A Meta-Survey of Generative AI in Education: Trends, Challenges, and Research Directions

Abstract

Education is experiencing a paradigm shift, evolving from traditional learning methods to computer-tool-based education, and now toward the integration of Generative Artificial Intelligence. While classical methods offer structured and standardized learning, they often do not fully address individual learner needs and accessibility. The rise of digital technologies introduced adaptive learning platforms, online classrooms, and interactive educational tools, expanding the reach and flexibility of educational systems. Today, Generative Artificial Intelligence tools are redefining the education landscape by personalized learning experiences, automating content generation, and providing real-time feedback. Intelligent tutoring systems and personalized assessments empower students with customized learning pathways that enhance engagement and academic performance. This paper presents a meta-survey that systematically examines the role of Generative Artificial Intelligence in education, following PRISMA guidelines to analyze trends, frameworks, and research outcomes across a curated body of academic literature. Special attention is given to the emergence of commercial Generative Artificial Intelligence tools, which are increasingly embedded in learning environments. A structured comparison framework and research questions guide the review, offering insights into how Generative Artificial Intelligence technologies are shaping pedagogical practices, influencing assessment, and raising new ethical and technical challenges. The paper also explores future directions, highlighting how Generative Artificial Intelligence is driving the emergence of new learning models.

1. Introduction

The landscape of educational technology is undergoing a profound transformation, driven by rapid advancements in Generative Artificial Intelligence (Gen AI). While the transition from traditional to digital learning models has been underway for decades, the public release of large language models (LLMs) such as ChatGPT (version 4, OpenAI, released March 2023) in late 2022 accelerated research and development at an unprecedented scale. Gen AI tools now demonstrate significant capabilities in generating educational content, supporting personalized learning, and mimicking aspects of human creativity [1]. These technologies are reshaping pedagogy by enabling adaptive tutoring systems, customized content recommendations, and interactive learning environments that respond to the needs of individual students [2,3].

The evolution of education has historically followed technological advances: from classical methods centered on books and lectures [4,5], to digital and e-learning models that expanded access and flexibility, particularly during the COVID-19 pandemic [6,7], and now to Generative AI. Each phase has reshaped pedagogy in distinct ways, but Gen AI represents a qualitatively new stage [8]. Unlike earlier technologies that primarily delivered or organized content, Gen AI actively generates, adapts, and personalizes learning experiences, raising novel opportunities and challenges for research and practice [2].

This rapid growth has been accompanied by a surge in academic publications. A preliminary analysis indicates that more than 40 survey studies on the integration of Gen AI in education have been published since 2022. However, this proliferation has led to a fragmented research landscape. Many reviews focus on specific tools (e.g., chatbots), single educational levels (e.g., higher education) [9], or isolated themes such as ethics and academic integrity [10]. As a result, there is still no consolidated synthesis that compares definitions, frameworks, and applications across contexts. Moreover, few reviews systematically address the pedagogical and institutional impact of commercial Gen AI tools that are increasingly being adopted in academic practice.

To address these gaps, this paper presents a systematic meta-survey of the existing literature on Gen AI in education. By adopting a tertiary review approach—a survey of surveys—we synthesize fragmented findings into a coherent overview of the state of the art. Our methodology is guided by two principles:

• Comprehensiveness: The review follows the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) guidelines [11], a rigorous and reproducible protocol that ensures transparent selection of high-quality studies.
• Effectiveness: The meta-survey approach aggregates findings from a broad range of secondary studies, enabling generalization and synthesis that make the rapidly expanding research landscape more accessible and actionable.

The structure of the paper is as follows: Section 2 outlines the research objectives and methodology. Section 3 analyzes publication and citation trends from 2022 to April 2025. Section 4 synthesizes the collected literature, focusing on key trends, tools, and frameworks. Section 5 discusses future research directions and emerging learning models enabled by Gen AI. Finally, Section 6 concludes with implications for educators, researchers, and policymakers.

2. Methodology

A systematic analysis of Gen AI applications in education was conducted, covering the period from 2022 to April 2025 (since ChatGpt’s emergence). The study followed the PRISMA guidelines to ensure a comprehensive and structured approach to the literature review and data synthesis, and after completing the record screening and report selection process, we analyzed a total of 40 works. From these, we conducted an additional manual selection, resulting in a systematic analysis literature of 21 papers. The research included the following:

• Original studies that explore new Gen AI applications in education.
• Comprehensive surveys and review papers analyzing theoretical frameworks, educational objectives, and case studies that highlight the opportunities and challenges of integrating Gen AI into education.

The undertaken search plan is summarized in

Table 1. Search plan.

Source	Criteria
Database	IEEEXplore, MDPI, Springer, Elsevier, Scopus, Google Scholar
Date of Publication	2022–2025
Keywords	Generative Artificial Intelligence/Education
	Gen AI tools/ Education
	LLMs/Education
Language	English
Type of Publication	Survey, Review, or SLR
Inclusion Criteria	Papers published after 2022 (after ChatGPT’s emergence)
	Papers must be written in English.
	Papers that are peer-reviewed and published in journals or conference proceedings.
	Comprehensive studies on the use of Gen AI in educational contexts.
	Studies emphasizing Gen AI as a central component of innovative educational solutions.
Exclusion Criteria	Papers before 2022 (before ChatGPT’s emergence)
	Papers not written in English.
	Papers not peer-reviewed.
	Contributions with minimal relevance or limited depth regarding educational applications of Gen AI.
	Papers where “Gen AI” is mentioned without proper context or detailed analysis.

, while Figure 1 depicts the flow-chart of the PRISMA-based selection process.

2.1. Objectives

This review aims to explore the state of the art in Gen AI at the intersection of AI and education. The survey was conducted to analyze current research trends, identify key challenges related to Gen AI, and address the following research questions (RQs):

• (RQ1): What are current definitions of Gen AI ?
• (RQ2): What are the current methodologies and frameworks for integrating Gen AI into educational systems?
• (RQ3): Which educational levels have been the primary focus of research on Gen AI ?
• (RQ4): What are the primary technical, pedagogical, and ethical challenges associated with the adoption of Gen AI in education?
• (RQ5): How effective are Gen AI tools in fostering critical thinking and problem-solving skills among students?

