AI-Integrated Scaffolding to Enhance Agency and Creativity in K-12 English Language Learners: A Systematic Review

Abstract

This systematic literature review examines how artificial intelligence (AI)-integrated tools are being used to scaffold writing instruction for K-12 English as a Foreign Language (EFL) students, with a particular focus on preserving student agency and creativity. Drawing from sociocultural and cognitive theories of writing, this study synthesizes findings from 14 peer-reviewed empirical studies published between 2018 and 2024. Studies were analyzed using a thematic synthesis approach to identify how AI tools function as cognitive supports, creative collaborators, and language enhancement systems. The results reveal that AI tools, when implemented thoughtfully, can reduce cognitive load, foster ideation, and support self-regulated learning without undermining student autonomy. Three major categories of AI scaffolding emerged, i.e., cognitive support tools (e.g., grammar correction, idea organization), creative support systems (e.g., story generation, narrative feedback), and language enhancement tools (e.g., vocabulary expansion, stylistic refinement). Key themes include the importance of balancing AI support with student decision making, aligning tool use with theoretical models, and ensuring equitable access. Limitations in the current literature include short-term study durations, a lack of standardized metrics for creativity and agency, and the predominance of research in EFL contexts. Future research should expand on ESL settings, adopt rigorous methodologies, and explore long-term impacts on writing development.

1. Introduction

In today’s globalized education landscape, English writing functions as more than a language skill—it is a gatekeeper to academic achievement and access to future opportunities. For K-12 students in an English as a Foreign Language (EFL) context, proficiency in written English often determines participation in advanced schooling, access to higher education, and long-term success in the global workforce [1]. Yet, for these learners, writing in a second language is not only cognitively demanding, but also it requires the development of creative expression in a language that is not their own. Language barriers can lead to academic isolation and restrict students’ capacity to explore and articulate original ideas, diminishing their creative potential in the classroom [2].

Amid these challenges, artificial intelligence (AI) has gained momentum as a potentially transformative tool in writing education. From grammar correction and idea generation to content organization and automated feedback, AI-integrated tools promise to streamline various stages of the writing process [3]. For EFL students facing a heavier linguistic and cognitive load than their L1 peers, these tools may offer scalable, individualized scaffolding that helps them engage more fully in writing tasks and focus on higher-order thinking. Early research suggests that AI can support students in overcoming challenges with transcription, organization, and fluency, which may, in turn, foster greater creative engagement in writing [4,5,6].

However, the potential benefits of AI are accompanied by concerns about how its integration in writing instruction may affect student agency. As AI tools become more sophisticated, there is a growing risk that students may rely on them for fundamental aspects of composition—such as generating ideas, developing content, and shaping expression. This raises critical questions about whether AI serves as a scaffold that enhances creativity or a substitute that undermines it [7]. The balance between support and over-reliance is especially delicate for EFL learners, whose writing development depends on building confidence in their own creative capacities. If not carefully implemented, AI may inadvertently suppress the very cognitive engagement and creativity that writing instruction seeks to cultivate.

To address this complex intersection, researchers have begun to frame AI as a “machine-in-the-loop” collaborator—an intelligent co-participant in writing processes rather than a replacement for the human writer [8,9]. This collaborative perspective calls for pedagogical strategies that preserve student creativity while leveraging the strengths of AI to scaffold growth [10]. Such strategies are particularly important in EFL contexts, where learning to write involves both mastering a language and discovering a scholarly identity.

Recent literature reviews underscore the transformative role of AI in education, particularly in higher education contexts. Castillo-Martínez et al. [11] conducted a systematic review of 85 studies, identifying AI’s benefits in enhancing research efficiency, supporting scientific writing, and improving academic productivity, while also highlighting challenges related to ethics and the necessity for human oversight. Bond et al. [12] synthesized 66 systematic reviews, revealing a predominance of AI applications in adaptive learning systems and predictive analytics, and emphasized the need for greater ethical, methodological, and contextual considerations in future research. Chu et al. [13] analyzed 138 articles, noting a significant increase in AI-related publications and identifying trends such as the shift in research focus from the US to China, and the emergence of new applications like intelligent tutoring systems and AI assistants. These studies collectively demonstrate AI’s potential to revolutionize higher education by enhancing learning experiences, personalizing instruction, and streamlining administrative tasks.

However, the existing body of research predominantly focuses on adult learners in higher education, leaving a significant gap in understanding AI’s impact on younger learners in K-12 settings, particularly in the English as a Foreign Language (EFL) context. In the context of K-12 education, Woo et al. [6] explore how secondary students interact with AI-generated text during composition writing, revealing complex cognitive processes beyond mere text insertion. Additionally, the CGScholar AI Helper Project investigates the effects of AI-driven feedback on 11th-grade students’ writing development, suggesting that customized AI feedback can support writing skills in diverse, low-income school settings [14]. lTwijri and Alghizzi [15] provide a systematic review examining how AI integration in EFL classrooms influences learners’ affective factors, such as motivation, engagement, anxiety, and self-confidence. Drawing from 22 empirical studies, they find consistent evidence that AI applications—especially chatbots and intelligent tutors—can enhance emotional engagement and reduce anxiety when effectively implemented. While these findings underscore AI’s value in shaping learners’ emotional responses, the review focuses primarily on general EFL instruction and affective outcomes, with minimal attention to the cognitive and compositional dimensions of writing. In contrast, the present review centers specifically on K-12 EFL writing instruction, emphasizing how AI-integrated tools support student agency and creativity in the writing process. Rather than addressing affective factors alone, this study explores how AI tools scaffold idea generation, content development, and metacognitive regulation in writing tasks—the critical aspects of cognitive engagement and identity formation in language learning. This focus responds to a notable gap in the literature, offering a more granular understanding of how AI mediates students’ intellectual participation in EFL writing.

Thus, in order to contribute to the field’s understanding of how best to leverage AI tools to promote K-12 EFL students’ writing proficiency, this study involved a systematic literature review focusing on the theme of optimizing AI-integrated scaffolding to enhance both student agency and creativity. By examining how AI is currently used to support critical components of the writing process, such as idea generation, content organization, and feedback, this review seeks to answer the following guiding question: how do AI-integrated tools support the development of student agency and creativity in EFL students’ writing in a K-12 classroom?

