AI has been a revolutionary tool for all those who communicate ideas. It has also revolutionised how people access information, and impacted some of the ways that people learn. Going into mainstream use in November 2022 with the rollout of OpenAI’s prompt-based generative text tool Chat GPT, an expanded landscape soon opened into view, incorporating generated speech, imagery, video and more. We all now live in that world, and for publishers it has been quite a challenge. First came confusion with questions such as: ‘why are we receiving so many more submissions?’, ‘surely people cannot use generative AI to write complex academic and technical papers with no human author engaged?’, ‘How can we detect this/how can we tell?’, ‘What kind of policy should we have on generative AI?’, ‘Should we also use generative AI in our editing’?, ‘If we can reliably detect it, should we reject all of it on sight?’, and so on. This article is not a systematic study of the issue at large. It is a focused account of how E-International Relations has answered such questions as those sketched above, with some personal reflections following. That being said, we do not exist in a vacuum and notes and experiences have been shared, distilled, and iterated upon by us as we developed our ‘answers’ or ‘approach’.
Get E-International Relations delivered to your inbox, free. As you sign up, consider becoming a paid subscriber, or make a donation, to support our work.
The first thing we did was rewrite our submissions policy in early 2023 to include the words ‘your thoughts and your words must be your own’, and we continued with our prior editorial patterns based on a 2-step human review of all submissions (growing in number as they were). Our editors became sharper to the AI-like writing forms and ‘tells’ (you come to know it when you see it) and we got rid of those with automated rejections swiftly. However, as time passed, the way authors were using AI, and the tools themselves, evolved into a hybrid zone where you often could not easily tell, and in pieces that were clearly part-generated (in style by an author using AI-generative editing, or in substance) there were sometimes insights worthy of publication and we sometimes decided to roll ahead and publish those if we saw value, but we could not definitively prove AI was used beyond permitted assistive uses.
So, we treaded carefully, certainly not taking a zero-tolerance approach, and we kept thinking, learning, and reflecting. My role as the Editor-in-Chief and Publisher was central to this, as any new policy would need to be written, and overseen, by me (with the support of the team I lead). It is important also to note that every editor at E-International Relations, including myself, volunteers their time. And it was this issue that became one of two hinge points that got us to where we are today.
After a few years of this, I became sick of reading AI text. Other editors felt the same. It didn’t matter anymore if a human started a draft and then had AI-reshape it, or it was something more nefarious with little human involvement. The result was the same. It was boring, lifeless, formulaic. It was not this, but that. One of the joys of being an editor in the ‘before times’ was taking a piece of writing with genuine scholarly insight within it but presented a little poorly (perhaps the author wasn’t a good writer, or were writing in a second language), and digging into it and getting it into shape. That felt like a good use of my time. As an editor, you sharpen your skills, you learn more about the subject you are editing, you become a better writer yourself. It’s a net positive. The age of AI writing takes much of that away. And, as has become clear, it also takes a lot away from the person on the other side, the ‘author’ (though that word often needs to be in scare quotes these days). Now, people are turning off large developmental areas of intellect and skill – and a whole generation is not developing them much at all. That is terrifying.
A second hinge point in this journey came when we moved E-International Relations to Substack. We had been cross posting/mirroring here for a couple of years, but in late July 2026 we decided to fully migrate and make Substack the home for our publication. The change made me rethink the relationship I had as a publisher with our readers. Now, myself and my fellow editors were not just pressing publish and putting something on a website, but we were also sending something to thousands of email inboxes. It was more direct relationship, and it made me think about many things, AI being one of them. Around the same time as we migrated, Substack plugged Pangram (an AI detector) into its back-end publishing system so our editors can scan everything for AI. On top of that, publishers can enable readers to generate their own AI scans (in summary form) of published works when using Substack reader. We decided to enable this function for our readers (going for full transparency), and at the same time to rewrite our AI policy and draw a line in the sand.
So, two hinge points for those TLDR folks. 1. Fatigue with AI writing and 2. A temporal series of events tied to being on Substack led to the rewriting of our submissions guide and our publishing agreement. Both went live in mid-September.
The central through-line to our new policy is zero-tolerance for any AI generated text: “works that appear to be AI generated in whole, or in part, will be rejected on sight.” Of course, we have ways to elaborate on this, and back-end processes to make sure we apply it consistently. Authors must send us an AI declaration with their submission confirming that their words and your thoughts are their own, and AI has not been used to generate any text. We do permit strictly assistive uses of AI (such as light editing, etc.) and these must be declared if used and must not result in anything that materially reshapes/rewrites text (as that text would then be generative and trigger our zero-tolerance policy). This is an arms race of sorts, and a moving target, and we won’t get every decision right, but we want E-International Relations to be a place for humans who are reading and writing and learning from each other. That’s the bottom line. We understand that some people may feel the position we have arrived at is not a fit for their AI use cases, but they are, of course, welcome to seek publication elsewhere.
