Introduction
The Individualized Education Program (IEP) is a foundational element of inclusive education systems, designed to ensure that students with special educational needs (SEN), including those with disabilities, have access to a free, appropriate and individualized education. Functioning both as a legal mandate and a pedagogical tool, the IEP outlines students’ current academic and functional performance, sets individualized annual goals, identifies necessary accommodations and modifications and specifies procedures for progress monitoring (Özdemir et al., 2020).
Despite comprehensive legal frameworks and detailed procedural guidelines, IEPs often exhibit inconsistencies in their design and implementation. Research indicates that many IEPs lack pedagogically grounded goals and omit legally required components. For instance, a large-scale study of 206 IEPs from 60 schools in Turkey revealed widespread misalignment between the documented needs of students and the actual content of their IEPs, resulting in generic and superficial plans that fail to address individual learning profiles (Rakap, Yucesoy-Ozkan, Kalkan, 2019). Further challenges arise in the formulation of IEP goals and objectives, which frequently lack specificity, measurability and relevance. In particular, many IEPs fail to meet specific standards and, in some cases, frame objectives in terms of teacher actions rather than student outcomes (Ruble et al., 2010).
In response to these persistent issues, emerging technologies, particularly those powered by artificial intelligence (AI), offer promising avenues for innovation in IEP design. Among these, ChatGPT, a large language model (LLM) developed by OpenAI, has demonstrated potential in mainstream educational contexts by supporting personalized learning, lesson planning, content generation and assessment design (Herft, 2023; Rahman, Watanobe, 2023). Given the IEP’s need for customized goal-setting and instructional alignment, such AI tools could function as intelligent assistants, helping educators draft SMART goals (Specific, Measurable, Achievable, Relevant, and Time-bound), design interventions and construct rubrics tailored to individual students (Adiguzel, Kaya, Cansu, 2023). Likewise, a mixed-methods study with Greek special educators demonstrated that ChatGPT can facilitate IEP development by clarifying goals, generating instructional materials and structuring lesson plans, while teachers emphasized the necessity of maintaining professional judgment and the relational dimensions of teaching (Giaouri, Charisi, 2025).
However, while the theoretical potential of ChatGPT is considerable, its integration into special education, and specifically into IEP development, remains an underexplored area. There is a notable lack of empirical studies assessing its impact on the quality, consistency, and efficiency of IEPs. Additionally, significant ethical issues, such as data protection, teacher autonomy and student autonomy, remain minimally studied. This paper focuses specifically on ChatGPT, an AI-powered language model developed by OpenAI. While broader developments in AI are occasionally referenced for context, no empirical data from other AI platforms or tools are examined. The examples and applications provided are illustrative rather than comprehensive.
The study addresses a critical gap by exploring the central research question: What are the educational implications, opportunities and challenges of using ChatGPT to support the development and implementation of IEPs for students with Special Educational Needs (SEN), including those with disabilities? By synthesizing research from the fields of special education, AI in education and inclusive pedagogy, this article evaluates the extent to which tools like ChatGPT can contribute to pedagogically sound and ethically responsible improvements in IEP practices.
Advantages, challenges, and ethical implications of ChatGPT in general and special education
ChatGPT offers significant benefits in education, such as enhancing student independence and providing immediate feedback (Mhlanga, 2023; Rahman, Watanobe, 2023). Moreover, it supports students with special educational needs and/or disabilities by delivering personalized, step-by-step guidance and integrating assistive technologies (Alshahrani, 2023; Liang et al., 2023). Additionally, AI-driven personalization of instruction improves engagement and comprehension by tailoring content to the individual needs of each learner (Adiguzel, Kaya, Cansu, 2023; Chen et al., 2022). ChatGPT also facilitates the design of individualized lesson plans and educational materials, thereby increasing both quality and accessibility (Gupta, Raturi, Venkateswarlu, 2023; Tlili et al., 2023). In assessment, it streamlines grading processes and provides immediate feedback (Cotton, Cotton, Shipway, 2023; Mondal et al., 2023). Finally, teacher professional development is enhanced through AI training, promoting the appropriate and ethical use of these tools (Li, Ye, 2023).