1.1. Theoretical Framework

To understand how AI can support writing education without compromising student autonomy, this literature review draws on two foundational theories: sociocultural theory and cognitive theories of writing. These theories frame AI as a tool that must be integrated thoughtfully to both scaffold student learning and respect the cognitive and social processes inherent in writing development.

1.1.1. Sociocultural Theory and Collaborative Writing

Sociocultural theory, as developed by Vygotsky [16], posits that learning is a socially mediated process where cultural tools and social interactions play a central role. In the context of writing instruction, AI can function as a “machine-in-the-loop” collaborator, positioning students as active decision makers while AI provides structured support. This approach aligns with Vygotsky’s principle of scaffolding, which Rogoff [17] further developed through her concept of guided participation, where learners develop proficiency in a target skill, such as writing, through both direct feedback and implicit mentorship experiences. In writing contexts, AI tools can serve as procedural facilitators [18], providing strategic support that helps make the invisible processes of writing visible and manageable for students.

By offering suggestions, structure, and feedback traditionally provided by teachers or peers, AI tools like ChatGPT (https://openai.com/index/chatgpt/, accessed on 16 April 2025) and QuillBot (https://quillbot.com/, accessed on 16 April 2025)) can enhance the collaborative nature of the writing process without diminishing students’ ownership of their work [19]. This aligns with Rogoff’s [17] notion of apprenticeship in thinking, where learners develop expertise through guided participation in culturally organized activities. The AI system can function as both an expert guide and a collaborative partner, adjusting its level of support based on the student’s developing capabilities.

Furthermore, past research highlights how AI-enabled computer-mediated collaborative environments support the development of “group cognition,” where understanding and meaning are jointly constructed through the features and interactions that the online tools facilitate [20]. Englert’s concept of procedural facilitation [18] provides a framework for understanding how AI can scaffold the writing process by making expert strategies visible and accessible to novice writers. This collective scaffolding encourages students to engage more deeply with content while AI provides accessible, structured feedback, maintaining the collaborative and participative spirit that sociocultural theory advocates.

Nevertheless, the effective integration of AI in these ways requires that educators structure interactions thoughtfully, ensuring that students remain the primary authors and decision makers in their writing. Rogoff’s [17] emphasis on the transformation of participation reminds us that learning occurs through the changing patterns of involvement in sociocultural activities. Therefore, AI support should be designed to gradually transfer responsibility to students as they develop greater competence. When implemented in this manner, AI’s contributions may act as complementary aids, reinforcing students’ sense of agency and promoting a supportive, guided learning experience [21].

1.1.2. Cognitive Theory and Writing Development

Cognitive theories of writing, especially those of Flower and Hayes [22] and Bereiter and Scardamalia [23], examine the mental processes involved in writing, such as ideation, planning, organization, and revision. These models depict writing as a recursive, non-linear process requiring the simultaneous management of various cognitive tasks, from conceptualizing ideas to refining language choices. AI tools, which streamline lower-order tasks (e.g., grammar correction) and facilitate higher-order tasks (e.g., organizing content), align well with cognitive writing models by freeing cognitive resources, enabling students to focus on critical thinking and creativity [24].

From a cognitive perspective, AI functions as a scaffold that helps manage cognitive load, making the writing process more accessible, especially for young or multilingual students. For instance, Natural Language Generation (NLG) tools provide ideation prompts that foster creativity while still requiring students to independently select and develop ideas, preserving their agency [25]. By offloading some cognitive tasks, AI supports students in dedicating more mental resources to content development and organization, which are critical for high-quality writing [26].

1.1.3. Integrating Sociocultural and Cognitive Perspectives

Integrating sociocultural and cognitive perspectives provides a robust framework for understanding AI’s role in writing instruction, positioning AI as both a collaborative partner and a cognitive scaffold. Sociocultural perspectives, grounded in Vygotsky’s theories, emphasize the importance of collaborative interactions and external tools in supporting learning, framing AI as a facilitator of dialogic exchanges where students can interact and negotiate meaning, enhancing their cognitive skills. From a cognitive standpoint, AI tools can streamline lower-order tasks and assist with higher-order functions, aligning with theories by Flower and Hayes [22] and Bereiter and Scardamalia [23], which underscore the importance of freeing cognitive resources for creativity and critical thinking [24].

However, for optimal effectiveness, AI tools must support rather than supplant the cognitive and social dimensions of writing. Research advocates for designing AI tools that engage students as active participants, promoting critical thinking and self-regulation [27]. By combining sociocultural and cognitive theories, this framework emphasizes a balanced approach where AI acts as a complementary support system, preserving student agency while scaffolding their learning. This theoretical foundation informs the central question of this review: how do AI-integrated tools support the development of student agency and creativity in EFL students’ writing in a K-12 classroom? This dual-theoretical approach highlights that the integration of AI in writing instruction must consider both cognitive demands and the social nature of learning, thereby creating an environment where students are empowered, engaged, and creatively autonomous.

2. Materials and Methods

This systematic literature review follows the Preferred Reporting Items for Systematic Reviews and Meta-Analyses [28] guidelines to identify, analyze, and synthesize recent studies on the integration of AI in EFL writing in K-12 classrooms. As this review employed a qualitative synthesis approach rather than a meta-analysis, formal study-level risk-of-bias assessments (e.g., using quantitative bias assessment tools) were not conducted. Instead, studies were included based on their alignment with clearly defined inclusion and exclusion criteria. This approach is consistent with qualitative systematic review standards where thematic rigor, transparency of methodology, and theoretical alignment are prioritized. Similarly, no standardized effect size measures (e.g., risk ratios or mean differences) were reported or synthesized across studies, as the intent of this study was to offer a thematic summary of ways that AI-integrated tools might be employed within K-12 EFL writing instruction to maintain, rather than undermine, students’ creativity, agency, and cognitive engagement. The first author independently reviewed and coded all included studies and conducted the initial thematic synthesis. Themes were then discussed collaboratively with the second author to ensure conceptual clarity and consistency. No automation tools were used in the screening, coding, or data extraction processes, and no contact with study investigators was made for clarification or additional data.