With hope, our small part in this big debate will help move the needle in a positive direction. AI is an incredible tool for academics and students. But, like all tools, it must be used wisely.
With all that said, I would like to leave some personal reflections (with my editor/publisher hats off). Detection tools have gotten to the point in 2026 where they can be said to be reliable enough to use. These detection tools will continue to improve, and to closely follow, the further development of the generative AI tools they are chasing. With technical detection and human recognition of AI writing both growing, I think AI writing will become a marker placed on individuals in academia – and that marker will be a bad one. This includes those who think they are using AI just to ‘edit and polish’ their writing, but what they are actually doing is having a large language module materially rewrite their text in new words, often with new sentences and a different paragraph structure. In my opinion, this is not assistive use as if you do that, you are no longer the author.
I think the majority of academic/intellectual society will reach the same fatigue/distaste with generative AI in academic writing as I have expressed here (if they are not getting there already). For that reason, those who have come to rely on AI entirely, or rely on it a bit too much, will find that in time they come to regret it. When that reckoning comes, and it may not be at the same time for every person, every sector (etc.), it may invalidate their work, impact their career progression, or other unforeseen negative things. It may not even be fully their fault, as they have grown up in a world where they never developed strong communication skills in writing, or never developed their own voice, or the academic system failed them. Taking a health analogy, the best way to immunise oneself from this reckoning is to be ‘fully human’ in your writing. At the very least, this is what people ought to aspire to when they are asking (or expecting) others to spend time reading their work and engaging with their ideas. Not all AI writing is slop, but all AI writing is lifeless.
I would like to wrap my thoughts up with another analogy: AI is like smoking cigarettes. Most societies have come to recognise that while a person can smoke, and many do, it is not healthy for the individual or for the society at large to allow smoking to go unregulated, or to be seen in a positive light. Laws are needed to assist with this message (such as age and use restrictions, safety warnings etc.) and norms are also needed within different groups/use cases outside of formal law (such as smokers understanding non-smokers do not want to breathe their second-hand smoke etc.). To bring this to the academic realm, the argument isn’t that AI is equivalent to tobacco in its potential for harm to our health, it is that technologies can become normalised before societies fully understand their consequences – like the widespread use of tobacco in human society before we fully realised its dangers in the 1950s. The danger is not the occasional, conscious, use of AI tools to make certain life tasks easier; it is unreflective dependence disguised as convenience. As AI embeds itself more, my fear is that generations coming through will no longer have the skills or the consciousness to come to such a reflection. Skills will soften and become atrophied. The potential for genuine intellectual discovery will follow, becoming rarer in a world where it is needed more than ever.
The question therefore is whether people can still think, research and write when AI tools are taken away? At present, I think they can, for the most part – but those who act as leaders or gatekeepers across academic and publishing ought to get on the right side of this and take leadership roles in the AI debate – similar to those in science and public health who contributed (and still contribute) to the momentum to get people to reconsider the wisdom of smoking cigarettes.
Dr Stephen McGlinchey is the Editor-in-Chief and Publisher of E-International Relations, and Senior Lecturer of International Relations at UWE Bristol. His publications include Foundations of International Relations (Bloomsbury 2022, 2nd ed forthcoming 2028), Global Politics in a Post Truth Age (2022), International Relations (2017), International Relations Theory (2017) and US Arms Policies Towards the Shah’s Iran (Routledge 2014).
Postscript: To underline the point, I have asked a leading AI model to tidy up and polish the 100% human written article above which I have pasted unedited below. Using the standard prompt logic someone would use when relying on AI tools. See for yourself how it looks and which you prefer to read. I’m not the best writer in the world, but it sure beats the lifeless version below. And, see if you can spot all the tells (these generated from an AI detector which picked it up when I asked it to compare the two) such as:
Formulaic signposting: “This article does not attempt to provide… Rather, it reflects on…”
Predictable contrasts: “On the one hand… On the other hand…” and “The concern is not simply X. It is Y.”
Excessive framing: repeatedly telling the reader why an issue matters before making the substantive point.
Three-part lists: “consistency, judgement, and a willingness to adapt”; “learning, independence, and intellectual growth.”
Generic transitions: “This experience reinforced an important lesson”, “Taken together, these factors…” and “Ultimately, the central question…”
Over-explained arguments: taking a relatively direct observation and expanding it into several layers of qualification and interpretation.
