Exploring the artificial intelligence interaction profiles of participants with different levels of teaching experience for lesson planning in the context of acids and bases

Abstract

In this multi-case study, the ChatGPT interaction profiles of participants with varying years of teaching experience were examined during lesson planning for the topic of acids and bases. Six participants (i.e., two senior pre-service science teachers, one induction year science teacher, two experienced science teachers, and an experienced science teacher educator) participated in the study. A lesson plan in the Content Representation (CoRe) format, a copy of dialogue with ChatGPT (i.e., conversation history with ChatGPT), a reflection paper on details of the lesson planning with ChatGPT, and online focus group interview data were collected from all participants. Both deductive and inductive analyses of the multiple data sources revealed four profiles, namely, Artificial Intelligence (AI)-reliant planners, AI-collaborative planners, AI-assisted plan refiners, and AI-independent planners. The profiles differed from each other in terms of the flow of lesson planning with AI, the interaction purposes of the participants with ChatGPT during lesson planning, and the extent to which the participants incorporated the content provided by ChatGPT into their lesson plan preparation. In light of the results, science pre-service teachers and teachers should be trained on what AI can offer them, how AI tools can be effectively utilized in science education, and the ethical considerations of AI use.

---

Introduction

With the advent of artificial intelligence (AI), education and training processes have evolved into a new dimension (Guner and Er, 2025). While AI tools play certain roles for students in the learning process, such as preparing homework and providing personalized learning (Li et al., 2023), they also offer convenience for teachers in terms of enriching the content of lesson plans (Clark et al., 2024), grading assignments, and providing personal feedback (Cao and Zhong, 2023). In the literature, besides the benefits of using AI tools, it has been reported that there are problems such as the loss of collaboration and innovation competencies of learners (AlBadarin et al., 2023), hallucinations, and biased and false information (Feldman-Maggor et al., 2025). Given the rapid integration of AI into education, there is a need for further research on how this technology can be effectively utilized by teachers and learners.

In this transition period when AI technologies are being used in education, “to unleash their [AI tools] full potential for education, it is crucial to approach these models with caution and critically evaluate their limitations and potential biases, understanding that they are tools to support teaching and learning and do not replace teachers” (Van den Berg and du Plessis, 2023, p. 1). In this study, through the use of the technological pedagogical content knowledge (TPACK) framework enlightening the integration of technology (i.e., particularly AI technology) (Mishra and Koehler, 2008; Celik, 2023) into educational settings, we aimed to examine pre-service, in-service teachers (i.e., both in-experienced and experienced), and a teacher educator's lesson planning with ChatGPT for acids and bases topic, a fundamental topic of middle school science and high school chemistry curricula (Drechsler and Van Driel, 2008). The acids and bases topic encompasses abstract concepts, which makes learning the topic challenging for learners (Cetin-Dindar and Geban, 2017) and leads to alternative conceptions (Nakhleh, 1994; Sesen and Tarhan, 2011).

Previous research has provided information on how instructors and pre- and in-service teachers utilize AI in lesson planning. To the best of our knowledge, there is a gap in the literature regarding the examination of how participants with varying levels of teaching experience utilize the models, their aims in interacting with the models, and whether they engage with the model's offer, and if so, how they handle the offer. Although Clark et al. (2024) argued that prior knowledge is an important factor in AI dialogue, the questions of how it changes the planning procedure and directs the dialogue with AI remain unanswered. By focusing on the points listed, this study has the potential to inform teacher educators about the use of AI tools in teacher education, at which points they should support teachers, and how science and chemistry teaching methods courses can be enriched with the use of AI.

Theoretical framework: from pedagogical content knowledge (PCK) to technological pedagogical content knowledge (TPACK)

PCK is the model that describes how a teacher's understanding of a subject is transformed to make the subject ‘teachable’. Shulman described PCK as it “represents the blending of content and pedagogy into an understanding of how particular topics, problems, or issues are organized, represented, and adapted to the diverse interests and abilities of learners, and presented for instruction” (Shulman, 1987, p. 8). As can be understood from this definition, PCK is a specific synthesis of content knowledge (CK) and pedagogical knowledge (PK). CK refers to a teacher's domain-specific knowledge of the subject matter to be taught. To help readers better understand the CK parts of the study, chemistry objectives stated in the respective curricula (Appendix 1) and common alternative conceptions reported by previous studies (Appendix 2) are provided in the Appendices. PK refers to a teacher's general knowledge about how students learn, manage the classroom, and motivate learners.