2.2. Search Strategy

The search was conducted in February 2025 using a wide range of keywords, and the dataset was subsequently updated with publication and citation information available up to April 2025. To extract the most relevant documents, search queries were conducted using predefined keywords and Boolean operators across multiple digital libraries. The search aimed to identify scientific publications that explore the application and the role of Gen AI in education. The keyword search string was structured around three key concepts: (i) Generative Artificial Intelligence, (ii) Gen AI tools, and (iii) large language models. Based on these terms and their synonyms, the following search string was formulated:

$$\begin{gathered} \mathbf{(}\mathbf{“}\mathbf{Generative} \mathbf{Artificial} \mathbf{Intelligence}\mathbf{”}\mathbf{OR} \mathbf{“}\mathbf{Gen} \mathbf{AI} \mathbf{tools}\mathbf{”}\mathbf{OR} \mathbf{“}\mathbf{Large} \mathbf{Language} \mathbf{Models}\mathbf{”}\mathbf{)} \\ \mathrm{AND} \mathbf{(}\mathbf{“}\mathbf{Education}\mathbf{”}\mathbf{OR} \mathbf{“}\mathbf{Learning}\mathbf{”}\mathbf{OR} \mathbf{“}\mathbf{Higher}\mathbf{Education}\mathbf{”}\mathbf{)} \end{gathered}$$

(1)

To ensure the identification of relevant results, the search string was applied to article titles with a constraint allowing a maximum gap of two words between key terms. Subsequently, considering the study’s focus on Gen AI research, the following search string was applied to the abstracts of the manuscripts:

$$\mathbf{(}\mathbf{Survey} \mathbf{OR} \mathbf{Review} \mathbf{OR} \mathbf{Literature}\mathbf{)}$$

(2)

2.3. Eligibility Criteria

To ensure that only highly relevant research is included in this analysis, inclusion and exclusion criteria were established. The articles were eligible for selection if they met all of the following inclusion criteria:

• If the work is a survey or literature review on the use of Gen AI in education.
• The work discusses the application of Gen AI in an educational context.
• If the work focuses on AI-generated content, personalized learning, adaptive assessment, or student support.
• If the work proposes a solution or framework involving Gen AI in education.
• If the work is based on Gen AI techniques.
• If the work addresses subject-specific applications (e.g., programming, engineering, etc.) within the broader context of Gen AI in education, rather than as a domain-specific focus alone.
• If the work includes a model or provides a formal definition of Gen AI.

The articles were excluded from the selection if they met one of the following exclusion criteria:

• Studies that discuss Gen AI in general but do not address educational application, such as in healthcare [12,13,14], finance [15,16], and nursing [17,18].
• Studies published in preprints and the gray literature, including theses, dissertations, technical reports, opinions, discussions, editorials, viewpoints, comments, tutorials, lessons, pilot studies, and presentations.
• Research on Gen AI that was not written in the English language [19,20].
• The terms related to Gen AI are mentioned only in the title, abstract, or keywords but are not addressed in the main body of the work [21].
• The study lacks a clear focus on the role of Gen AI in the field of education.
• Unpublished works [21,22].
• Studies focused on AI in education but do not involve Gen AI.

For all studies meeting the eligibility criteria, study metadata (e.g., publication year, venue, methodology, frameworks, and tools) were extracted by the first author and independently verified by the co-authors. Any disagreements regarding data extraction or classification were resolved through discussion until consensus was reached. No automation tools were employed in this process.

2.4. Study Selection

The study selection process followed PRISMA 2020 guidelines, as shown in Figure 1. A total of 130 records were initially retrieved. After removing 33 papers based on predefined technical criteria (e.g., unpublished works, pre-2022 publications, non-English papers, and incomplete content), 97 remained for screening. During the title, abstract, and keyword screening, two authors independently assessed all records, excluding 32 papers that provided only a superficial mention of Gen AI. Disagreements were resolved through discussion with the other co-authors until consensus was reached. A further 23 papers were excluded due to insufficient methodological detail ($n=8$) or limited Gen AI contribution beyond superficial mention ($n=15$). This left 40 papers for full-text review. In the final quality screening phase, which was conducted collaboratively by the research team, 19 papers were excluded for lack of rigor or relevance, leaving 21 secondary studies.

Out of the 40 eligible articles, 21 reviews met all four methodological quality indicators (see

Table 2. Quality assessment of included reviews (based on four indicators of methodological rigor).

Study	Clear RQs/Eligibility	Transparent Search Strategy	Inclusion/ Exclusion Reporting	Coherent Synthesis
[23]	Yes	Yes	Yes	Yes
[24]	Yes	Yes	No	Yes
[25]	Yes	Yes	Yes	Yes
[26]	Yes	Yes	No	Yes
[27]	Yes	Yes	Yes	Yes
[28]	Yes	Yes	Yes	Yes
[29]	Yes	Yes	Yes	Yes
[30]	Yes	No	No	Yes
[31]	Yes	Yes	Yes	Yes
[32]	Yes	Yes	No	Yes
[33]	Yes	No	No	Yes
[34]	Yes	Yes	Yes	Yes
[35]	Yes	Yes	Yes	Yes
[36]	Yes	Yes	No	Yes
[37]	Yes	Yes	Yes	Yes
[38]	Yes	Yes	Yes	Yes
[39]	Yes	Yes	No	Yes
[40]	Yes	Yes	Yes	Yes
[41]	Yes	Yes	Yes	Yes
[42]	Yes	Yes	No	Yes
[43]	Yes	No	No	Yes

) and were therefore included in the final synthesis. These 21 reviews form the basis of our in-depth synthesis and are reported in Table 4.

2.5. Risk of Bias Assessment

To ensure the validity of our synthesis, we assessed the methodological quality of the included secondary studies, guided by the principles of the PRISMA 2020 statement [11] Given that this is a meta-survey, a formal application of tools like ROBIS [44] or AMSTAR-2, which are designed for primary systematic reviews, was not directly applicable. Our objective was not to validate the individual data points within the reviews, but to appraise the methodological soundness of the reviews themselves.

We therefore implemented a qualitative assessment using four indicators of review quality: (i) articulation of research questions and eligibility criteria; (ii) use of a systematic and reproducible search strategy; (iii) transparency in reporting study selection and inclusion/exclusion; and (iv) provision of a coherent synthesis of findings aligned with stated objectives.

This review was conducted in accordance with the PRISMA 2020 guidelines [11]. The review was not registered in a database, and no prior protocol was prepared.