2.1. Inclusion Criteria

Studies were selected based on the following inclusion criteria:

• Population: Research focusing on K-12 students and/or teachers in the English as a Foreign Language (EFL) context. For this review, EFL refers to students learning English in countries where English is not the dominant language.
• Focus on AI Integration: Studies examining AI tools specifically designed for writing instruction, such as Natural Language Generation (NLG) tools, Automated Essay Scoring (AES), paraphrasing tools, and AI-driven feedback systems.
• Outcome Relevance: Studies addressing cognitive engagement, creativity, scaffolding, student autonomy, or teacher perspectives on AI use in writing.
• Publication Date: Only studies published between 2018 and 2024 were included to ensure relevance to recent advancements in AI technology.
• Language: Only studies published in English to ensure consistency in analysis.

2.2. Search Strategy

A comprehensive search was conducted across major academic databases, including Education Source, ERIC, and Web of Science. The following keywords and their combinations were used to capture relevant studies: “AI in writing education,” “AI scaffolding in writing,” “Natural Language Generation (NLG) in education,” “EFL writing,” “student agency with AI tools,” and “cognitive engagement in AI writing tools.” Filters were applied to narrow results by publication date (2018—2024) and source type (peer-reviewed articles).

2.3. PRISMA Flow Diagram

A PRISMA flow diagram was used to systematically illustrate the study selection process, which included four stages: identification, screening, eligibility, and inclusion. The flow diagram visually represents the number of studies retrieved from database searches, those screened based on relevance, those excluded according to inclusion criteria, and the final studies included in this review. See Figure 1.

As shown in Figure 1, an initial pool of 91 records was identified through database searches (ERIC, Education Source, and Web of Science). After removing 21 duplicates and 4 records unrelated to education, the remaining studies were screened by title and abstract. Records were excluded for several reasons, including being outside the publication date range (n = 6), not focused on K-12 contexts (n = 23), or lacking relevance to AI in writing instruction (n = 29). After applying all inclusion criteria, a final set of 14 empirical, peer-reviewed studies published between 2018 and November 2024 (when this search was initiated) was retained for data extraction and analysis.

2.4. Coding of Studies

The selected studies were organized and then coded according to the review’s theoretical framework and research question. Key information collected included the following:

• Study Metadata: Includes authorship, publication year, and setting to contextualize generalizability.
• AI Technologies Implemented: Captures the type, function, and interaction mode of AI tools used in writing instruction.
• Sample: Describes participant characteristics such as age, educational level, and language background to support subgroup analysis.
• Scaffolding Strategies/Design: Identifies how AI tools were embedded instructionally—e.g., feedback design, prompting structure, dialog models, or teacher mediation.
• Writing Activities/Outcomes: Summarizes the writing tasks and assessments used, indicating whether they targeted academic writing, narrative development, or revision.
• Key Findings on Creativity and Agency: Synthesizes outcome patterns linked to learner autonomy, ideation, self-regulation, and student-led decision making.

2.5. Data Analysis

Following the coding process, the data from the 14 selected studies were analyzed using a thematic synthesis approach. A thematic synthesis was chosen to allow the inductive and theory-driven interpretation of qualitative patterns across diverse study contexts. No software was used for thematic analysis; coding and theme development were conducted manually.

The analysis was structured in two major phases. First, studies were clustered by AI tool types and scaffolding strategies, allowing the identification of common instructional patterns (e.g., dialogic scaffolds, structured feedback loops, creative prompt expansion). Then, these findings were organized into three categories of scaffolding approaches: cognitive support tools, creative support systems, and language enhancement tools. This categorization allowed for comparison across different tool functions and instructional designs.

Second, the studies were analyzed for their reported associated effects on student outcomes, with a focus on dimensions of student agency and creativity. Subthemes were identified inductively within each dimension, including autonomy in writing decisions, self-regulated learning behaviors, control over the writing process, ideation support, linguistic experimentation, genre exploration, and creative risk taking. These themes are presented in the Results Section with supporting examples to illustrate how AI scaffolding strategies influence students’ engagement and development as writers.

Cross-case comparisons were also conducted to surface how specific tool designs and implementation contexts influenced these outcomes. For instance, dialog-based systems often promoted exploratory writing, while rigid paraphrasing tools constrained originality. Subthemes were inductively generated within the “Key Findings on Creativity and Agency” category and are reflected in the final table to support direct traceability between coded data and thematic claims.

3. Results

The systematic review of 14 studies published between 2018 and 2024 reveals emerging themes in how AI tools are being integrated into K-12 EFL writing instruction to support student agency and creativity while maintaining effective scaffolding.

Table 1. Included studies with AI technologies implemented and scaffolding strategies used.

