Teaching Scientific Writing as Process While the Process Is Changing

Ioana Cretu, Crețu R. Ioana PFA, Iași, Romania

In over two decades of teaching academic and medical English across different professional and cultural contexts, I witnessed several shifts in how research is conducted, written, and shared, but none as disruptive as the current one. Artificial Intelligence, increasingly ubiquitous, multimodal, agentic, is not simply adding to the toolkits of tech-savvy writers; it is upending what it means to think and communicate generally, and to draft, revise, and even decide what counts as evidence in scientific writing. I catch myself sliding back and forth between accepting and resisting change, in search of the ever-elusive sweet spot of tech-enabled productivity without surrender of identity and thought.

More than ever, the focus of my ESP work is now on process, hence the title. I first articulated my approach in a talk at the IATEFL 2025 Conference in Edinburgh (Crețu, 2025), then in a scholarly article accepted for publication in the journal Professional and Academic English, and most recently in a TESOL webinar on using AI’s shortcomings as learning opportunities (Crețu, 2026). These three moments, across different genres, have allowed me to promote process-based, non-linear, adaptive, multimodal, and tool-rich approaches precisely to keep science human-centered and critically grounded.

My own hybrid writing workflows reflect this philosophy and I mention this not out of self-indulgence, but because the way I write is inseparable from how I am able to help others with their writing. Mine involve constant movements between stages, formats, and methods to stimulate ideas and connections that a linear model would stifle in me. Ideas arrive while assessing a manuscript, planning a lesson, walking, gardening, cooking (with boundaries). I use the latest audio note-taker and transcription apps to ramble, transcribe, summarize, and edit thoughts and conversations. These imperfect tools act as external working memory as I offload ideas when alone and stay present when together. I deliberately counterbalance this with handwriting, using a Parker 51 fountain pen that is older than I am and moves better than I do. It wisely slows me down and restores a tactile relationship with ideas, which I need and is needed.

Once a critical mass of notes exists, I use two or three LLMs as mirrors, deciding what to keep, redo, add, how to organize, to stay concise, all within inevitable constraints of time and format. Importantly, I am under no illusion that the tech will preserve my thinking, and training it to imitate me is too cringe for my taste. No, the guardian of my thinking and my voice is not the tech; it’s me. Against this backdrop, the struggles of the PhD candidate whose manuscript I discuss next, and the contrasting perspectives of two other researchers experimenting with AI-mediated workflows, all non-native English speakers, are individual stories that illustrate the spectrum of learning needs and renegotiations.

Just before Edinburgh, I was contacted by a PhD-level clinician-researcher who had received two rapid rejections from international journals. ‘I’m at a standstill,’ she said. As I reviewed her manuscripts, I was struck not by sloppiness, but by the opposite: the English was clean, (too) sophisticated. The volume of work, impressive. The emotional toll was palpable in my client’s self-doubts and exhaustion (working full-time, raising a child, preparing for further professional exams). Non-compliance with the journals’ requirements justified the rejections, but there was more, deeper misalignment between aims, methodology, data, language, and between intention, language, effect on reader.

My initial focus was to help her see these layers of disconnect for herself. In the latest, much revised version, many issues are resolved, yet the revision still allows claims to outrun the data. Interview responses about beliefs and attitudes pass as ‘behavioral’ evidence. Terms such as ‘predictor’ are used without the statistical modelling that would justify them. Cautious descriptive findings are escalated to unsubstantiated assertions about ‘urgent’ policy needs. And the AI tool my client used to check accuracy, eliminate redundancies, and ensure consistency did half the job, keeping both American and British spelling, misplacing commas, repeating words, and sentences even in conclusions. As Hyland (2003) has long argued, problems in L2 academic writing are often not grammatical but rhetorical and epistemic: writers have difficulty negotiating disciplinary expectations, audience, and what the evidence entitles them to say.

My client’s struggles are informative for me as an ESP practitioner, and by no means singular. In time, I have developed as routine practice to collaboratively, non-judgmentally check alignment as illustrated here; sometimes, unexpectedly, valid further research ideas emerge from calling out ill-fitting methodological or linguistic choices. This invites the teaching and modelling of key distinctions, as needed: data type versus rhetoric, correlation versus causation, logical fallacies and biases that creep into interpretation. It also helps to read backwards as devil’s advocates, tracing every statement to its data and/or references, disentangling facts from inferences and hypotheticals. And now, I (re)introduce AI into the process as tool, not ghostwriter.