This process allowed us to systematically identify potential sources of bias at the review level. During our synthesis, findings from reviews that lacked transparency in these key areas were interpreted with caution and weighted as less robust in our overall analysis. Our evaluation revealed that 12 of the 21 reviews employed transparent search strategies, fewer than half explicitly reported eligibility criteria, and only 7 documented a formal synthesis approach. Table 2 summarizes the results, showing that while most reviews clearly stated research questions and provided coherent syntheses, the reporting of inclusion/exclusion and methodological transparency was inconsistent. This uneven rigor highlights the need for greater methodological standardization in Gen AI educational reviews.

3. Results

A total of 40 articles met the eligibility criteria. This full set is used to depict overall publication and citation trends (Figure 2 and Figure 3). Following the quality assessment described in Section 2, 21 reviews were retained for the in-depth synthesis presented in Section 4 and Section 5.

From January 2022 to April 2025, monthly output in Generative AI in education rose sharply (Figure 2 and Figure 3). Activity in 2022 is sporadic (three surveys in total). In 2023, production becomes more regular (seven in total), and 2024 exhibits a pronounced surge (twenty), accounting for roughly half of all included surveys. The partial window for 2025 (January–April) already contains 10 surveys, indicating continued momentum; because only four months are observed, totals are not directly comparable to full prior years, but the monthly pace is clearly elevated. This upswing temporally coincides with the mainstreaming and rapid uptake of large generative models beginning in late 2022. The descriptive counts use all 40 eligible works; the interpretive synthesis is restricted to the 21 high-quality reviews (Section 2).

Baseline comparator and monthly data construction: To contextualize these dynamics, we established a monthly baseline from a closely related AI+domain AI in Medical Diagnostics/Health and Medicine. A dedicated search was conducted in Scopus and cross-checked in Web of Science using the keywords “medical” AND “Generative AI” (and their controlled vocabulary equivalents). The retrieved records were filtered using the same eligibility criteria applied in our main review: (i) peer-reviewed journal articles or systematic reviews; (ii) exclusion of editorials, conference abstracts, and non-scholarly notes/letters; (iii) English-language publications; and (iv) availability of complete bibliographic metadata. Records were grouped by publication month to construct the baseline time series; items lacking monthly metadata were excluded from the monthly aggregation. Where month-level citation metadata were incomplete, monthly publication counts were used as a transparent proxy, and this limitation is explicitly noted in the figure captions. We selected this comparator because the AI-in-medicine literature is both mature and extensively characterized in prior bibliometric syntheses [45,46], providing a conservative and credible benchmark against which to interpret the acceleration observed in Generative AI in education. The comparison reveals that, while both domains exhibit steady growth over the observed period, the post-2023 rise in Generative AI in education is significantly steeper, consistent with a domain-specific surge in scholarly attention rather than a background effect of general AI research expansion.

Baseline comparison: Compared to Gen AI in Medical Diagnostics, both domains show strong post-2022 growth. However, Gen AI-in-education exhibits a steeper post-2023 acceleration (especially through 2024), indicating a domain-specific surge rather than merely reflecting a general AI upswing. This conclusion holds when baseline 2025 values are treated as partial/omitted in line with source availability.

Geographic Origin of the Reviewed Studies

To provide a broader understanding of the research landscape, we analyzed the geographic origin of the 21 included studies based on the affiliation of the first author (see

Table 3. Geographic origin of the 21 included studies (first author affiliation).

Region/Country	Number of Studies
China	4
United Arab Emirates	2
USA (incl. California)	2
UK	1
Portugal	1
Spain	1
Denmark	1
Poland	1
Sweden	1
Greece	1
India	1
Bangladesh	1
Qatar	1
Nigeria	1
Malaysia	1
Colombia	1
England	1
Total	21

).

The distribution highlights a concentration of contributions from Asia (China, India, Malaysia, Qatar, and Bangladesh), Europe (the UK, Portugal, Spain, Denmark, Poland, Sweden, Greece), North America (the USA, California), South America (Colombia), the Middle East (United Arab Emirates), and Africa (Nigeria). This uneven distribution suggests that research on Generative AI in education is heavily concentrated in Asia and Europe, while contributions from Africa and South America remain limited.

4. Literature Review

In the context of analyzing the application of Gen AI in the educational sector, there is a growing academic interest in the transformative potential of these technologies. Numerous studies have examined how Gen AI can enhance the educational experience through the development of adaptive and personalized learning tools.

To organize our meta-survey, we grouped the reviewed studies into three categories: general educational applications of Gen AI, research focused specifically on ChatGPT, and studies examining the use of commercial Gen AI tools in educational settings.

4.1. Educational Applications of Gen AI

This category encompasses papers that examine the general applications, educational outcomes, ethical considerations, and integration strategies of Generative AI across various educational levels and domains. These works provide foundational insights into how Gen AI is reshaping modern education.

The current literature highlights Gen AI transformative potential for education landscape, particularly in higher education, where it supports advancements in personalized learning, adaptive feedback, automated assessment, and the development of virtual instructors and intelligent tutoring systems [23,35,40,41]. Several implementation frameworks such as PAIGE, IDEE, and AI-CRITIQUE have been introduced to ensure ethical and effective deployment [23,24]. A recurring theme in this body of work is the ethical dimension of Gen AI use, including concerns over academic integrity, data privacy, intellectual property, and algorithmic bias. These challenges have prompted calls for institutional policies and clear ethical guidelines [24,30,37].

While positive outcomes are widely reported, including improvements in cognitive, technical, and interpersonal skills [41], scholars also caution against over-reliance on AI technologies. They advocate for educator engagement, critical AI literacy, and balanced integration strategies to avoid undermining independent thinking and pedagogical goals [26,33,36]. Notably, a gap persists in research focusing on K–12 education and teacher perspectives, suggesting a need for broader and more inclusive investigation [25,39].

Finally, bibliometric analyses confirm a sharp increase in Gen AI-related educational research since 2022, driven primarily by the release of ChatGPT [27]. This growth reflects a shift toward interdisciplinary, policy-oriented approaches aimed at ensuring responsible and impactful integration of Gen AI in teaching, learning, and assessment [28,33,40].

4.2. ChatGPT in Educational Practice

This category includes studies that specifically investigate the use of ChatGPT in educational settings, focusing on its pedagogical implications, perceived benefits, and emerging challenges across different educational levels.