Despite its advantages, the use of ChatGPT in education raises ethical and practical concerns. A primary issue is bias within training data, which may reinforce discrimination or stereotypes (Adiguzel, Kaya, Cansu, 2023). Furthermore, AI-based assessment can be inaccurate and potentially disadvantage certain groups (Baidoo-Anu, Owusu Ansah, 2023), while students often accept responses uncritically, increasing the risk of misinformation (Božić, Poola, 2023). Cultural and linguistic biases may marginalize data that do not originate from Western contexts (Kohnke, Moorhouse, Zou, 2023; Mhlanga, 2023). Privacy concerns and reduced human interaction have also been identified as significant issues (AlZu'bi et al., 2024), while “hallucinations”, the generation of false information, pose further risks (Rudolph, Tan, Tan, 2023). Academic integrity is also threatened, as ChatGPT is frequently used for assignments without genuine effort, leading to plagiarism (Jarrah, Wardat, Fidalgo, 2023).
Materials and methods
This study employed a descriptive and integrative literature review to examine emerging applications of ChatGPT in supporting inclusive practices in primary and secondary special education. Given the novelty of the topic, the review prioritized capturing recent empirical findings and practitioner perspectives rather than conducting a fully systematic review. To enhance methodological transparency and rigor, selected elements of the PRISMA framework were applied to guide study identification, screening and reporting. Specifically, the PRISMA 2020 stages of Identification, Screening, Eligibility and Inclusion were followed to document the flow of studies throughout the review process. These steps ensured transparency in database search, exclusion criteria and the final synthesis of the selected publications.
A structured search was conducted in Google Scholar, ResearchGate and ERIC for publications between January 2023 and June 2024, using keywords such as “ChatGPT,” “Artificial Intelligence,” “inclusive education,” “special education” and “Autism Spectrum Disorder.” Eligible studies were peer-reviewed, published in English with full-text access, reported classroom applications or interventions and focused on primary or secondary learners, including those with diverse learning needs. Studies were excluded if they addressed other AI tools, were purely conceptual or theoretical or lacked full-text availability.
The search yielded 73 records; 38 were excluded at the title/abstract stage and 24 full-text articles were removed due to non-compliance with inclusion criteria, leaving 11 studies for synthesis (see Table). Data were thematically coded into three domains: pedagogical affordances, implementation challenges and ethical/professional considerations. This approach provided a structured, evidence-informed understanding of ChatGPT’s integration in inclusive educational settings.
Table. PRISMA Flow Diagram for Study Selection
Identification
Records identified through database searching (Google Scholar, ResearchGate, ERIC): n = 73
Records after duplicates removed: n = 73
Screening
Records screened (titles/abstracts): n = 73
Records excluded (irrelevant or duplicate): n = 38
Eligibility
Full-text articles assessed for eligibility: n = 36
Full-text articles excluded: n = 24
Reasons: non–peer-reviewed, focused on other AI tools, or purely theoretical
Included
Studies included in final synthesis: n = 11
Results
The integration of ChatGPT into teaching and lesson design has demonstrated positive outcomes across diverse educational contexts. In the study by Karaman, Göksu (2024), third-grade mathematics lesson plans were developed with the support of ChatGPT (experimental group) and without it (control group). Effectiveness was evaluated using a 25-item multiple-choice pre- and post-test. Although the experimental group exhibited significant improvement in academic achievement, post-intervention comparisons between the two groups did not reach statistical significance. These findings suggest that ChatGPT can be at least as effective as traditional approaches, provided that teachers ensure flexibility and continuity of activities.
Similarly, Yilmaz Can, Durmuş (2024) conducted a content analysis of fourth-grade fraction lesson plans, comparing those generated with ChatGPT against traditional plans. Data sources included the lesson plans themselves, teacher observations, researcher notes and feedback, with triangulation applied to strengthen validity. Effectiveness was assessed through the 5E instructional model (Engage, Explore, Explain, Elaborate, Evaluate), focusing on both lesson structure and classroom implementation. Results indicated that ChatGPT-enhanced plans fostered teacher creativity and professional development, though they required adaptation to student levels, limiting flexibility to some extent.
Baytak (2024) analyzed 18 lesson plans produced with ChatGPT and Google Gemini across mathematics, science, literature and social studies for 12-year-old students. Content analysis combined cognitive mapping with expert evaluations of educational applicability, supported by qualitative coding in the open-source software Taguette. Findings showed strong similarities between AI-generated and human-designed plans, with clear learning objectives and alignment of activities. However, the tasks were predominantly quizzes, pointing to the need for more interactive content.
In another comparative study, Li, Liu, Yang (2024) examined five mathematics (Grade 8) and English (Grade 9) lesson plans produced by teachers and by ChatGPT (versions 3.5 and 4.0). Ten experts evaluated the plans against five domains of the Chinese Young Teachers’ Competition rubric: instructional objectives, teaching content, teaching process, teaching methods and innovation. Results showed that ChatGPT-generated lessons produced content and language interactions broadly comparable to those of human teachers, although the language tended to be standardized and colloquial. This evaluation method enabled a systematic comparison of the effectiveness of AI-generated and teacher-generated lesson designs.