In the related literature, there are different PCK models, such as the hexagonal model (Park, 2005) and the Refined Consensus (RCM) model (Carlson and Daehler, 2019). The subcomponents are more or less similar across different PCK models. For example, alternative conceptions students may have about the subject and the difficulties they encounter, as well as learning outcomes in the science curriculum and instructional and assessment strategies. In contrast to other models, Carlson and Daehler's (2019) RCM model distinguishes between three separate realms of PCK: collective PCK (cPCK), personal PCK (pPCK), and enacted PCK (ePCK). While pPCK is formed by an individual teacher, cPCK is “held by multiple educators in field” (p. 82). ePCK is the one that is employed by an individual teacher during teaching and/or planning instruction for a specific topic taught to a specific grade level. In addition to different PCK models, the PCK construct has been enlarged to teach specific constructs (e.g., PCK for Science, Technology, and Mathematics [STEM] (Aydin-Gunbatar et al., 2020), PCK for Nature of Science [NOS]) (Hanuscin et al., 2011; Faikhamta, 2013). Additionally, PCK was enriched with technology integration. The interactions between technology, content, and pedagogy created a new derived construct: technological pedagogical content knowledge (TPACK) (Angeli and Valanides, 2009). Although Carlson and Daehler (2019) did not establish a connection with TPACK for the three different realms they proposed for PCK, this categorization for PCK is also applicable to TPACK (i.e., personal TPACK, enacted TPACK).

Mishra and Koehler (2006) proposed a TPACK framework based on Shulman's (1986,1987) idea of PCK. “At the heart of good teaching with technology are three core components: content, pedagogy, and technology, plus the relationships among and between them.” (Koehler and Mishra, 2009, p. 62). TPACK involves the knowledge of content, pedagogical knowledge, and technological tools to advance the teaching and learning process (Mishra and Koehler, 2006) (Fig. 1).

Fig. 1 TPACK construct and its components (Koehler and Mishra, 2009).

Technology knowledge (TK) is defined as an understanding of technological possibilities, an interest in emerging technologies, and the ability to use technology (Mishra and Koehler, 2006). TK is crucial for teachers to support students' understanding of the subject and material taught (Hasriadi and Nurul, 2023). Teachers with a high level of TK have the ability to integrate technological tools and equipment into their professional and daily lives. They can practically comprehend the support or obstacles technology offers to solve problems (Koehler and Mishra, 2009).

CK, PK, and TK interactions create secondary-level knowledge types, such as Technological Content Knowledge (TCK), which refers to understanding the interactions between CK and TK. It includes knowledge about technologies used in the relevant domain (e.g., chemistry; e.g., utilizing simulations for ionic compound dissolution in water) (Mishra and Koehler, 2006). Technological Pedagogical Knowledge (TPK) refers to the use of technology for educational purposes (Mishra and Koehler, 2006; Mishra et al., 2010). Regardless of the domain (i.e., chemistry or history teaching), teachers' use of Canva and similar technologies to prepare mini-quizzes can be used as an example for TPK.

Finally, TPACK is a special combination of all core elements. TPACK refers to the understanding of the most effective pedagogical strategies that can be applied, as well as the most appropriate technological tools that can be used to teach a particular domain. It is formed by the complex interactions of all knowledge components (Mishra and Koehler, 2006). For example, a chemistry teacher who observes that high school students struggle with abstract thinking uses animation to illustrate ion and electron movement while teaching the topic of electrochemical cells. The chemistry teacher integrates animation technology into the lesson to make the reactions in the half cells, the changes in the electrodes (e.g., the cathode gaining mass), and the ion and electron movements more concrete.