The growing body of literature reflects a strong interest in leveraging ChatGPT for educational purposes, particularly in higher education. Studies demonstrate its effectiveness in supporting assignment design, research assistance, and personalized learning pathways [32,38]. ChatGPT’s natural language processing capabilities enhance student engagement, accessibility, and instant feedback, contributing to more adaptive and inclusive learning environments [31,38]. However, significant concerns persist regarding the quality, bias, and factual reliability of its outputs, as well as the risks of academic dishonesty and over-reliance on AI-generated content. These challenges underscore the need for ethical strategies and educator oversight to safeguard academic integrity and ensure meaningful learning [31,34,38].

Research in K–12 contexts, though limited, points to ChatGPT’s potential to improve language accessibility, student motivation, and personalized instruction [29]. successful implementation requires coordinated efforts among educators, policymakers, and stakeholders to address risks such as data privacy, unequal access, and teacher preparedness. Notably, ChatGPT is not merely reshaping learning delivery but is also redefining the educator’s role, shifting from knowledge transmission to facilitation and personalized guidance [31]. Overall, the literature calls for a balanced integration of ChatGPT that aligns with pedagogical goals, supported by ethical frameworks and institutional policies to maximize its benefits and mitigate unintended consequences [34].

4.3. Commercial Gen AI Tools in Education

This category includes papers that review or analyze branded, real-world Gen AI applications (e.g., Bard (Google, 2023), Gemini (Google DeepMind, released 2023), RoomGPT (open-source project, 2023)) and their use in authentic educational settings, especially technical and applied fields.

The paper [42] provides a comprehensive review of state-of-the-art Gen AI models, including GANs, GPT, autoencoders, diffusion models, and transformers, with a focus on their architectures, tasks, and real-world applications. It highlights the rapid rise of commercial tools like ChatGPT, Bard, Stable Diffusion, and Adobe Firefly, highlighting their transformative impact across sectors including education, healthcare, business, and entertainment. In education, the study showcases how Gen AI tools are being used for personalized content creation, tutoring, code generation, and scientific visualization, while also acknowledging challenges such as misinformation, hallucinated outputs, and privacy concerns.

The authors of [43] explore the use of Gen AI in ICT engineering education, with an emphasis on programming and ethical instruction. The study highlights several commercial Gen AI tools actively used in educational settings, including ChatGPT (GPT-3.5, GPT-4), GitHub Copilot, Amazon Code Whisperer, Replit, Android Studio Bot, and OpenAI Codex, among others. These tools support various educational applications, including automatic feedback on code, bug detection, multilingual coding, and intelligent tutoring. The authors include real-world case studies from universities worldwide, showcasing how these tools enhance student learning, programming skills, and engagement, while also acknowledging ethical and practical limitations such as over-reliance, prompt engineering challenges, and bias in large language models.

• The Dynamic Landscape of Commercial Gen AI Tools:
  While early research (2022–2024) catalogued specific commercial tools such as ChatGPT, Bard, GitHub Copilot, and Adobe Firefly, such inventories risk rapid obsolescence. The commercial Gen AI ecosystem evolves rapidly, with foundation-model vendors frequently releasing new features, subsuming third-party tools, or launching academic-facing access programs. For example, in mid-2025, Perplexity AI partnered with SheerID to provide free access to Perplexity Pro for verified students worldwide [47], fundamentally shifting adoption dynamics and institutional decision-making. Similarly, OpenAI released ChatGPT Edu across the California State University system, offering access to more than 500,000 students and faculty [48].
  Beyond access initiatives, new agentic systems demonstrate capabilities that extend beyond traditional tutoring. For instance, Manus, an AI agent highlighted by researchers, is capable of generating full online course modules in minutes, exemplifying the rise of agentic task execution in education [49]. In parallel, Google’s NotebookLM emerged in 2025 as a study-assistant platform capable of summarizing course readings, integrating with academic workflows, and providing contextualized responses [50,51]. These innovations illustrate how commercial tools are increasingly embedding themselves into institutional platforms and student study habits.
  In light of this volatility, a more durable analytic lens is to abstract from specific brands to capability-based categories. Commercial tools in education can be framed into the following clusters: (i) retrieval-augmented tutoring and document-grounded assistance, (ii) agentic workflow generators, (iii) multimodal and creative generation, and (iv) LMS-integrated study platforms. Mapping risks and governance controls to these categories (rather than to specific brands) provides a framework that remains relevant even as individual tools evolve.
  Recent frameworks, such as the CRAFT model proposed by APRU [52], and adoption studies by Ithaka [53], reinforce this approach by linking capabilities with ethical, cultural, and policy dimensions. Large-scale surveys confirm these shifts: in the UK, the 2025 HEPI survey found that 92% of students now use Gen AI (up from 66% in 2024) and 88% report using it for assessments [54]; globally, more than 86% of students use AI in their studies, with 25% engaging daily [55]. Such patterns indicate that categories of capability and adoption, rather than static tool lists, offer more durable value to both researchers and practitioners.
• Emerging Categories: Research-Centric, Document-Grounded, and Agentic Systems Beyond early single-chat systems such as ChatGPT, the educational use of Gen AI since late 2024 has expanded into qualitatively distinct categories that require differentiated analysis and oversight. These categories include the following:

  –

  Retrieval-grounded research assistants: Tools such as Perplexity integrate live web search with citation trails, enabling students to generate responses anchored in real-time sources. The educational benefits include improved citation quality and source evaluation practices, while risks involve over-reliance on automatically retrieved content or uncritical trust in surfaced sources. Appropriate controls may include explicit citation rubrics, critical source evaluation exercises, and verification of reference accuracy [47].

  –

  Document-grounded tutors: Platforms like Google’s NotebookLM exemplify study tools that operate over instructor-provided materials (e.g., syllabi, PDFs, or course readings) [50]. Their alignment to curricular objectives allows for personalized scaffolding of reading comprehension and formative assessment. However, risks include bypassing independent reading and undermining student accountability. Case studies in higher education have shown that NotebookLM can support reading compliance and study accountability, but also require careful instructor oversight [51]. Controls therefore include embedding NotebookLM sessions into structured coursework, monitoring reading compliance, and designing assignments that require synthesis beyond the provided documents.

  –

  Agentic systems: Emerging systems such as Manus can plan and execute multi-step educational workflows, including automated generation of complete course modules. These affordances introduce both opportunities (e.g., rapid prototyping of instructional content, automation of routine tasks) and significant risks (e.g., outsourcing entire coursework, opaque decision-making processes). Mitigation measures include process logging, version control trails, oral defenses, and data diaries to verify authorship and learning integrity [49].