A conventional academic conclusion: restating the central argument, broadening its implications, and ending with a general statement about the future.
A flattened authorial voice: replacing some of the original’s sharper, more personal observations with balanced, impersonal and institutionally acceptable prose.
Introduction: A New Era of Academic Writing
In recent years, generative artificial intelligence has emerged as one of the most significant technological developments of our time. Since the widespread introduction of ChatGPT in 2022, AI has transformed the ways in which people communicate, access information, learn, and produce written work. What began as a text-based tool has rapidly expanded into a much broader ecosystem encompassing speech, images, video, and other forms of content generation.
For academic publishers, this transformation has created both opportunities and challenges. On the one hand, AI offers the potential to improve efficiency, support communication, and make complex ideas more accessible. On the other hand, it raises fundamental questions about authorship, originality, academic integrity, and the future of scholarly communication. How should publishers respond to the growing number of submissions? How can they distinguish legitimate assistance from the wholesale generation of academic work? And what responsibilities do editors have in an environment where the boundaries between human and machine-generated writing are increasingly difficult to define?
This article does not attempt to provide a comprehensive analysis of generative AI across academia. Rather, it reflects on how E-International Relations has approached these challenges, the lessons learned along the way, and the broader implications for academic writing and publishing.
From Uncertainty to Editorial Practice
Like many publishers, E-International Relations initially had to navigate an unfamiliar and rapidly changing landscape. Our first response was to make our expectations explicit: authors’ thoughts and words must be their own. We retained our established editorial process, which involved two stages of human review, while becoming increasingly attentive to the linguistic patterns and stylistic signals commonly associated with AI-generated text.
In the early stages, some submissions could be rejected relatively quickly because their writing displayed recognisable AI-like characteristics. Over time, however, the situation became more complicated. The tools themselves improved, and authors began using them in different ways. Some relied on AI to generate substantial portions of their work; others used it to reshape, polish, or reorganise writing that had begun as a human draft.
The result was a grey area in which definitive judgements became more difficult. Some submissions appeared to contain AI-generated material while also offering insights that might otherwise have justified publication. In such cases, editors had to weigh the apparent use of AI against the intellectual value of the work, while recognising that it was not always possible to establish whether a submission had exceeded the boundaries of acceptable assistance.
This experience reinforced an important lesson: there is no simple, universal answer to the question of how academic publishers should respond to generative AI. Effective policy requires judgement, consistency, and a willingness to adapt as the technology evolves.
The Human Cost of AI-Generated Writing
Over time, however, a more fundamental concern came to the fore: editorial fatigue. After years of reading AI-shaped prose, I and other editors increasingly found ourselves confronted by writing that was technically fluent but intellectually lifeless. Whether a text had been generated almost entirely by AI or substantially rewritten by it, the outcome often felt much the same. The prose was formulaic, predictable, and lacking in distinctive voice.
This matters because editing has traditionally been more than the correction of grammar or the improvement of sentence structure. One of the most rewarding aspects of academic publishing is helping authors communicate ideas that are genuinely interesting, even when those ideas are initially expressed imperfectly. An author may be inexperienced, may struggle with style, or may be writing in a second language. Through careful editorial work, an argument can become clearer without losing the individuality that made it worth publishing in the first place.
In this sense, editing is a collaborative intellectual process. Editors learn from the work they read, refine their own judgement, and contribute to the development of an author’s voice. When AI substantially rewrites the text, some of that process is displaced. The immediate result may look more polished, but the deeper educational value can be diminished.
The implications extend beyond publishers. Writing is not simply a way of presenting ideas; it is also a way of developing them. By drafting, revising, struggling with structure, and finding the right words, writers build intellectual habits and communication skills. If these activities are increasingly outsourced to AI, there is a risk that people will lose opportunities to develop the very capacities that academic education is supposed to cultivate. This is not merely a question of style. It is a question of learning, independence, and intellectual growth.
A Turning Point: Substack and Transparency
A second turning point came with E-International Relations’ move to Substack in late July 2026. The move changed more than the technical location of the publication. It also prompted a reconsideration of the relationship between publisher and reader. Publishing was no longer simply a matter of placing an article on a website; it increasingly involved delivering that article directly to readers’ inboxes and cultivating a more immediate relationship with an audience.
At approximately the same time, Substack incorporated Pangram, an AI-detection tool, into its publishing system. The platform also made it possible for readers to generate summary-style AI scans of published work. E-International Relations chose to enable this feature as part of a commitment to transparency.