Today's teachers need to master TPACK as much as possible to develop innovative, creative, and effective classroom implementation, given their students' engagement and interest in technology (Li et al., 2022). Teachers should integrate technology into teaching and learning (Lee and Zhai, 2024). As a result, TPACK has become an essential component for teacher training.

To teach a conceptual understanding of science, according to the TPCK [TPACK] framework, teachers need to focus on learners’ difficulties (e.g., visualization of particles) in learning science topics (e.g., dissolution of NaCl salt in water), determine how to help learners overcome their issues, then look for technological applications and integrate them into the lesson. Thus, TPCK is fruitful in helping learners, teachers, and teacher educators in solving instructional problems (Yerdelen-Damar et al., 2017, p. 396).

Hence, more support and emphasis are required in teacher education regarding technology infrastructure to ensure a broader application of the scope of TPACK in learning and teaching processes in the coming years (Hasriadi and Nurul, 2023).

From TPACK to AI-based TPACK

The TPACK model can explain the knowledge necessary for teachers to integrate AI-based tools into education (Celik, 2023). However, for teachers to optimally benefit from the use of AI in education, they need to understand the nature and potential of the pedagogical contributions of AI-based tools and applications (Xu, 2020). “Integrating AI tools into learning activities requires an understanding of GenAI [generative AI] literacy, which includes students' basic operational skills and awareness of GenAI's capabilities, limitations, ethical considerations, and potential impact on learning outcomes” (Lee and Zhai, 2024, p. 1685). Teachers with the necessary knowledge of AI can utilize such applications and tools to enhance student motivation and engagement (Wang et al., 2023) and more effectively select appropriate AI-based tools for teaching purposes (Edwards et al., 2018). In short, understanding AI and its integration into teaching planning and implementation within the TPACK framework is crucial (Celik, 2023; Karataş and Aksu Ataç, 2024; Feldman-Maggor et al., 2025).

The impact of AI on the educational process necessitates a critical reevaluation of the interaction among technology, pedagogy, and content (Ning et al., 2024). In response to the rapid adoption of AI in the field of education, Celik (2023), Feldman-Maggor et al. (2025) and Lorenz and Romeike (2023) proposed that an AI dimension should be added to the TPACK construct. According to Celik (2023), “as teachers have more knowledge to interact with AI-based tools, they will better understand the pedagogical contributions of AI” (Celik, 2023, p. 8) (Fig. 2). To conclude, the TPACK construct is evolving into the AI-TPACK construct. AI-TPACK is an AI-based framework designed to help teachers achieve their educational goals more effectively.

Fig. 2 Intelligent-TPACK framework with its components (Celik, 2023, p. 8).

AI-TK refers to teachers' awareness of AI and their competencies in communicating with AI applications and using AI applications at a basic level. AI-TCK refers to teachers' utilizing and interacting with AI to improve their content expertise. AI-TPK refers to teachers' awareness of the pedagogical elements in the content of AI applications. Ethics refers to having the necessary knowledge to understand the results of interactions with AI applications, distinguish between right and wrong, and evaluate them (Celik, 2023). “In addition to teachers’ technological and pedagogical knowledge, their ethical assessments play an important role in effective AI integration.” (Celik, 2023, p. 2). Examples of ethical issues include biases related to gender, fairness, the ambiguity surrounding the developers of the AI tool, the knowledge provided by AI tools, and inclusiveness. Moreover, hallucination is another issue that teachers consider when using AI. Feldman-Maggor et al. (2025) revealed an example of ChatGPT 3.5's hallucination. A chemistry teacher engaged in a dialogue with AI and asked the students to identify possible alternative conceptions in chemistry to the model. The teacher also asked for references for those possible alternative conceptions. The AI provided publications that did not exist. To conclude, the Institute for Ethical AI in Education (2022) stated that for the better use of AI in education, the issues listed above should be introduced to users (e.g., teachers and learners) to make them aware of both the potential enrichment and the drawbacks of the new technology. Ultimately, AI-based decisions should be critically evaluated by users in light of the points outlined above.