Study (Author and Year)	AI Technologies Implemented	Sample	Scaffolding Strategies/ Design	Writing Activities/ Outcome	Findings
Agaidarova et al. (2025) [29]	Write & Improve platform	134 senior-year high school students in Kazakhstan	Provides students with structured writing assignments, initial training on the Write & Improve platform, and immediate automated feedback to support their writing development in an online learning environment; also includes clear guidelines, assessment criteria aligned with IELTS standards, and the use of tools like Grammarly to ensure academic integrity	Using the Write & Improve platform to provide automated feedback on students’ weekly writing assignments, accompanied with traditional teacher feedback	AI feedback was equally effective as teacher feedback in improving writing proficiency, but did not enhance creativity or agency. Tool functioned as a corrective mechanism rather than a scaffold for student ownership or divergent thinking.
Alammar and Amin (2023) [4]	QuillBot, Grammarly, Ginger, SpinnerCeif	25 high school students in Saudi Arabia	Students were trained to use automated paraphrasing tools for several stages of writing academic papers: outlining, drafting, proofreading, and editing	Students submitted writing outcome reviewed and edited with AI tools at each stage for data analysis	Over-reliance on AI limited original content generation, constraining the development of creative agency.
Coenen et al. (2021) [21]	Wordcraft editor: open-ended dialog AI system Meena	8 middle school students in China	Uses a dialog model to enable flexible, task-specific interactions via conversation and then applies few-shot learning to adapt the model to each interaction type without retraining, and also embeds the AI capabilities into a familiar text editor UI for seamless human–AI collaboration during story writing	Students completed informal, self-directed creative writing with interactive AI assistance tool	High creative engagement and agency were observed as students shaped narrative direction through iterative prompting.
Hawanti and Zubaydulloevna (2023) [30]	ChatGPT	73 high school students in Indonesia	15 min per session to interact with ChatGPT to receive instant feedback, grammar correction, idea generation, and writing support, while maintaining structured teacher feedback and peer review sessions	Students were given academic writing assignments on selected topics like Purwokerto City, student organizations, early marriage, etc.	Limited evidence of enhanced creativity or student-led direction
Hsiao and Chang (2023) [31]	Custom AI-powered writing tool: Linggle Write, Linggle Read, and Linggle Search	43 high school students in Taiwan	Teachers demonstrated how to use the AI-powered tools, and facilitated structured discussions to help students interpret AI feedback and apply it during writing and revision tasks	Students practiced using Linggle Write, Read, and Search in real-world tasks, wrote 100-word reflective paragraphs about their experiences, converted them into slides, and gave 2 min presentations.	While revision accuracy improved, creativity was bounded by template-like output and limited student control over feedback integration.
Karim and Mustapha (2020) [32]	Digital mind mapping software	32 high school students in Malaysia	Instructor gives step-by-step instruction on mind mapping techniques and students try writing using it	Students were provided with AI software (iMindMap) to help them brainstorm and organize ideas before moving to later stages of writing	While creativity was scaffolded structurally, agency remained instructor-directed. AI played a minimal interactive role.
Kim et al. (2022) [10]	Custom-built chatbot on Chatfuel platform (Facebook messenger chatbot)	69 11th grade students in South Korea	Three-stage graduated corrective recast approach delivered through an AI chatbot, moving from implicit feedback (repetition and clarification requests) to semi-explicit feedback (elicitation with prompts) to explicit feedback (direct recasts with correct forms)	Students completed Elicited Writing Tasks by constructing English caused-motion sentences based on picture and word cues	Students showed improvement in grammatical precision but limited autonomy. No meaningful development in creative composition or self-directed exploration.
Lee and Maeng (2023) [33]	ChatGPT	30 first-year high school students in South Korea	Gives initial training on using AI chatbots, then gives structured writing assignments and surveys to gauge students’ perceptions of the benefits, concerns, and ethical considerations of utilizing chatbots for English learning	Students independently completed routine homework assignment writing	Some expressed increased independence in editing, but overall writing performance remained the same.
Marzuki et al. (2023) [5]	Grammarly; Quillbot; ChatGPT; Wordtune	38 high school students in Indonesia	Allows students to obtain real-time feedback, grammar corrections, vocabulary suggestions, and writing assistance, while providing teacher guidance on appropriate tool usage and monitoring students’ dependence levels on these tools	Students can freely choose from a range of English writing tasks, such as assignments, projects, or work-related writing	Some students showed increased lexical diversity and confidence, but inconsistent agency. Concerns over AI dependency suggested mixed effects on self-directed development.
Ranalli (2021) [34]	Grammarly Premium	6 high school students in China	Adopt multi-level scaffolding approach that combined technical training with hands-on practice: first teaching students how to code and customize their AI writing tools, then providing instruction on prompting techniques, text evaluation, and integration of AI-generated content into their writing	Student submitted an essay or written assignment that they had previously written for a course, including EFL writing course essays, essay test responses (e.g., history class), blog posts (e.g., for a design course), etc.	Results showed high engagement, increased agency in tool use, and flexible integration of AI-generated text, contributing to both linguistic creativity and learner autonomy.
Song and Song (2023) [7]	ChatGPT	50 12th grade students in China	Three-step learning cycle where teachers first provided structured templates and guidance (Cornell notes, writing/slide templates), then allowed students to practice using AI tools in authentic contexts, and finally transitioned to student-centered presentations—creating a “learning loop” that gradually built learner autonomy through the progressive reduction in support.	Students completed the IELTS writing test	Students gained confidence in using ChatGPT for IELTS tasks. Evidence of growing ownership and self-direction, though creativity limited by the assessment format.
Utami et al. (2023) [35]	Write & Improve and AI Kaku, Eskritor, Grammarly, Plot Generator, Poem Generator, Speech-to-Text, Text-to-Speech, Smodin	58 high school students in Indonesia	Teachers present and simulate how to use the tools, after which students worked in small groups to explore and apply the tools’ features like idea generation, introduction writing, and text expansion to help them with the writing tasks.	Students worked in groups of 2–4 to complete academic/scientific papers, progressing through the stages of planning, drafting, revising, and editing	Collaborative use promoted shared agency and some novel language generation, indicating a modest impact on both dimensions.
Woo et al. (2023) [36]	Four custom NLG tools developed on Hugging Face platform: Next Sentence Generator; Next Word Generator; Next Paragraph Generator 1; Next Paragraph Generator 2	4 secondary school students in Hong Kong, China	Provides students access to various NLG tools in voluntary creative writing workshops, where students can freely experiment with prompting the AI tools with their own text to generate new ideas, then they evaluate and selectively incorporate the AI-generated words and sentences into their stories.	Each student wrote an original English short story (maximum 500 words) using both their own words and AI-generated text, created through prompting custom-built Natural Language Generation (NLG) tools.	A high creative output and the self-guided integration of AI-generated text showed strong gains in both agency and creative expression.
Woo et al. (2024) [6]	Custom AI-NLG tools using Hugging Face language models	23 secondary school students in Hong Kong, China	Two-workshop approach where students first learned technical skills for creating and customizing AI writing tools, followed by practical instruction on writing strategies including prompting techniques, text evaluation, and the integration of AI-generated content into their stories.	Students completed creative short story writing	Students exercised significant agency in deciding when and how to use AI output, enabling deep engagement with their own voice.

provides an overview of the selected studies, highlighting the specific AI technologies implemented and the scaffolding strategies employed in each case.

Table 1 illustrates the diverse range of AI tools being utilized in EFL writing contexts, from general-purpose language models like ChatGPT to specialized writing assistance tools such as Grammarly (https://www.grammarly.com/), QuillBot, and custom-built Natural Language Generation (NLG) systems. The scaffolding strategies documented in these studies demonstrate varying approaches to maintaining student agency while leveraging AI capabilities, ranging from structured teacher-guided implementations to more flexible, student-centered approaches.

The current research landscape shows significant geographical and contextual imbalances. While East Asian contexts, particularly China, are well represented, there is limited research on other EFL contexts, especially in developing countries where technological infrastructure and access may differ substantially. Furthermore, most studies have focused on urban educational settings, leaving rural and resource-constrained environments underexplored. This gap is particularly significant given that the effectiveness of AI scaffolding may vary considerably across different socio-economic and technological contexts.