  These categories illustrate a transition in educational use of Gen AI: from brand-based discussions to capability-based analysis. Framing tools in terms of what they enable, and how associated risks can be managed, provides a foundation for more resilient adoption strategies. For example, the APRU whitepaper emphasizes the importance of linking AI use with institutional culture, rules, and access policies [52], while the Ithaka report highlights the need for frameworks that remain valid as tools evolve [53]. At the same time, practitioner reports show measurable benefits, such as teacher time savings and reduced workload stress, underscoring why such structured approaches are urgently needed [56].

Table 4. Analyzed academic works.

Title	Cit.	Methodology	Pub Venue	Database	Prisma	Use Case	Framework	GAI Tools	Review Objective
[23]	23	Topical Review (80 Papers)	IEEE Access, 2024	IeeeXplore	Yes	Education	No	No	Examines the impact of Gen AI on academic, highlighting the opportunities and challenges of AI, and exploring proposed frameworks.
[24]	29	Survey (24 Papers)	Information, 2024	MDPI	No	Higher Education	Yes	LLMs	Proposes a framework for ethical integration of AI in HE, emphasizing adaptive regulations, AI literacy, and responsible AI governance.
[25]	34	Systematic Mapping Review (407 Papers)	Review of Education, 2024	Wiley Online Library	No	Education	No	ChatGPT	The study highlights GAI educational impact, emphasizing gaps in K-12 (i.e., kindergarten through 12th grade in the U.S. system, corresponding to primary and secondary education) integration research.
[26]	67	Survey (26 Papers)	IEEE Access, 2024	IeeeXplore	No	Education	No	No	Gen AI enhances education through personalized learning, collaboration, and ethical practices.
[27]	66	Systematic Review (355 Papers)	Education Sciences, 2024	MDPI	Yes	Higher Education	No	ChatGPT-4, Gemini	Explores the current state, trends, and research gaps in Gen AI for teaching and learning in HE.
[28]	36	Systematic Literature Review (37 Papers)	Information, 2024	MDPI	Yes	Higher Education	No	ChatGPT, GPT-3.5, DALL-E	Analyzes empirical research on GAI in HE, categorizes its application areas, and proposes future research directions and policy implications.
[29]	155	Systematic Review (13 Papers)	European Journal of Education, 2024	Wiley Online Library	Yes	K-12	No	ChatGPT	Explores ChatGPT’s impact on K-12 education, highlighting its potential and the challenges, emphasizing the need for structured guidelines and collaborative efforts for effective integration.
[30]	35	Review (45 Papers)	Higher Education, 2024	Springer	No	Higher Education	No	ChatGPT	Examines the impact of Gen AI on HE, highlighting its potential benefits and challenges in academic integrity and assessment.
[31]	44	Systematic Review (70 Papers)	Frontiers in Education, 2024	Frontiers	Yes	Education	No	ChatGPT	Examines the benefits and challenges of ChatGPT, its impact on student engagement and learning outcomes, ethical considerations, and its effects on educators.
[32]	43	Systematic Literature Review (57 Papers)	International Journal of Educational Research, 2024	Elsevier	No	Higher Education	No	ChatGPT	Examines ChatGPT’s adoption in HE by analyzing trends, user intentions, applications for various stakeholders, research limitations, and future directions.
[33]	29	Review (23 Papers)	Higher Education, 2024	Springer	No	Higher Education	No	LLMs	Explores the use of GAI chatbots in HE reveals limited empirical research, minimal theoretical grounding, and polarized discourse.
[34]	205	Systematic Review (40 Papers)	International Journal, 2023	Ceeol	Yes	Education	No	ChatGPT	Explores ChatGPT’s transformative role in education, and its key applications with potential challenges.
[35]	572	Review (217 Papers)	Sustainability, 2023	MDPI	Yes	General education, medical, engineering, HE	No	ChatGPT GPT-4	Explores the transformative impact of GAI in education, highlighting its potential benefits across disciplines, and addressing ethical concerns like academic integrity, biases, and responsible use.
[36]	6	Systematic Literature Review (48 Papers)	Frontiers of Digital, 2024	Springer	No	Education	No	No	Examines studies on GAI in education and its role in learning support, instructional design, assessment, and feedback then addressing ethical concerns, AI literacy.
[37]	27	Research Article (36 Papers)	IEEE Access, 2024	IeeeXplore	No	Higher Education	Yes	ChatGPT	Explores the use of ChatGPT in education and proposes a decision-making framework to guide policymakers and educational institutions in ensuring its responsible and ethical implementation.
[38]	53	Systematic Review (112 Papers)	Education, 2024	MDPI	Yes	Education	No	ChatGPT	Explores benefits and challenges of ChatGPT in education, focusing on engagement, accessibility, ethical concerns, and academic integrity.
[39]	2	Scoping Review (24 Papers)	IEEE Access, 2025	IeeeXplore	Yes	Engineering and Computing Education	No	DALL·E, GitHub Copilot, RunwayML	Examines the implications of integrating GAI in engineering and computing education across K-12 to tertiary levels, revealing its benefits and challenges.
[40]	89	Scoping Review (32 Papers)	International Journal of Educational Technology in Higher Education, 2024	Springer	Yes	Higher Education	No	ChatGPT	Examines the potential of GAI in personalized feedback and self-regulated learning, challenges in academic integrity, and the need for teacher training, innovative pedagogy, and revised assessment policies.
[41]	1	Scoping Review (32 Papers)	Journal of Education, 2025	Wiley Online Library	Yes	Higher Education	Yes	No	Assesses the impact of GAI on academic skill development in higher education and proposes strategies for its ethical and effective integration.
[42]	215	Topical Review (20 Papers)	IEEe Access, 2024	IEEEXplore	No	Education, business, healthcare, media	No	ChatGPT, Bard, Adobe Firefly, RoomGPT, RunwayML, DALL·E, Jukebox	Provides a comprehensive technical and applied review of Generative AI models and tools, including their educational potential and societal impact.
[43]	9	Literature Review (26 Papers)	2024 47th MIPRO ICT and Electronics Convention (MIPRO), 2024	IEEEXplore	No	Engineering Education	No	ChatGPT, GitHub Copilot, Amazon CodeWhisperer, OpenAI Codex, Replit, Android Studio Bot, Tabnine, DeepCode	The review discusses how GAI supports programming and ethical competencies in ICT education, and identifies key tools and real-world teaching scenarios.

summarizes the main contributions of a selection of academic research in this field, with particular attention to the innovative aspects of each study to help address the research questions of this paper. Analyzed academic works form the dataset for RQs 1–5. The table highlights the innovative contributions of each study, which are systematically synthesized in Section 4.