The development of physics exercises for students aged 15–16 was investigated by Küchemann et al. (2023), comparing tasks created with ChatGPT 3.5 and those based on a school textbook. Effectiveness was assessed by evaluating the quality and correctness of the exercises, alongside user experience measures using the System Usability Scale (SUS) and the Technology Acceptance Model 2 (TAM2). Results indicated no significant differences in task accuracy between the two groups, though both faced limitations regarding sufficiency of information. Participants rated ChatGPT as useful but noted difficulties in tailoring exercises, highlighting both its potential and its constraints in educational practice.
In South Korea, Lee, Zhai (2024) examined 29 pre-service elementary teachers who integrated ChatGPT into 14 instructional strategies for science education. Using a modified TPACK-based evaluation instrument, qualitative analysis revealed positive expectations including the promotion of higher-order questioning, support for self-directed learning and personalized assistance while concerns centered on information accuracy and possible student overreliance. The outcomes showed alignment with instructional goals but emphasized the need for teacher training to maximize effective integration.
Zemljak (2023) analyzed 58 ChatGPT-generated STEM lesson plans for fourth-grade students. Three randomly selected teachers independently evaluated the plans in terms of curriculum alignment, appropriateness of teaching methods, completeness of materials, feasibility of implementation and overall structure. Results revealed deficiencies in learning objectives and deviations from curricular timelines, underscoring the need for active teacher involvement in refining AI-generated plans and ensuring alignment with curriculum standards.
In the field of special education, Rakap (2023) evaluated 22 novice special educators in designing IEP goals for five preschool children with Autism Spectrum Disorder (ASD). Effectiveness was assessed using the R-GORI instrument, with analysis of goal quality, prior training and preparation time. The ChatGPT group produced significantly higher-quality goals, required less preparation time and generated more comprehensive and individualized targets. In a subsequent study, Rakap, Balikci (2024) used a mixed-methods design. Quantitative analysis with a two-sample t-test showed higher-quality goals for the ChatGPT-assisted group, while qualitative analysis revealed a greater proportion of communication, social, motor/sensory and self-care goals compared with the control group, which focused more on pre-academic and behavioral goals.
The impact of ChatGPT-supported worksheets was examined by Rizos, Foykas, Georgakopoulos (2024) in a nine-hour mathematics intervention with two 14-year-old students (one with dyslexia and one with ASD). Data collection included interviews, observation and online questionnaires. Effectiveness was measured in terms of participation, self-confidence, achievement of learning goals and student attitudes. Findings demonstrated increased engagement, confidence and goal attainment, along with more positive attitudes.
Most recently, Waterfield et al. (2025) employed a convergent mixed methods design to analyze the quality of IEP goals created either independently by teachers or collaboratively with ChatGPT. Quantitative analysis was conducted using a modified R-GORI tool, with evaluations performed by participants and verified by two independent doctoral-level reviewers, while statistical analysis was performed in Stata/BE 18.5. The qualitative component included 17 semi-structured interviews. Results indicated positive teacher attitudes toward AI for reducing workload and enhancing IEP goal quality but also raised concerns about preserving the human dimension of special education and ensuring sufficient training for effective AI integration.
Discussion
The integration of ChatGPT into educational contexts has attracted growing attention in recent literature, particularly within general primary (Baytak, 2024; Karaman, Göksu, 2024; Lee, Zhai, 2024; Yilmaz Can, Durmuş, 2024; Zemljak, 2023) and secondary education settings (Küchemann et al., 2023; Li, Liu, Yang, 2024), with comparatively limited exploration in the field of special education (Rakap, 2023; Rakap, Balikci, 2024; Waterfield et al., 2025). Geographically, much of this research has been conducted in Turkey, with additional contributions from East Asia, Europe and the United States.
Methodologically, the reviewed studies employed a variety of designs. Some adopted experimental or quasi-experimental frameworks involving control and treatment groups (Karaman, Göksu, 2024; Küchemann et al., 2023; Rakap, 2023; Rakap, Balikci, 2024; Yilmaz Can, Durmuş, 2024), while others implemented ChatGPT with all participants as a tool for classroom activities or lesson planning (Lee, Zhai, 2024; Li, Liu, Yang, 2024; Waterfield et al., 2025; Zemljak, 2023). Participants ranged from in-service teachers (Karaman, Göksu, 2024; Li, Liu, Yang, 2024; Rakap, 2023; Rakap, Balikci, 2024) to students across various grade levels (Küchemann et al., 2023; Lee, Zhai, 2024; Yilmaz Can, Durmuş, 2024; Zemljak, 2023).