These developments created an opportunity to reconsider our approach. Editorial fatigue had already made the limitations of AI-shaped writing increasingly apparent, while the platform’s new features raised the prospect of greater visibility around the question of AI use. Taken together, these factors encouraged us to revise our submissions guidance and publishing agreement.
The result was a clearer and more demanding policy. The central principle is straightforward: work that appears to be AI-generated, either wholly or in part, will be rejected. Authors must provide a declaration confirming that their thoughts and words are their own and that AI has not been used to generate text. Limited assistive uses, such as light editing, remain permissible, provided they are declared and do not materially reshape or rewrite the work.
No detection system is infallible, and the technology will continue to evolve. Nevertheless, the policy establishes a clear standard for what E-International Relations wants to be: a place where people write, read, think, and learn from one another.
Beyond Policy: A Personal Reflection
Stepping away from my role as editor and publisher, I want to consider the issue from a more personal perspective. My concern is that AI-generated writing may increasingly become a negative marker within academic life. As detection tools improve and readers become more familiar with the characteristics of AI-shaped prose, writing that has been materially generated or rewritten by AI may come to be regarded as evidence of diminished intellectual independence.
This concern applies not only to those who ask AI to produce an entire article, but also to those who believe they are merely using it to edit or polish their work. There is an important distinction between correcting a grammatical error and asking a language model to replace sentences, restructure paragraphs, and recast an argument in different words. In the latter case, the technology is no longer simply assisting the author. It is participating materially in the act of authorship.
It is possible that many people who currently depend on these tools will come to regret that dependence. Their work may be viewed less favourably, their professional development may be affected, or they may discover that they have not developed the writing skills they need when AI is unavailable. These outcomes will not necessarily be the product of individual irresponsibility alone. Educational systems, institutional expectations, and the normalisation of automated assistance may also play a role.
The challenge, therefore, is not simply to tell individuals to use AI responsibly. It is to create environments in which human judgement, intellectual effort, and the development of a distinctive voice remain valued. Not every piece of AI-generated writing is poor. Yet even fluent prose can lack the qualities that make writing intellectually engaging: personality, curiosity, uncertainty, and the unmistakable presence of a mind working through a problem.
AI and Smoking: An Analogy for the Risks of Normalisation
A useful way to think about these concerns is through an analogy with smoking. The comparison is not that AI use presents the same direct health risks as tobacco. Rather, it highlights how a practice can become normalised before society fully understands its longer-term consequences.
For many years, smoking was widely accepted and often presented positively. Over time, scientific evidence, public-health campaigns, regulation, and changing social norms reshaped how it was understood. The lesson is not that every new technology should be prohibited. It is that convenience and popularity are not, by themselves, reliable indicators of long-term benefit.
The same principle may apply to generative AI in academic writing. Used selectively and consciously, AI can make certain tasks easier. The greater danger lies in unreflective dependence: a gradual shift from using a tool to support one’s work towards relying on it to perform the intellectual work itself. As this dependence becomes normal, the skills that make independent writing possible may weaken through disuse.
There are consequences for individuals, but also for academic culture. If students and scholars increasingly outsource the work of developing arguments and expressing ideas, opportunities for genuine discovery may become less frequent. The concern is not simply that academic prose will become more uniform. It is that the process through which people learn to think, research, and communicate may be altered in ways that are difficult to reverse.
The analogy also points to the importance of collective responsibility. Formal rules matter, but so do informal expectations. Just as societies developed norms around smoking in shared spaces, academic communities can develop clearer norms around AI use, authorship, and transparency. Publishers, universities, educators, and researchers all have a role to play in shaping those expectations.
Conclusion: Keeping Human Thought at the Centre
Ultimately, the central question is a simple one: can people still think, research, and write when AI tools are taken away? For the most part, I believe they can. But this capacity should not be taken for granted. Skills require practice, and intellectual independence requires opportunities to exercise it.
Academic publishers and institutional leaders therefore have a responsibility to engage seriously with the implications of generative AI. That means moving beyond the assumption that technological adoption is inevitable or inherently beneficial. It means developing clear policies, supporting transparent practices, and protecting the conditions in which people can develop their own ideas and voices.
Generative AI is a powerful tool, and it would be mistaken to deny its potential benefits. But the existence of a tool does not determine how it should be used. The task is to ensure that convenience does not come at the expense of learning, that fluency is not confused with insight, and that technological efficiency does not displace human intellectual development.
The future of academic writing will depend not only on what AI systems become capable of doing, but also on what scholars, publishers, educators, and readers decide is worth preserving. If we want academic work to remain a genuinely human exchange of ideas, we must be prepared to defend the value of thinking and writing for ourselves.