This study contributes to the emerging research on generative AI and writing pedagogy by exploring how college writing students make use of GAI when offered instruction in a range of responsible uses and latitude to integrate it into their writing process as they see fit. We analyzed chat log data and papers from participants recruited from six sections in which students were guided in experimenting with ChatGPT Plus and permitted to use it to produce up to 50% of submitted work. Through a combination of AI and human thematic content analysis of student chat logs, we found that in 18.6% of prompts, students asked ChatGPT to write for them. The rest of the prompts involved work leading up to or in support of the writing process. Human thematic content analysis of papers showed that students used ChatGPT to generate 8.2% of the writing they submitted. The most common rhetorical purpose of the AI-generated text they included was discussion/analysis/synthesis. English as a foreign language students (EFLs) in the sample prompted ChatGPT to clarify understanding less often than non-EFLs and integrated less AI-generated text into their papers, with a particularly notable difference in their use of AI-generated summaries. This unexpected finding merits further research, but it suggests that EFLs may use GAI for somewhat different purposes than non-EFL peers.

This special issue of the Journal of Writing Research brings together seven empirical studies of the relationship between writing and generative AI, examining what can be systematically observed and measured about the functioning of generative AI in educational and professional writing contexts. Collectively, the studies demonstrate the necessity and value of methodological pluralism for investigating a complex, rapidly evolving phenomenon. In their contributions, the researchers use experimental comparisons, mixed-methods intervention designs, corpus-based analyses, computational linguistic techniques, and qualitative interpretive approaches. Taken together, these methods enable lines of inquiry that no single approach could sustain: comparisons of AI and human performance in professional writing tasks; analyses of how writers at different ages and levels of expertise engage AI tools; examinations of how assessment systems register and respond to AI-generated prose; and investigations of how human readers interpret texts with ambiguous authorship. By foregrounding both the affordances and limitations of different methodological traditions, the articles present a multifaceted approach to the study of writing and generative AI.

Since the launch of ChatGPT, the use of and debate around generative AI has grown rapidly. Professionals whose work depends on writing have expressed concern about the potential impact of such tools on their roles. But are these concerns justified? Can ChatGPT truly take on the responsibilities of a professional writer? This study investigates that question by comparing the performance of ChatGPT with that of professional editors tasked with optimizing business communication. We conducted two studies, using both qualitative and quantitative methods. In the first, three experienced editors were asked to rewrite four business letters. Their editing processes were recorded using the Microsoft Snipping Tool, and immediately afterward, we conducted retrospective interviews using stimulated recall. These interviews were transcribed and analyzed. Insights from the observations and interviews informed the design of the prompt instructions used in the second study. In the second study, we asked ChatGPT to revise the same four letters using three different prompt types. The Simple prompt instructed the model to “make this text reader-focused.” The B1 prompt referred explicitly to the CEFR B1 language level, requiring ChatGPT to tailor the text for intermediate readers. Finally, the Process prompt simulated the editing steps observed in the professional editors’ workflows. To evaluate outcomes, we conducted both a qualitative comparison of the revised texts and a quantitative readability analysis using LiNT, a validated tool developed for Dutch texts. Our results show that the human editors substantially improved the readability of the original letters, reducing the use of unfamiliar words, shortening complex sentences, and increasing personal engagement through pronoun use. Among the AI outputs, ChatGPT B1 achieved results most comparable to the editors, both in readability and accuracy. In contrast, ChatGPT Simple fell short in terms of clarity and introduced errors through faulty inferences. Surprisingly, ChatGPT Process also underperformed compared to ChatGPT B1 and the human editors. Only the editors' and ChatGPT B1versions were free from errors. In the discussion, we reflect on how generative AI is reshaping the concept of writing within organizations, the skills required to produce effective written communication and the impact on writing pedagogy. Rather than replacing human editors, we argue that generative AI can play a valuable role as a collaborative tool in the organizational writing process.