Literature review

AI usage profiles

Research in recent years has developed multifaceted profiles of how ChatGPT is used in education. To the best of our knowledge, these profiling studies have primarily focused on research with students and have been conducted in fields other than science education. For example, Guner and Er (2025) studied the interactions of 136 university students with ChatGPT. The profiles illustrate the extent to which students utilize AI tools and how their use evolves over time with effective guidance interventions. During the profiling process, the ways in which students interacted with AI in three separate sessions were examined, and changes in these interactions were statistically analysed using the McNemar-Bowker test. In addition, by visualizing the transitions between profiles through Sankey diagrams and transition tables, students' profile changes over time were revealed in detail. With this method, students were categorized according to their patterns of interaction with AI, and the effects of guidance interventions on these patterns could be monitored. Similarly, Stojanov et al. (2024) examined college students' trust in ChatGPT across different tasks and uncovered trust profiles such as “versatile low trusters”, “information seekers”, “proactive students”, and “assignment delegates”. These profiles helped us understand which tasks students used ChatGPT for more frequently and how they influenced their trust. In the profile identification process, students who exhibited similar trust patterns were grouped together using Latent Profile Analysis (LPA), taking into account their trust levels in various tasks. With this method, each group was defined according to the way they used ChatGPT, and profiles were created based on their trust levels. In another similar study, where participants were students, the use of ChatGPT by students and their experiences with these interactions were analysed and classified. Students reported using ChatGPT in their introductory programming courses to write and debug code, as well as receive step-by-step guidance. ChatGPT was considered as a useful tool, especially for novice programmers (Scholl and Kiesler, 2024). Wang et al. (2023) examined the AI-supported English language learning processes of 16 primary school students and identified four different AI usage profiles. In the profile identification process, cluster analysis and epistemic network analysis (ENA) methods were used to analyse the ways students interacted with AI. These analyses provided an in-depth understanding of how students utilize AI and how these interactions influence their learning processes. Students who adopted an in-depth learning approach benefited from the personalized guidance provided by AI, while those who used AI passively or mechanically performed poorly. In a study among 489 mathematics teachers in China, Wijaya et al. (2024), whose participants were teachers unlike the other four studies, examined teachers' AI literacy and confidence through five profiles: those with high literacy and confidence, those with medium-high literacy and confidence, those who balance with traditional methods, those with a cautious approach to AI, and those with low literacy and confidence. During the profiling process, the researchers employed the LPA method, considering teachers' AI literacy and confidence levels. This method enabled the categorization of teachers' literacy and trust levels into latent groups, which are related to each other, in order to understand how teachers interact with AI. In the study, LPA was used to identify different interaction patterns in how teachers use AI and how these interactions are reflected in teachers' teaching styles and attitudes towards AI. LPA analysis revealed the most appropriate profile structure considering the relationships between teachers' AI literacy and confidence. These studies emphasized that to use AI tools more efficiently in education, teaching strategies should be customized according to these profiles, considering the diversity of student and teacher interactions.

AI use in lesson planning

Teachers prepare lesson plans to organize classroom practices and plan instruction systematically (Lee et al., 2022). In recent years, AI tools, such as ChatGPT, have been increasingly used in the lesson-planning processes of pre-service teachers (PSTs). When we examine the studies conducted with PSTs, we see how ChatGPT can be an effective tool in lesson planning; however, some difficulties are also encountered. For example, Corp and Revelle (2023) examined the interactions between PSTs and ChatGPT in lesson planning. The study showed that although PSTs used ChatGPT to create and evaluate lesson plans, some participants were unable to complete the process due to the complexity of the guide. Most participants modified their responses and made additional assessments to customize their lessons. Similarly, in another study by Lee and Zhai (2024), 29 science PSTs used ChatGPT to develop lesson plans in accordance with the curriculum objectives. The study revealed that PSTs were generally able to utilize ChatGPT in a manner compatible with the curriculum and teaching strategies, but they encountered some difficulties. It was observed that ChatGPT effectively provided functions such as generating content and encouraging student engagement, but it was insufficient in lessons that required visual information. Participants emphasized the importance of diversifying information sources and developing critical thinking skills due to the potential for misinformation. This study demonstrates that ChatGPT has value as a potential tool in science teaching; however, PSTs require additional training and support for effective use. Another study by Goodman et al. (2024) examined the use of ChatGPT in the 5E lesson-planning process of PSTs. The participants developed lesson plans with AI, then evaluated these plans and discussed the deficiencies. The study reveals that ChatGPT generates erroneous information in scientific content and struggles to create lesson plans that adhere to the 5E framework. But it is successful in developing inquiry skills. Participants learnt to ask focused questions to make ChatGPT's answers more efficient. In conclusion, all three studies emphasize that AI can be a helpful tool in PSTs' lesson planning processes, but human judgment should always be in play, and guidance is critical for effective integration.

