise. LSE Impact Blog, London School of Economics and Political Science, 23 June 2026. Available at: LSE Impact Blog article [blogs.lse.ac.uk]
Lepp, H. and Smith, D.S. (2025).“You Cannot Sound Like GPT”: Signs of language discrimination and resistance in computer science publishing. Proceedings of the ACM Conference on Fairness, Accountability, and Transparency (FAccT ’25). Available via arXiv: View paper
Y. He, & Y. Bu, Academic journals’ AI policies fail to curb the surge in AI-assisted academic writing, Proc. Natl. Acad. Sci. U.S.A. 123 (9) e2526734123, https://doi.org/10.1073/pnas.2526734123 (2026).