This section presents the results organized according to the research questions (RQ1–RQ5). Each subsection highlights how the analyzed studies contribute to answering these questions.

4.4. RQ1: What Are the Current Definitions of Generative AI?

With the rapid advancement of artificial intelligence technologies particularly following the release of ChatGPT, Gen AI has become a key focus of research in the educational domain. However, definitions of Gen AI vary considerably across the literature, reflecting differences in technical perspectives, application contexts, and disciplinary orientations.

The reviewed literature offers a variety of definitions for Gen AI, reflecting both technical perspectives and educational applications. These definitions typically fall into two thematic categories: (1) Gen AI as autonomous content creation technology, and (2) Gen AI as a form of human-like creativity.

• Generative AI as Content Creation Technology
  Gen AI is commonly described as a class of AI systems capable of producing original content by learning from large datasets. These definitions highlight the underlying architectures such as LLMs (e.g., GPT), GANs, VAEs, and NLP models. For instance, Santos and his colleagues [43] and Bengesi [42] refer to Gen AI tools as content-producing models that enhance tasks like feedback, code generation, and personalized learning. Similarly, McGrath [33] and Daniel [41] emphasize the autonomous nature of these tools, while Xia [40] describes the ability of Gen AI to generate synthetic data across multiple modalities. In addition, Ahmed [23] and Mittal [26] define Generative AI as models capable of producing novel content or generating new material from existing sources, while Ogunleye [27] highlights its use in LLM-powered systems for content generation, tutoring, and feedback.
• Generative AI as Human-like Creativity
  The studies discuss Gen AI in terms of its ability to produce human-like content across various media formats. For example, Yusuf [25] describes Generative AI as AI models that can create human-like content across multiple modalities. These discussions often emphasize the use of advanced AI techniques to generate content that mimics human creativity and cognition. Several studies focus on the application of Gen AI in educational contexts, particularly in K–12 education, highlighting its potential to generate content that enhances learning experiences [29]. Other authors explore how Gen AI models analyze existing digital media to simulate human-level creativity, enabling the generation of text, images, and other forms of content [35]. Additionally, there is a significant focus on Gen AI capabilities in language interpretation, summarization, and prediction, which has led to considerable interest and discussion regarding its implications for educational practices [36].
  Our synthesis reveals inconsistency in definitions across reviews, ranging from purely technical descriptions to broader pedagogical framings (see

  Table 5. Definitions of Generative AI across selected studies.

  Study	Definition of Gen AI	Category
  [23]	Machine learning models that can produce novel, original content without being explicitly programmed.	Content Creation
  [25]	AI models that create human-like content across multiple media formats.	Human-Like Creativity
  [26]	Specialized AI that generates new content from previous materials.	Content Creation
  [27]	LLM-powered systems performing complex tasks like content generation, tutoring, and feedback.	Content Creation
  [33]	AI based on or self-supervised models in generating original outputs (text, images).	Content Creation
  [35]	Framework using existing media to simulate human creativity.	Human-Like Creativity
  [29]	Subset of AI tools generating content via advanced techniques rather than human creation.	Human-Like Creativity
  [40]	Subset of AI generating new content including synthetic data.	Content Creation
  [42]	Models generating data resembling training data (GANs, GPT, etc.).	Content Creation
  [41]	AI systems generating text, images, and music from learned patterns.	Content Creation
  [36]	AI with human-like capabilities in language generation, summarization, and prediction.	Human-Like Creativity
  [43]	Models using LLMs and NLP for content generation in engineering education.	Content Creation

  ). This conceptual fragmentation complicates comparative analysis and underscores the need for a more standardized definition of Gen AI in education.

4.5. RQ2: What Are the Current Methodologies and Frameworks for Integrating Gen AI into Educational Systems?

Although the rapid rise of Gen AI has prompted widespread experimentation in educational settings, the academic literature still offers a limited number of structured methodologies or frameworks for its integration. Among the studies reviewed, only a few propose concrete models or evaluative tools to guide the implementation of Gen AI in educational environments.

Ahmed and his colleagues [23] provide one of the most comprehensive contributions by reviewing four notable frameworks: PAIGE, AI-CRITIQUE, DATS, and IDEE. These frameworks collectively address personalization, ethical considerations, assessment, feedback loops, and instructional design. They aim to guide the responsible deployment of Gen AI tools such as EduChat [57] and CodeHelp [58] within educational curricula.

Ref. [24] proposes a conceptual framework grounded in AI literacy, human oversight, adaptive regulation, and responsible governance. Their model stresses ethical integration of large language models in higher education, calling attention to transparency, inclusivity, and safeguards against plagiarism and bias.

Ref. [41] introduces the UTAUT-EG model, an adaptation of the Unified Theory of Acceptance and Use of Technology (UTAUT) with an ethical governance focus. This framework connects the use of Gen AI tools to the development of cognitive, technical, and interpersonal academic skills, while promoting principles of transparency, accountability, and responsible adoption.

Ref. [37] applies the Analytical Hierarchy Process (AHP) to prioritize ethical concerns surrounding ChatGPT use in academia. The authors proposed a decision-making framework assisting educational institutions in choosing governance strategies, revealing a preference for restriction over legislation as a more adaptable policy approach.

Existing approaches primarily focus on integrating Gen AI into specific educational contexts, addressing particular challenges, and optimizing learning in certain fields or courses. The absence of a standardized framework that leverages the full potential of Gen AI across various educational settings represents a significant challenge.

Existing approaches primarily focus on integrating Gen AI into specific educational contexts, addressing particular challenges, and optimizing learning in certain fields or courses (see

Table 6. Frameworks for integrating Gen AI into education.