Figure 2 reveals a clear trend of increasing research interest in the integration of AI in EFL writing instruction, particularly after 2022. There is minimal activity from 2020 to 2022. However, 2023 marks a significant surge, with eight studies published—over half of the total sample—indicating a major uptick in scholarly attention, likely driven by the widespread adoption of advanced AI tools such as ChatGPT. The number slightly drops in 2024, with four studies published as of November, though this still reflects a sustained research interest compared to earlier years. This trend aligns with the technological advancements, corresponding with the release and integration of more capable large language models, prompting a wave of studies investigating their educational use.

The rapid evolution of AI technology presents a unique challenge for research currency. Many studies conducted before 2023 examined AI tools that have since been superseded by more advanced versions or entirely new platforms. For instance, research on earlier versions of automated writing assistants may not fully reflect the capabilities and implications of current large language models like GPT-4. This rapid technological change creates an ongoing need for updated research that addresses the latest AI developments and their educational applications.

3.1. AI Scaffolding Approaches in EFL Writing

The analysis of these studies reveals three distinct categories of AI-integrated scaffolding approaches in K-12 EFL writing contexts: cognitive support tools, creative support systems, and language enhancement tools. Each category demonstrates unique affordances for supporting student writing development while maintaining learner agency and engagement.

3.1.1. Cognitive Support Tools

The analysis of the selected studies highlights the growing role of AI tools as cognitive supports in EFL writing classrooms. These tools are primarily used to reduce the mental load associated with writing by assisting students in areas such as grammar correction, idea organization, and drafting efficiency. By offloading lower-order concerns, AI tools enable learners to concentrate on more cognitively demanding aspects of writing, such as planning, content development, and revision, thereby supporting deeper engagement and improved writing outcomes.

Several studies demonstrate how cognitive scaffolding was implemented across different instructional designs. For example, Song and Song [7] used ChatGPT in a high school EFL context, where students were guided through a three-stage learning loop involving templates, teacher support, and student-centered practice. Students were found to engage actively with AI-generated suggestions, selectively modifying or rejecting content, which indicated maintained agency and thoughtful decision making. Similarly, Alammar and Amin [4] examined the use of Grammarly and QuillBot in the Saudi Arabian context. Their findings revealed that students used the tools not just for mechanical correction, but to reduce language-related stress during drafting. This cognitive relief allowed them to invest more energy into organizing ideas and improving coherence, supported by metacognitive reflection on tool use.

Longer-term implementation was explored in Hsiao and Chang [31], who conducted an 18-week online course for Taiwanese high school students using Linggle Write, Read, and Search. Teachers scaffolded students’ interaction with these tools through explicit demonstrations, structured writing templates, and guided peer presentations. Students used AI support to manage grammar and vocabulary issues independently, which enabled them to dedicate more attention to overall writing structure and message clarity. Reflective writings and flow questionnaires suggested that students experienced increased engagement and confidence in managing the writing process autonomously.

Agaidarova et al. [29] similarly utilized the Write & Improve platform to support Kazakhstani students preparing for the IELTS writing task. The platform’s automated feedback, in conjunction with teacher support and tools like Grammarly, created a hybrid scaffolding system that encouraged iterative revision and reduced dependence on teacher correction. This structure allowed students to work at their own pace while developing self-monitoring strategies.

The positive effect of cognitive scaffolding on student confidence was further supported by Marzuki et al. [5]. Their mixed-methods study revealed that tools such as QuillBot and Wordtune not only improved content organization but also bolstered students’ belief in their ability to complete complex writing tasks. Teachers also played a role in guiding appropriate tool use and helping students avoid over-reliance.

Finally, Ranalli [34] adopted a multi-level scaffolding approach in which high school students received technical instruction on customizing and prompting AI tools, including Grammarly Premium. This hands-on approach helped students develop both procedural and metacognitive skills, allowing them to make more informed decisions about integrating AI feedback into their writing.

3.1.2. Creative Support Systems

Although a less frequently documented application in the literature, the integration of AI-powered creative support systems represents a significant and promising advancement in the scaffolding of EFL writing instruction. These tools enable learners to engage more deeply with ideation, storytelling, and genre experimentation—areas that are often challenging for students with limited linguistic resources. AI tools serving as creative partners offer not just assistance but inspiration, helping students explore alternative expressions, narrative pathways, and stylistic experimentation while preserving their authorship.

For example, Kim et al. [10] investigated a custom-built chatbot designed to support narrative construction in a South Korean high school setting. The study found that iterative feedback loops between AI and students helped build more complex storylines and sustained engagement throughout the writing process. Importantly, AI suggestions acted as creative catalysts rather than directive templates, allowing students to maintain their creative agency and decision-making power during composition.

Similarly, Woo et al. [36] explored student use of four custom Natural Language Generation (NLG) tools in a voluntary creative writing workshop. Students were given autonomy to interact with tools like the Next Sentence Generator and Next Paragraph Generator, evaluating and selectively incorporating AI-generated content into their narratives. The findings revealed a high degree of experimentation, with students blending human-authored and AI-generated text to enhance plot coherence, character development, and stylistic depth. The open-ended and dialogic nature of these tools encouraged creative risk taking, particularly among students who previously expressed anxiety about storytelling in English.

Additionally, Coenen et al. [21] examined the use of a dialog-based AI system embedded in a familiar text editor interface. Middle school students in China used the tool to co-construct creative texts in a flexible, self-directed environment. The few-shot learning approach enabled AI to adapt to different tasks without retraining, which promoted fluid collaboration between human and machine during writing. Students valued the tool for its conversational support, using it to refine narratives, expand ideas, and experiment with genre conventions.

3.1.3. Language Enhancement Tools

The third category encompasses tools designed to enhance language use and stylistic sophistication in EFL writing. Karim and Mustapha’s [32] examination of AI-powered vocabulary enhancement systems showed significant improvements in lexical diversity among users. This finding was complemented by Marzuki et al.’s [5] investigation of style variation tools like Wordtune, which demonstrated enhanced ability among students to adapt their writing tone and style according to different contexts.

Ranalli [34] provided important insights into the role of automated grammar support, finding that precise feedback on language use led to more sophisticated sentence structures without compromising student agency in the revision process. This aligns with Song and Song’s [7] observations of improved lexical diversity and syntactic complexity among students using AI-enhanced writing tools.

3.2. Influence of AI Tools on Student Agency in EFL Writing

The analysis of recent studies reveals four key dimensions through which AI tools influence student agency in EFL writing contexts: autonomy in writing decisions, self-regulated learning behaviors, control over the writing process, and voice and identity expression.