This study focuses on a generative AI approach to facilitate qualitative analysis in Writing Studies research. We gathered 13,336 one-sentence to one-paragraph responses written by 3,334 incoming students in a directed self-placement program administered at a large R1 U.S. university. In these responses, students describe their high school writing experience and college writing expectations. In stage one of the project, we pilot the use of Retrieval-Augmented Generation to expedite the selection of relevant responses for a topic—in this case, students’ positive self-assessments as writers. The selected responses were then compared to a random sample and rated by three faculty with writing expertise. In stage two, these faculty generated codes and themes from a subset of the responses, incorporating ChatGPT-4 through the stages of thematic analysis. Results show that the use of AI expedites and enhances qualitative analysis, but human participation in the process is still essential. We suggest a machine-in-the-loop framework with which Writing Studies researchers can more readily integrate generative AI to study large corpora of student writing.

With the emergence of generative artificial intelligence, dialogue systems like chatbots are redefining traditional concepts of authorship and impacting critical aspects of writing. In educational contexts, previous research has pointed out new opportunities associated with using chatbots for writing instruction and support. This study involved 108 students across 10 classes in Norwegian K-12 education, examining how they employed educational chatbots as a support tool in L1 writing assignments. Through an inductive, data-driven thematic analysis of 895 student prompts, five recurring patterns emerged: information requests, structural guidance, example requests, content creation, feedback on text, and follow-up clarification. Aggregated results show that information requests were the most common pattern, particularly among younger students, whereas content creation and feedback on text were more prevalent among secondary and upper secondary students. Illustrative examples from the conversations revealed that generative AI extensively produced content on student’s behalf, even when students primarily sought scaffolding. The study proposes that effective scaffolding of writing through educational chatbots requires not only refining students' prompting strategies but also enhancing system designs that better support pedagogical use of generative AI.

In this study, we examine the feasibility of augmenting student-written essays with those generated by large language models (LLMs) for scoring essays. We found that with correct instructions, generative AI systems such as GPT-4 and GPT-4o can generate essays similar to those written by students in terms of surface-level linguistic features, although material differences may still exist. Systematic analyses revealed that scoring models trained with synthetic data perform comparably to models trained using student essays, but the performance varies across prompts and the sizes of the model training sample. The augmented models could alleviate large discrepancies between human and AI scores on the subgroup level that may be introduced by a lack of training samples for a particular subgroup or due to inherent biases in LLMs. We also explored an established method – DecompX – on token importance to identify and explain AI predictions. Future research directions and limitations of this study are also discussed.

This empirical exploration investigates how integrating a handwritten diary with a generative AI writing companion can strengthen elementary school students' writing habits and interests in a naturalistic classroom setting. The AI companion serves as a personalized assistant, offering real-time ideas, suggestions, and feedback. By encouraging students to handwrite daily experiences and emotions, then digitize their entries, the approach fosters both reflection and skill development. Over 18 weeks, 32 students from grades three to five (average age 10.5 years old) recorded their diary in Chinese and interacted with the AI companion. This exploratory study employed a pre-post, single-group design, analyzing diary entries, interaction logs, and questionnaire data to assess changes in writing participation and interest. The findings indicate three major outcomes: a notable increase in writing participation, reflected by a rise in the number of ideas and entry length; an enhanced level of writing interest, demonstrating the effectiveness of merging traditional handwriting with AI tools; and improved writing behavior through more frequent and diverse writing activities. When students encountered challenges—such as topic selection or content organization—the AI companion supplied up to three suggestions, preventing information overload and preserving independent thinking. Overall, this interactive, AI-supported environment transformed writing from a solitary task into a dynamic, collaborative process, boosting motivation and quality. The study thus illustrates how strategically blending handwritten diary with innovative AI systems can enrich writing education and sustain students' long-term engagement, while acknowledging its exploratory nature and the need for further research to establish causal links.