Many studies centered on STEM subjects such as mathematics, physics and biology (Karaman, Göksu, 2024; Lee, Zhai, 2024; Yilmaz Can, Durmuş, 2024; Zemljak, 2023), while some incorporated interdisciplinary approaches blending science and humanities (Baytak, 2024; Li, Liu, Yang, 2024). Results generally suggested that ChatGPT-generated lesson plans were comparable in quality to those created by teachers, showing alignment with curriculum goals, fostering collaborative learning and improving instructional efficiency. Reported benefits included enhanced student performance, novel instructional ideas and reduced lesson planning time (Karaman, Göksu, 2024; Yilmaz Can, Durmuş, 2024). In the context of special education, ChatGPT demonstrated promise in generating more coherent and goal-aligned IEPs while decreasing teacher workload (Rakap, 2023; Rakap, Balikci, 2024).
However, the overwhelmingly positive portrayal of ChatGPT’s educational applications may obscure several critical issues and methodological limitations. For instance, some studies reported barriers related to limited technological infrastructure and digital literacy among teachers (Baytak, 2024), rigid language outputs that lacked local contextual nuance (Li, Liu, Yang, 2024) and difficulties in tailoring ChatGPT’s responses to specific classroom dynamics (Lee, Zhai, 2024). Additionally, instances of inflexible or incomplete lesson plans, unrealistic timelines and omissions in materials preparation were noted (Yilmaz Can, Durmuş, 2024; Zemljak, 2023). Concerns about over-reliance on automated tools at the expense of teacher creativity and the importance of preserving the "human touch" in education were also raised (Waterfield et al., 2025).
Conclusions
This study explored the emerging use of ChatGPT in special education, focusing on lesson planning, instructional design and IEP goal development (Rakap, 2023; Rakap, Balikci, 2024; Waterfield et al., 2025). Findings indicate that ChatGPT can enhance creativity, streamline planning and assist novice educators in formulating goals for diverse learners (Karaman, Göksu, 2024; Rizos, Foykas, Georgakopoulos, 2024; Yilmaz Can & Durmuş, 2024). AI-generated lesson plans were generally comparable to those of teachers but often required adaptation to curriculum standards and individual needs (Baytak, 2024; Küchemann et al., 2023; Li, Liu, Yang, 2024; Zemljak, 2023). In special education, ChatGPT improved clarity and coherence of IEP goals, particularly in communication, self-care and social skills, while reducing teacher workload. Educators emphasized the importance of human oversight and using AI as a supplement rather than a replacement (Waterfield et al., 2025). Limitations included prompt precision, standardized outputs and lack of contextual nuance (Lee, Zhai, 2024; Li, Liu, Yang, 2024), highlighting the need for targeted teacher training.
Building on these findings, AI integration should be guided by inclusive, culturally responsive and ethically sound frameworks. Drawing on “Artificial Intelligence in Early Education” (Papadakis, 2025) future directions include participatory design with children and educators, professional development programs, longitudinal and mixed-method research, and embedding AI literacy within national curricula. These approaches aim to ensure that AI tools foster meaningful learning, equity and developmentally appropriate practices from the earliest educational stages. Ultimately, ChatGPT should complement human teaching, empowering educators to create inclusive, collaborative and ethical learning environments. In line with UNESCO guidance (Giannini, 2023), education in the era of generative AI must critically assess when, by whom and for what purposes AI tools are used, ensuring human-centered, safe and socially responsible implementation.
Limitations. This review presents several methodological limitations. Its descriptive and integrative nature, guided by selected PRISMA elements rather than a fully systematic process, restricts the comprehensiveness and reproducibility of findings. Although the search strategy included major databases and clear inclusion criteria, the relatively small number of eligible studies (n = 11) limits the representativeness of the evidence base. The review relies exclusively on published, English-language sources, which may exclude relevant non-English or grey literature. Furthermore, variations in research design, educational contexts and participant characteristics across studies reduce comparability and hinder the generalization of conclusions. As most studies examined short-term or pilot implementations, long-term effects of ChatGPT in special education remain unclear. Future research should adopt systematic, longitudinal and mixed-method approaches to strengthen empirical validity and deepen understanding of ChatGPT’s pedagogical, ethical and practical implications in inclusive education.