3.2.1. Autonomy in Writing Decisions

Recent empirical studies demonstrate that AI integration significantly influences students’ decision-making autonomy in EFL writing contexts. In a study of 23 secondary school students in Hong Kong, Woo et al. [6] asked participants to use AI-generated text during a creative story-writing task. Students had full control over whether to prompt, accept, revise, or reject AI-generated suggestions, and their final stories reflected varying levels of AI integration—ranging from minimal to extensive use. By analyzing student writing products, which distinguished between human- and AI-written text, and applying rubric-based evaluations alongside cluster analysis, the study revealed that students made independent and strategic decisions about how to use AI. Higher-performing students, in particular, demonstrated greater agency by selectively incorporating AI output to enhance content and organization, rather than relying on it passively. This finding is also supported by Marzuki et al.’s [5] mixed-methods investigation, which revealed that students perceived AI tools as empowering for content creation and revision processes. However, their study also highlighted an important tension: while students felt more confident in their writing decisions, some expressed concerns about becoming overly dependent on AI assistance.

3.2.2. Self-Regulated Learning Behaviors

The integration of AI tools appears to foster self-regulated learning behaviors among EFL students. Song and Song [7] conducted a comprehensive mixed-methods study combining pre/post-tests with interviews, finding that AI tools like ChatGPT encouraged self-regulation by improving both writing skills and motivation. Their findings align with Hsiao and Chang’s [31] study, which documented enhanced cognitive, emotional, and behavioral engagement through AI tool usage. This study particularly emphasized how AI assistance helped students adopt effective learning strategies, maintain positive attitudes, and actively engage with language tasks while addressing concerns about over-reliance on technology.

3.2.3. Control over Writing Process

Research indicates that AI tools can significantly enhance students’ control over various aspects of the writing process. Marzuki et al.’s [5] study, using teacher interviews, found that tools like Quillbot and Wordtune enhanced students’ ability to manage content creation and organizational clarity independently. Similarly, Utami et al. [35] found that senior EFL students demonstrated increased autonomy in managing and refining academic writing when using multiple AI tools. These findings suggest that AI integration can enhance students’ agency in both structural and content-related aspects of writing.

3.2.4. Voice and Identity Expression

Recent studies highlight the role of AI tools in supporting students’ expression of voice and identity in EFL writing. Marzuki et al.’s [5] mixed-methods investigation found that tools like Wordtune supported expressive writing, enhancing students’ ability to convey individual voice while maintaining academic standards. This finding is complemented by Utami et al.’s [35] case study with Indonesian high school students, which revealed that AI tools enabled students to experiment with different narrative structures and writing styles, leading to more personalized expression in their writing.

3.3. Influence of AI Tools on Student Creativity in EFL Writing

The analysis of selected studies revealed four key dimensions through which AI scaffolding supports creative writing development in EFL contexts: ideation support, linguistic experimentation, genre exploration, and creative risk-taking behaviors.

3.3.1. AI-Enhanced Ideation and Brainstorming

Recent studies demonstrate AI tools’ significant influence on idea generation and development in EFL writing. Marzuki et al. [5] found that AI writing assistants like Wordtune effectively supported the brainstorming process by offering multiple rewrite options, helping students maintain writing flow while exploring diverse expressions. This finding was corroborated by Woo et al. [36], whose investigation of AI-generated text revealed comparable benefits for both high- and low-performing students in terms of content development and organizational clarity. Notably, the democratizing effect of AI support on creative ideation suggests its potential for reducing achievement gaps in EFL writing classrooms.

3.3.2. Novel Language Combinations

Studies indicate that AI scaffolding facilitates sophisticated linguistic experimentation among EFL learners. Song and Song [7] documented how ChatGPT interactions led to improved vocabulary diversity and sentence structure complexity. Similarly, Hsiao and Chang [31] found that personalized AI suggestions through tools like Linggle Search enhanced students’ ability to create cohesive and contextually appropriate language combinations. These findings suggest that AI scaffolding can expand learners’ linguistic repertoire without altering the original content of their writing.

3.3.3. Genre Exploration and Development

AI tools have shown particular promise in supporting genre experimentation. Coenen et al. [21] examined professional writers’ use of AI assistants like Wordcraft for genre exploration and world building, highlighting both opportunities and challenges in maintaining authorial voice. Similarly, Woo et al. [6] demonstrated how generative AI tools helped students overcome creative barriers in story writing across various genres. The studies collectively suggest that AI scaffolding can expand students’ genre awareness and production capabilities while raising important questions about voice authenticity.

3.3.4. Support for Creative Risk Taking

Research indicates that AI scaffolding can foster creative risk taking in EFL writing. Hawanti and Zubaydulloevna [30] found that AI-powered tools encouraged students to experiment with unfamiliar language structures and styles, reducing anxiety around linguistic experimentation. This finding aligns with Karim and Mustapha’s [32] work on reflective thinking mechanisms in AI writing environments, which showed improved creative outcomes and increased self-efficacy among EFL learners. The reduction in cognitive load appears to create space for greater creative experimentation.

4. Discussion

4.1. Synthesis of Key Findings and Implications

This systematic review of AI-integrated scaffolding in K-12 EFL contexts reveals several interconnected themes that illuminate how AI tools can effectively support writing development while preserving student agency and creativity. Through the analysis of recent studies (2018–2024), three significant themes emerge in the successful integration of AI tools: the complementary relationship between cognitive support and creative development, the dynamic balance between AI scaffolding and student autonomy, and the critical alignment between theoretical frameworks and practical implementations.

4.1.1. Complementary Relationship Between Cognitive Support and Creative Development

The findings demonstrate that AI tools’ cognitive support functions can enhance rather than inhibit creative development when properly implemented. Studies by Song and Song [7] and Woo et al. [36] reveal that, by reducing cognitive load in areas such as grammar correction and basic organization, AI tools free up mental resources for higher-order creative tasks. This relationship aligns with cognitive writing theories [22] while extending their application to AI-supported environments. Notably, when AI tools manage lower-order concerns, students demonstrate increased willingness to engage in creative risk taking and experimentation with complex language structures [30].