Study	Methodology	Focus Area
[23]	PAIGE, AI-CRITIQUE, DATS, IDEE	- Personalization; - Ethics; - Instructional design; - Adaptive feedback
[24]	Ethical Integration Framework	- AI literacy; - Human oversight; - Responsible governance; - Academic integrity
[41]	UTAUT-EG (Unified Theory of Acceptance and Use of Technology—Ethical Governance)	- Academic skill development; - Ethical AI adoption; - Transparency
[37]	Analytical Hierarchy Process (AHP) Decision Framework	- Ranking ethical risks; - Governance strategy selection; - Policy design

). The absence of a standardized framework that leverages the full potential of Gen AI across various educational settings represents a significant challenge.

Notably, only 7 of the 21 reviews employed a formal synthesis method, and fewer than half reported explicit eligibility criteria. This uneven methodological rigor highlights a lack of standardization, limiting comparability across studies.

4.6. RQ3: Which Educational Levels Have Been the Primary Focus of Research on Gen AI?

Referring to RQ3, our primary focus was on the academic context, and the main educational level targeted in Gen AI research is higher education, while K–12 education remains significantly underrepresented. Most of the selected papers focus on how Gen AI is being integrated, discussed, or evaluated within universities and other tertiary institutions.

Ogunleye and his colleagues [27] and Jensen [30], highlight the explosive growth of Gen AI-related publications since 2022, particularly within the context of higher education. These works emphasize themes such as academic integrity, student engagement, automated assessment, and intelligent tutoring systems, all situated in university environments.

Other authors [32,38] explore Gen AI applications, especially ChatGPT, for research support, assignment generation, and student-centered learning, reinforcing the dominance of higher education as the main setting. McGrath [33] confirms that the majority of the academic discourse surrounding Gen AI appears in the higher education space, including fields beyond traditional computer science, indicating a strong institutional interest in adopting AI for teaching and learning.

On the other hand, K–12 education receives much less attention. Zhang [29] highlights the potential of tools like ChatGPT for personalized instruction, language development, and learner motivation, but also points out that policy constraints, teacher training gaps, and ethical concerns have delayed the broader adoption of Gen AI in schools.

Bengesi [42] and Santos [43] focus on the use of commercial Gen AI tools such as ChatGPT, Bard, Adobe Firefly, GitHub Copilot, Replit, and Amazon CodeWhisperer. While their studies span multiple sectors, they primarily highlight educational use cases in higher education, especially in ICT, computer science, and engineering. For instance, Santos and his colleagues [43] present real-world university-level examples where these tools are used for code generation, debugging, intelligent tutoring, and ethical instruction. Similarly, Bengesi [42] emphasizes how commercial tools support content creation, tutoring, and scientific visualization in tertiary settings.

While higher education dominates the current literature, K–12 remains systematically underrepresented, with only one review addressing this level in depth. This gap points to an urgent need for comparative research across educational stages.

4.7. RQ4: What Are the Main Challenges of Integrating Gen AI into Education?

Referring to RQ4, the integration of Gen AI into the education sector introduces a complex array of challenges. A consistent concern is the reliability and accuracy of Gen AI outputs, with several studies [31,38,42] reporting that Gen AI tools such as ChatGPT and Bard are prone to hallucinated, biased, or misleading responses. These issues raise significant concerns about trust, particularly in academic settings where factual correctness and critical thinking are essential.

Educators also face uncertainty in adapting pedagogy to AI-driven learning environments. As noted by Jensen [30] and McGrath [33], Gen AI alters traditional educational processes such as student engagement, content creation, and assessment practices. This is further compounded by a lack of institutional readiness and structured support, which may lead to over-reliance on AI tools and a decline in deeper learning and creativity, as emphasized in [26,32].

Ethical risks represent another major challenge, especially in relation to plagiarism, academic integrity, data privacy, and AI bias. Qadhi [24] and Bukar [37] stress the importance of ethical governance and human oversight, proposing frameworks to ensure responsible implementation and address regulatory gaps.

Moreover, several studies [35,41] raise broader concerns regarding equitable access and AI literacy. These studies emphasize that without inclusive training, accessible design, and strong policy frameworks, Gen AI risks exacerbating existing educational inequalities.

In addition to general concerns of bias and misinformation, the integration of Gen AI into education creates potential conflicts between innovation and academic integrity. While these tools can stimulate creativity, provide personalized learning pathways, and reduce administrative burdens, their unregulated use may undermine the authenticity of student work, blur boundaries of plagiarism, and complicate assessment practices. Moreover, disparities in AI literacy and access to digital infrastructure, particularly in low-resource educational systems, risk widening existing inequalities, as educators and learners without adequate training may misuse or underutilize Gen AI. These gaps highlight the urgent need for AI literacy programs, clear institutional policies, and governance frameworks that balance pedagogical innovation with safeguards for integrity, equity, and accountability. Addressing these dimensions is essential for ensuring that Gen AI adoption enhances rather than undermines educational quality and inclusiveness (see

Table 7. Challenges of integrating Gen AI into education.

Challenge Category	Studies	Key Issues
Reliability and Accuracy	[31,38,42]	Gen AI outputs are prone to hallucinated, biased, or misleading responses. This undermines trust in academic settings where accuracy and critical thinking are essential.
Pedagogical Adaptation	[30,33]	Educators face uncertainty in adapting pedagogy to AI-driven environments. AI alters engagement, content creation, and assessment, often without institutional support.
Institutional Readiness	[26,32]	Lack of structured support and teacher preparation leads to over-reliance on Gen AI tools, potentially weakening student creativity and deep learning.
Ethical Risks	[24,37]	Challenges include plagiarism, academic integrity, data privacy, and algorithmic bias. The need for ethical governance and human oversight is emphasized.
Equity and AI Literacy	[35,41]	Without inclusive training, equitable access, and policy frameworks, Gen AI may worsen educational inequalities and exclude underrepresented learners.

).

Although ethical risks such as plagiarism and bias are widely acknowledged, very few reviews connect these issues to governance or institutional frameworks. This indicates a gap between identifying challenges and proposing actionable solutions.

4.8. RQ5: How Effective Are Gen AI Tools in Fostering Critical Thinking and Problem-Solving Skills Among Students?

Referring to RQ5, Gen AI has demonstrated transformative potential in various educational contexts. Specifically, Gen AI tools are increasingly recognized for their capacity to enhance students’ critical thinking and problem-solving abilities. Several papers discuss the potential of Gen AI in supporting higher-order thinking through personalized feedback, adaptive learning pathways, and intelligent tutoring systems.