This study investigated the impact of using ChatGPT 3.5 as a prewriting brainstorming tool on the overall quality of persuasive writing among five gifted seniors majoring in Arabic at the College of Education, Kuwait University. Giftedness, in this study, was not defined by innate advantages such as intelligence quotient (IQ) but was instead viewed from a multidimensional perspective, focusing on academic performance, writing skills, and personal traits that reflect intellectual engagement. Four participants were typically developing gifted students, while one participant was twice exceptional, both gifted and autistic. An integrated single-subject design with multiple probes across multiple baselines was used, with each participant serving as their own control. Repeated measures were used throughout the baseline, intervention, and maintenance phases to monitor intraindividual variability and examine the effectiveness of the intervention. The results indicated a significant increase in mean scores for persuasive essays from baseline to intervention for all participants, with continued improvement during maintenance for all but the twice-exceptional student, whose mean maintenance score remained unchanged from the intervention. While promoting ChatGPT 3.5 as a valuable brainstorming tool for persuasive writing, this study emphasizes its complementary role and recommends that writers engage in brainstorming using multiple resources before writing.

Generative AI has the potential to transform writing in schools and universities. This makes it necessary to develop training programmes for writing with AI, especially for students in teacher training. So far, however, little is known about the students' initial preconditions on which the trainings can be based upon. Evidence so far has come mainly from observational studies and questionnaire studies examining the frequency and type of AI use. However, the students themselves were not considered, nor the extent to which they can be categorised into groups. In other words, the focus has been on the writing rather than on the writers. To address this gap, the present article analyses data from a survey of N=505 students. To identify writer profiles, i.e. groups of students with comparable characteristics, we apply two-step cluster analysis. The students are clustered based on their use of AI for writing, as well as their level of awareness of AI applications, AI literacy, digital media literacy and writing-related self-concept. The results reveal four clusters, the two largest of which are characterised by the fact that students tend not to use AI, sometimes because they apparently have no awareness of AI, sometimes despite having such awareness. Merely one cluster, which describes 20% of the students, is characterised by regular use of AI for writing. The results therefore provide a useful insight for planning training in the context of university teaching.

Artificial intelligence-based Language Tools (AILTs) are being increasingly used in essay writing in higher education. Its application promotes global and multicultural perspectives in education and plays a critical role in advancing scholarly communication and research dissemination. However, these benefits cannot be measured without also considering student perspectives. This study analyzes the positive and negative aspects identified by students regarding the use of AILTs in their written texts at university. A total of 314 undergraduate and graduate education students were surveyed, and results were analyzed using the Reinert method. The results show that positive aspects are linked to the three pillars of text construction (planning, textualization, and revision). The negative aspects highlight concerns about academic integrity and student competencies. These findings can help guide teachers on how they can promote the responsible and beneficial use of AILTs.

This study examines links between human ratings of writing quality and the incidence of argumentative features (e.g., claims, data) in persuasive essays along with relationships among these features and their distance from one another within an essay. The goal is to better understand how argumentation elements in persuasive essays combine to model human ratings of essay quality. The study finds that, in most cases, it is not the presence of argumentation features that is predictive of writing quality but rather the relationships between superordinate and subordinate features, parallel features, and the distances between features. This finding has not only theoretical value but also practical value in terms of pedagogical approaches and automated writing feedback.

This article introduces a Special Issue that gathers a collection of effective tools to promote the teaching and learning of writing in school-aged and university students, across varied contexts. The authors present the theoretical rationale and technical specificities of writing tools aimed at enhancing writing processes (e.g., spelling, revising) and/or at providing writers with automated feedback to improve the implementation of those processes. The tools are described in detail, along with empirical data on their effectiveness in improving one or more aspects of writing. All articles conclude by indicating future directions for further developing and evaluating the tools. This Special Issue represents an important contribution to the field of technology-based writing instruction, in a moment in which online teaching and learning tools have shifted from being an instructional asset to a necessity. We hope that in the future the validation of each tool can be expanded by reaching out to different populations and cultural contexts.