However, this complementary relationship appears contingent on thoughtful implementation. Studies by Alammar and Amin [4] indicate that the excessive reliance on AI support can potentially undermine this benefit, suggesting the need for the careful calibration of AI assistance levels. The implications suggest that educators should strategically introduce AI tools in phases, gradually increasing complexity while maintaining focus on creative development.

4.1.2. Dynamic Balance Between AI Scaffolding and Student Autonomy

A second key theme emerges in the dynamic relationship between AI scaffolding and student autonomy. Research consistently shows that effective AI integration requires maintaining a delicate balance between support and independence. Marzuki et al.’s [5] findings demonstrate that students exhibit increased confidence and writing fluency when AI tools serve as consultative resources rather than directive authorities. This aligns with Vygotsky’s concept of the zone of proximal development, where AI functions as a more capable peer while preserving student agency in the learning process.

The implications of this balance are particularly significant for EFL contexts, where language barriers often complicate the writing process. Utami et al. [35] revealed that when AI tools are positioned as collaborative partners rather than authoritative sources, students maintain stronger ownership of their writing while benefiting from linguistic support. This suggests the need for pedagogical frameworks that explicitly position AI tools as resources for consultation rather than sources of definitive answers.

4.1.3. Alignment Between Theory and Practice

A third significant theme emerging from the review concerns the alignment between theoretical frameworks and practical implementations of AI tools in EFL writing instruction. Successful integration often reflects principles from both sociocultural and cognitive theories, even when these frameworks are not explicitly acknowledged in the studies. Implementations that yield positive outcomes tend to support both the social dimensions of learning and the cognitive demands of writing. This implicit alignment carries important implications for future implementation strategies.

The findings suggest that the effective integration of AI tools requires explicit attention to how these technologies support both cognitive processes and social interactions in writing development. Additionally, there must be clear frameworks for transitioning levels of support as students grow more independent in their writing capabilities. Rather than adopting entirely new instructional models, successful integration depends on the strategic embedding of AI tools within existing pedagogical approaches.

These insights highlight the need for a nuanced understanding of how scaffolding, autonomy, and engagement interact in the writing process. Tool selection alone is insufficient; broader considerations related to implementation strategy, instructional alignment, and responsiveness to student development are critical. Furthermore, the review indicates that successful integration is contingent not only on the capabilities of the AI tools themselves but also on the educational context in which they are employed. This underscores the need for comprehensive professional development programs that equip educators to navigate both the technical functionalities and pedagogical implications of AI. Finally, the research points to the importance of ongoing assessment and recalibration of support to sustain an optimal balance between guidance and independence.

4.2. Pedagogical Implications

The integration of AI tools in EFL writing instruction presents both opportunities and challenges that require the careful consideration of pedagogical approaches. The analysis of the reviewed literature reveals several key implications for teaching practice that can help optimize AI integration while preserving EFL students’ agency and creativity as they gain needed language skills required for access to advanced schooling, access to higher education, and long-term success in the global workforce.

4.2.1. Teacher Mediation and Scaffolding Strategies

The findings emphasize the critical role of teachers in effectively mediating the use of AI tools in writing instruction. Song and Song [7] demonstrate that the successful integration of AI depends largely on teachers’ ability to design writing tasks that harness AI capabilities while sustaining students’ cognitive engagement. This pedagogical approach reflects Vygotsky’s concept of scaffolding, wherein instructional support must be carefully calibrated to align with students’ zones of proximal development. A staged approach to AI integration is recommended, gradually introducing tools and features as students build writing proficiency and autonomy.

Research by Marzuki et al. [5] and Alammar and Amin [4] further identifies specific strategies for effective teacher mediation. These include providing explicit instruction on when and how to use various AI features, regularly monitoring the impact of AI on students’ writing processes, and implementing structured reflection activities that encourage students to articulate their decision making when using AI. Additionally, teachers are advised to strategically combine AI-generated feedback with their own to offer students complementary and balanced perspectives on their writing.

4.2.2. Balancing AI Support with Student Agency

A critical pedagogical consideration emerging from the research is the need to maintain an appropriate balance between AI support and student autonomy. Woo et al. [36] and Kim et al. [10] emphasize the importance of designing writing tasks that require students to engage critically with AI-generated suggestions rather than accepting them passively. To support this goal, several instructional practices are recommended. First, writing protocols can be implemented that require students to justify their acceptance or rejection of AI suggestions, thereby fostering critical engagement. Second, tasks can be designed to allow students to compare multiple AI-generated alternatives, encouraging informed decision making. Third, assessment rubrics should explicitly value original thinking and creative expression alongside technical proficiency, reinforcing the importance of student voice. Finally, educators should promote metacognitive reflection on the role of AI in the writing process, helping students become more aware of how these tools influence their writing decisions.

4.2.3. Differentiated Implementation Approaches

The research reveals the necessity of differentiated approaches to AI integration based on student proficiency levels and learning needs. This suggests a need for customized AI tool configurations tailored to individual student profiles, allowing the technology to better support diverse learning trajectories. Implementation strategies must remain flexible to accommodate varying levels of language proficiency, ensuring that students at different stages of development can engage meaningfully with AI tools. Additionally, targeted support should be provided for students who may face challenges with technology integration, helping to prevent further marginalization. Finally, the ongoing assessment of AI tool effectiveness across different student populations is essential to ensure that the tools are meeting learners’ needs and contributing positively to their educational outcomes.

4.2.4. Creating Supportive Learning Environments

The findings indicate that the effectiveness of AI integration in EFL writing instruction is heavily influenced by the broader learning environment. Drawing on sociocultural theory and evidence from the reviewed studies, several key environmental factors emerge as critical to successful implementation. First, establishing clear guidelines for the appropriate use of AI tools helps ensure that students understand both the affordances and limitations of these technologies. Second, creating collaborative learning opportunities that integrate AI support with peer interaction enhances the social dimension of learning, allowing students to negotiate meaning and build knowledge collectively. Third, fostering classroom cultures that value both creativity and technical proficiency encourages students to engage with AI tools not just for functional support but also as a means of creative expression. Finally, providing regular opportunities for students to give feedback on the usefulness and impact of AI tools promotes reflective use and helps educators make informed adjustments to instructional design.

These pedagogical implications underscore the complexity of integrating AI tools effectively in EFL writing instruction. Success requires careful attention to multiple factors, from teacher preparation to assessment design, all while maintaining focus on the primary goal of developing students’ writing abilities and maintaining their agency in the learning process.