Daniel [41] reports significant improvements in cognitive skills, including critical thinking, metacognition, and problem-solving, especially when Gen AI tools are implemented with ethical guidance and teacher participation. Similarly, studies by Ali and Xia [38,40] highlight the ability of Gen AI tools to support self-regulated learning and reflective thinking by providing instant feedback and offering alternative perspectives during learning tasks.

However, multiple studies also raise concerns about over-reliance on Gen AI, which may weaken students’ independent reasoning. Baig and Mittal [26,32] argue that students may begin to delegate cognitive tasks to AI systems, limiting their engagement in deep problem-solving processes unless educators purposefully design learning activities that promote active learning and critical analysis.

Jensen [30] adds that many current implementations of Gen AI tend to reinforce existing educational models rather than fundamentally transform how students approach inquiry, creativity, and autonomous learning. Furthermore, despite the optimism surrounding the potential of Gen AI, there is currently limited empirical evidence in most of the reviewed literature to conclusively assess its impact on critical thinking, particularly in comparison to traditional pedagogies.

While our analysis focuses on outcomes reported in academic studies (see

Table 8. Reported impacts of Gen AI on critical thinking and problem-solving.

Study	Reported Effect	Focus Area	Remarks
[41]	Positive impact on cognitive, technical, and interpersonal skills	Critical thinking, metacognition, problem-solving	Gen AI enhances learning when paired with ethical guidance and teacher support
[38]	Supports self-regulated learning and reflective thinking	Instant feedback, adaptive pathways	Enhances autonomous learning processes
[40]	Encourages reflective learning and self-regulation	Personalized feedback, learning support	Gen AI offers multiple learning perspectives
[32]	Concern about over-reliance on AI	Risk to student independence	Students may delegate thinking tasks to AI without proper instructional design
[26]	Cautions about loss of deep learning	Overuse of Gen AI tools	Emphasizes the need for guided implementation to prevent cognitive disengagement
[30]	Reinforces existing models, lacks transformative impact	Pedagogical inertia	Gen AI supports rather than redefines learning practices

), it is important to note that a wide range of commercial Gen AI tools have recently been introduced into educational environments. However, their effectiveness in fostering critical thinking and problem-solving remains under-examined. More empirical research is needed to evaluate the long-term impact of these tools on learning outcomes across diverse educational contexts.

Evidence of Gen AI’s effectiveness in fostering critical thinking and problem-solving remains limited and often anecdotal, underscoring the absence of large-scale empirical validation.

It is important to acknowledge the limitations of the current literature. First, most studies are conceptual or review-based, with a lack of empirical evidence evaluating Gen AI’s effectiveness in authentic classroom environments. Second, there is a clear imbalance in focus: the majority of research centers on higher education, while K–12 education remains underrepresented despite its critical role in shaping early AI literacy. Third, nearly all included studies are published in English, leading to the near absence of perspectives from non-English-speaking regions, which limits the global generalizability of findings. Addressing these limitations requires a more inclusive research agenda, incorporating diverse educational levels, cultural contexts, and methodological rigor.

5. Future Trends

Generative AI is not only transforming the educational landscape but also catalyzing the development of new learning models that extend beyond traditional classroom structures. To better understand these emerging directions, we asked the following question: What recent trends suggest how Gen AI will drive the development of new learning models in education?

Recent research suggests that the field is moving beyond single-chat interfaces and toward capability-based systems. A growing trend is the rise of agentic AI systems, capable of multi-step planning and task execution. Early platforms such as Manus illustrate how these systems can generate entire projects or learning modules with minimal input. While these capabilities hold promise for productivity and innovation, they also raise concerns about coursework automation, requiring new assessment strategies such as process logging, version trails, or oral defenses [49].

Another major development is the use of document-grounded and retrieval-augmented AI tutors. Tools such as NotebookLM allow students to query and generate study guides directly from course readings, while Perplexity integrates live web search with citation trails [50]. These tools shift pedagogy toward reading accountability and source evaluation, but they also require redesigned assignments and policies to ensure independent learning and academic integrity [51].

Institutional adoption is also accelerating. In 2025, several universities and ed-tech providers launched large-scale student access programs, such as Perplexity’s free academic licenses and OpenAI’s partnership with California State University [47,48]. This signals a transition from isolated experimentation to structured, institution-wide integration of Gen AI in curricula and assessment.

Future models are also expected to emphasizehuman–AI co-creation, where learners collaborate with AI agents to generate content, solve problems, and engage in entrepreneurial projects [53]. This development requires redefining the role of educators as AI facilitators and literacy coaches, emphasizing skills such as prompt design, critical evaluation, and ethical navigation [52].

Finally, governance frameworks are emerging as a critical trend. Moving beyond general concerns of bias or privacy, recent reports call for capability–risk–control approaches, for example, mapping the risks of retrieval grounding, document grounding, and agentic automation to appropriate institutional safeguards. This requires not only technical solutions (e.g., detection systems for academic dishonesty [59]) but also institutional readiness, inclusive policymaking, and frameworks that balance innovation with educational integrity.

Collectively, these trends suggest that by 2025–2026, Gen AI will be understood not as a single application but as an evolving ecosystem of capabilities. This transition carries profound implications for pedagogy and governance, requiring the ongoing adaptation of educational models, assessment practices, and institutional frameworks.

6. Conclusions

Generative AI has significant potential to transform education by enabling personalized learning experiences and producing interactive content tailored to individual learners. This meta-survey provides a comprehensive synthesis of current research on the integration of Gen AI in education, drawing on systematic literature reviews, scoping studies, and thematic analyses across diverse contexts. Using the PRISMA methodology, 21 surveys were examined and their contributions analyzed. The findings indicate that, although research on Gen AI has proliferated rapidly particularly since the emergence of ChatGPT, important gaps remain.

The analysis also highlights the emerging role of commercial Gen AI tools, which are increasingly being incorporated into teaching and learning practices across multiple disciplines. Despite their expanding presence, relatively few studies have explored these tools in depth, with more focused discussions only beginning to appear in 2024. Much of the existing literature continues to emphasize broader AI themes or isolated applications. Nevertheless, commercial tools are expected to become more widely adopted and more systematically studied in the near future, reflecting their growing educational impact.

Overall, the evidence suggests that Gen AI represents more than an enhancement to traditional instruction; it is a transformative force with the capacity to reshape educational models. By identifying current trends, gaps, and opportunities, this meta-survey establishes a foundation for further research and offers insights that may inform the integration of Gen AI into future educational practices.