Teaching the craft of written science communication is an arduous task that requires familiarity with disciplinary writing conventions. With the burgeoning of technological advancements, practitioners preparing novice research writers can begin to augment teaching and learning with activities in digital writing environments attuned to the conventions of scientific writing in the disciplines. The Research Writing Tutor (RWT) is one such technology. Grounded in an integrative theoretical framework, it was designed to help students acquire knowledge about the research article genre and develop research writing competence. One of its modules was designed to facilitate revision by providing different forms of automated feedback and scaffolding that are genre-based and discipline-specific. This study explores whether and how the features of the RWT may impact revision while using this module of the tool. Drawing from cognitive writing modeling, this study investigates the behaviors of a multidisciplinary group of 11 graduate-student writers by exploring how they interacted with the RWT's features and how this interaction may create conditions for enhanced revision processes and text modifications. Findings demonstrate promising potential for the use of this automated feedback tool in fostering writers' metacognitive processing during revision. This research adds to theory on cognitive writing models by acknowledging the evolving role of digital environments in writing practices and offering insights into future development of automated tools for genre-based writing instruction.

There is increasing evidence that automated writing evaluation (AWE) systems support the teaching and learning of writing in meaningful ways. However, a dearth of research has explored ways that AWE may be integrated within different instructional contexts and examined the associated effects on students’ writing performance. This paper describes the AWE system MI Write and presents results of a mixed-methods study that investigated the integration and implementation of AWE with writing instruction at the middle-school level, examining AWE integration within both a traditional process approach to writing instruction and with strategy instruction based on the Self-Regulated Strategy Development model. Both instructional contexts were evaluated with respect to fostering growth in students’ first-draft writing quality across successive essays as well as students’ and teachers’ experiences and perceptions of teaching and learning with AWE. Multilevel model analyses indicated that during an eight-week intervention students in both instructional contexts exhibited growth in first-draft writing performance and at comparable rates. Qualitative analyses of interview data revealed that AWE’s influence on instruction was similar across contexts; specifically, the introduction of AWE resulted in both instructional contexts taking on characteristics consistent with a framework for deliberate practice.

The inclusion of peer feedback activities into the academic writing process has become common practice in higher education. However, while research has shown that students perceive many features of peer feedback to be useful, the actual effectiveness of these features in terms of measurable learning outcomes remains unclear. The aim of this study was to investigate the linguistic and review features of peer feedback and how these might influence peers to accept or reject revision advice offered in the context of academic writing among L2 learners. A corpus-based machine learning approach was employed to test three different algorithms (logistic regression, decision tree, and random forests) on three feature models (linguistic, review, and all features) to determine which algorithm offered the best predictive results and to determine which feature model most accurately predicts implementation. The results indicated that random forests is the most effective way of modeling the different features. In addition, the feature model containing all features most accurately predicted implementation. The findings further suggest that directive comments and multiple peer comments on the same topic included in the feedback process seem to influence implementation.

This study examines the potential for computational tools and human raters to classify paragraphs based on positioning. In this study, a corpus of 182 paragraphs was collected from student, argumentative essays. The paragraphs selected were initial, middle, and final paragraphs and their positioning related to introductory, body, and concluding paragraphs. The paragraphs were analyzed by the computational tool Coh-Metrix on a variety of linguistic features with correlates to textual cohesion and lexical sophistication and then modeled using statistical techniques. The paragraphs were also classified by human raters based on paragraph positioning. The performance of the reported model was well above chance and reported an accuracy of classification that was similar to human judgments of paragraph type (66% accuracy for human versus 65% accuracy for our model). The model’s accuracy increased when longer paragraphs that provided more linguistic coverage and paragraphs judged by human raters to be of higher quality were examined. The findings support the notions that paragraph types contain specific linguistic features that allow them to be distinguished from one another. The finding reported in this study should prove beneficial in classroom writing instruction and in automated writing assessment.