4.3. Limitations and Future Research Directions

4.3.1. Methodological Limitations

The analysis of the reviewed literature reveals several significant methodological constraints that merit attention. First, many studies relied heavily on short-term implementations of AI tools, typically spanning only one academic semester or less [5,7]. This limited temporal scope restricts our understanding of how AI scaffolding influences long-term writing development and student agency. Additionally, the predominance of self-reported data through surveys and interviews, while valuable for understanding student perceptions, may not fully capture the actual effect of AI tools on writing quality and cognitive development. Moreover, few studies employed rigorous experimental or quasi-experimental designs; instead, most studies used qualitative or correlational approaches, such as interviews, surveys, or observational data. In the absence of controlled research designs, it is not possible to establish causal links between AI tool use and observed improvements in writing outcomes, creativity, or learner autonomy.

Another notable limitation across studies is the inconsistent operationalization of key constructs such as “student agency” and “creativity” in AI-assisted writing contexts. While studies like Song and Song [7] and Alammar and Amin [4] attempted to measure these constructs, the field lacks standardized metrics for evaluating how AI integration influences student autonomy and creative expression. This inconsistency makes it challenging to draw definitive conclusions about the effectiveness of different AI scaffolding approaches.

In addition to the limitations found in the studies reviewed, this systematic review itself has several methodological constraints that should be acknowledged. First, the final inclusion of only studies conducted in K-12 EFL contexts may limit the generalizability of the findings to broader second language (ESL) or first language (L1) writing instruction. While the EFL population provides a valuable perspective on how AI scaffolding supports language learners, it remains unclear to what extent these findings are transferable to native English-speaking students or multilingual learners in English-dominant educational settings. Second, the inclusion criteria prioritized studies published in English, which may have led to the exclusion of relevant research published in other languages, especially from underrepresented regions. Additionally, although the review employed a systematic search strategy, it is possible that some relevant studies were missed due to variations in keyword indexing or publication venues. It is also important to acknowledge that this is a rapidly expanding space for both research and technology, necessitating future reviews that build on the methods and findings of the present study. Lastly, because the review focused specifically on the writing-related applications of AI, studies examining other AI-supported language skills, such as reading or speaking, were excluded, even though those dimensions may indirectly influence writing development. These methodological boundaries narrow the scope of generalizability but also highlight areas for future exploration.

An additional limitation of this review is that it did not incorporate formal coding or appraisal of study quality. While all included studies were peer-reviewed, we did not apply systematic criteria to assess methodological rigor or risk of bias. This omission reflects a broader trend in recent AI-in-education syntheses, where conclusions have sometimes been distorted due to over-reliance on descriptive synthesis without evaluating study credibility. As recent critiques have noted, failure to consider study quality can lead to overstated findings or the misinterpretation of AI’s effectiveness [37]. Future reviews should integrate quality assessment tools to contextualize findings more precisely and guide the interpretation of heterogeneous study designs.

4.3.2. Future Research Directions

Future research on AI-integrated scaffolding in writing instruction should pursue several promising directions. First, expanding the contextual scope to include ESL (English as a Second Language) settings is critical. Studies in English-dominant countries such as the United States, Canada, and the United Kingdom could provide insights into how immersion in English-speaking environments influences students’ engagement with AI tools. Unlike EFL learners, ESL students navigate contexts where English is both the instructional and social language, introducing distinct cultural, cognitive, and linguistic dynamics. Comparative studies between EFL and ESL populations could help determine how contextual variables—such as linguistic exposure and educational infrastructure—shape the effectiveness of AI scaffolding, and whether adaptations are necessary for different learner populations.

Second, there is a clear need for longitudinal research that tracks the sustained impact of AI use on students’ writing development. Such studies can shed light on the evolution of student agency and creativity over time and examine whether long-term exposure to AI tools leads to dependency or fosters greater independence. Third, future work should include comparative analyses using rigorous experimental or quasi-experimental designs to determine the relative effectiveness of different AI tools, feedback strategies, and interface designs. These comparisons should consider age, proficiency level, and cultural background to identify scalable and adaptable practices.

Equity and access also remain pressing concerns that future research must systematically address. While AI-integrated tools hold promises for enhancing writing instruction, their benefits may not be equitably distributed across diverse educational environments. Studies should examine how these tools can be adapted for implementation in resource-constrained contexts, particularly in underfunded schools, rural areas, and low-income countries where infrastructure, device access, and digital literacy may be limited [38]. Design-based research is needed to explore lightweight, offline-compatible, or mobile-first AI writing supports that accommodate infrastructural constraints without sacrificing the core pedagogical value.

Finally, future studies must critically examine the emerging risk of AI dependency in writing tasks. While scaffolding is essential, over-reliance on automated feedback can diminish students’ sense of ownership, hinder metacognitive development, and reduce opportunities for productive struggle—an important component of learning [39]. Prior studies have noted that excessive automation in feedback processes may undermine learners’ ability to internalize writing strategies and transfer skills across contexts [40]. Future research should investigate how different levels and types of AI scaffolding affect learners’ autonomy, self-efficacy, and writing confidence over time, especially in formative writing contexts. Longitudinal and experimental studies are needed to determine the optimal calibration of support—balancing immediate guidance with gradual withdrawal to foster internalized strategies. Designing AI systems that can detect student readiness and adjust scaffolding dynamically represents a promising but underexplored direction for supporting sustainable, learner-centered writing development.

5. Conclusions

Through analyzing the synthesized findings from the 14 reviewed studies, this systematic review reveals that AI-integrated tools support EFL students’ writing development in secondary education by functioning as cognitive, linguistic, and creative scaffolds—but their impact on learner agency and creativity is contingent on instructional design, tool affordances, and learner interaction patterns. Specifically, dialog-based AI systems and tools with customizable prompting interfaces were more effective in fostering creative expression and autonomy than static, correction-focused tools. This review also finds that the degree of student training, teacher mediation, and alignment of tools with authentic writing goals are critical factors mediating the effectiveness of AI in sustaining student-centered learning. Furthermore, while some studies reported promising outcomes in student ideation, genre exploration, and self-regulated writing, others highlighted risks of over-reliance and limited transformative impact when scaffolds were overly prescriptive. These findings emphasize that optimizing AI in writing instruction requires not only technological adaptation but also pedagogical intentionality to preserve and promote learners’ cognitive and creative agency.