This resonates strongly with process-oriented traditions in TESOL that Raimes (1991) described over three decades ago, when writing was reframed as thinking, drafting, revising, and negotiating meaning rather than merely producing error-free text. Today, AI systems can make inconsistencies visible, but they can also amplify them if unchecked. ‘Writing is thinking’ (Nature, 2025) and no amount of plausible text generation can substitute human judgement about what is relevant, defensible, and true.

The perspectives of two other researchers I collaborate with further illustrate why workflow design matters just as much as, if not more than, language proficiency. A young clinician with C1 English, new to research, reacted to new AI tools first with awe (‘wow,’ ‘mind-blowing,’ ‘very impressive’); then, as she began to test them, first together, then independently, she regrouped: ‘Artificial Intelligence can be a very important tool for everything that means selecting information, interpreting it, facilitating, saving time, summarizing information, having easier access, such as all the apps through which we can transform information in(to) an audio file, for example, so these could be, and for me they are, very, very important tools. I acknowledge that my experience regarding the reliability of the information is limited, with all the tools that exist at the moment.’ She is open, curious, excited, and we make sure she is also critically aware of AI reliability and other issues.

By contrast, a seasoned researcher and university lecturer with B1–B2 English and extensive disciplinary expertise has been approaching AI with great caution: ‘As far as I am concerned, it is very tempting to use AI for writing emails, for much faster searching of bibliographic references, and for extracting the main ideas for courses. However, because I am extremely critical by nature, I am afraid of making a blunder by choosing to rely only on that, out of fear that I might miss the correct meaning of a word or concept (which happened to me when I tried to obtain a license to use a medical questionnaire). I am slow to make these tools my own. I do acknowledge that I try to test the ones that are recommended by peers who publish articles using those platforms. What helps me enormously, however, are the scenarios we build together, imperfect while we work on them, but perfect for my discussions with students, for courses, exams, supervising. When I am under serious time pressure, I admit that they help a great deal, but I would not go onto any platforms without your recommendation first.’

All three researchers I mentioned are capable and committed; each need workflows that match their linguistic abilities, cognitive styles, risk tolerance, and professional responsibilities. For ESP practitioners, their stories carry an important lesson. Teaching stable genres with relatively predictable pathways to publication no longer applies; our role is to guide learners through uncertainty, experimentation, and recalibration. For that, we should design challenging yet safe thinking environments for learners to exercise their control of data, interpretation, language, and technology. This is more demanding work, yes, but necessary if we want researchers to be responsible AI-assisted authors of the science in sentence, for the benefit of us all.

References

Cretu, I. (2025). Teaching Scientific Writing as Process While the Process Is Changing. IATEFL International Conference, Edinburgh, 8-11 2025. Available online at https://www.youtube.com/watch?v=6cUnbBB-gVw

Cretu, I. (2026). From Frustration to Facilitation: Turning AI’s Shortcomings into Learning. TESOL ESP IS webinar, 29 January 2026. Available online at https://youtu.be/YlV9vGYdEpQ?si=Jv7aGK1GxZygNgCp

Hyland, K. (2003). Second Language Writing. Cambridge University Press.

Nature. (2025). Writing Is Thinking. Nature Reviews Bioengineering, 3, 431. https://doi.org/10.1038/s44222-025-00323-4

Raimes, A. (1991). Out of the woods: Emerging traditions in the teaching of writing. TESOL Quarterly, 25(3), 407–430. https://doi.org/10.2307/3586978

Acknowledgment

I confirm that the excerpts of client-authored texts and reflections included in this paper were used with consent granted contractually at the outset of each professional collaboration. Signed contracts explicitly permit my right to use such samples for educational and demonstrative purposes, provided they are anonymized and embedded within my original analytical commentary. No personally identifiable information is disclosed.

Ioana Crețu, PhD, is an independent ESP and EMP specialist, scientific writing consultant, OET Preparation Partner, and certified medical translator. She partners with Romanian and international healthcare professionals to advance effective clinical and scientific communication (https://ioanacretu.ro). Her background includes continuous teaching of medical and academic English since 2003, including at university, and a BA, MA, PhD, and TESOL certificate on relevant topics.