The findings of the studies in which the participant group consists of experienced teachers and higher education instructors emphasize that ChatGPT can be an effective tool in teachers' and instructors’ lesson planning processes. However, guidance and critical thinking are important for correct use. Van den Berg (2024) worked with lecturers and trainers from various educational levels, including higher education (both private and public), primary schools, and technical vocational education. In the study, Van den Berg (2024) examined how the participants used AI tools, particularly ChatGPT, in their teaching processes. The educators facilitated their teaching by utilizing AI in tasks such as lesson planning, text translation, creating assessment questions, and promoting critical thinking. The study emphasized the potential transformation of AI in education, while also acknowledging limitations such as accuracy, bias, and reliability. Educators have used AI for personalized teaching and more efficient lesson presentations; however, it was concluded that institutional guidance is necessary for effective integration. In another study by Clark et al. (2024), four experienced general chemistry instructors used ChatGPT-4 to develop their university-level lesson plans for teaching historical experiments (e.g., Thompson's Cathode Ray experiment). The participants utilized ChatGPT in various tasks, including creating lesson plan outlines, discussing teaching strategies, explaining calculations, tailoring explanations to student levels, and developing assessments. The dimensions of instructional objectives, background knowledge, instructional strategies, resources, and assessment were addressed in the research process. The findings showed that ChatGPT helped prepare lesson outlines, suggest resources, discuss teaching strategies, explain calculations, create explanations tailored to student levels, and design assessments. However, it was insufficient in producing visual aids and slides. In this context, it is stated that ChatGPT can complement the teacher's role in creating tutorial content, but cannot substitute it (Clark et al., 2024).

Unlike all the summarized studies, Powell and Courchesne (2024) did not work with any participant group. They analysed the lesson plan documents prepared by ChatGPT and examined ChatGPT's efficiency in creating lesson plans. The study employed a case study design to investigate the effectiveness of generative AI in creating lesson plans aligned with the Massachusetts curriculum framework. The lesson plan was found to be aligned with the 5E instructional model and the curriculum, but it contained missing details and inaccuracies. These findings, similar to previous studies, emphasize the need for teachers to be cautious when utilizing AI outputs. It was also concluded that careful guidance is needed in teacher training to enable educators to use AI effectively. In this context, while the integration of AI in education enhances active learning strategies in science classrooms to increase student engagement and promote critical thinking and deep understanding of scientific concepts, it requires a focus on professional development training for teachers to use these technologies effectively and a human-centred approach and attention to ethical issues such as data privacy (Vorsah and Oppong, 2024).

Although there are studies in the literature on PSTs’ and experienced teachers’ lesson planning with ChatGPT, no study, to the best of our knowledge, compares PSTs, induction year, experienced in-service, and teacher educators’ lesson planning process with AI. In addition, existing studies generally collected a single type of data source (e.g., lesson plan). Given the domain- and topic-specific nature of the TPACK (Chai et al., 2013), elaborating on the lesson-planning process for a specific topic is necessary. With participants’ varying levels of teaching experience and prior knowledge, as well as multiple data sources, this study has the potential to inform the literature regarding the different profiles of AI use for lesson planning, specifically in the context of the acids and bases topic.

The main and sub-research questions guiding the study are:

What are the profiles of participants with different years of teaching experience in interacting with ChatGPT to prepare lesson plans for teaching acids and bases for middle school students?

i. How do participants utilize ChatGPT in the context of lesson planning?

ii. What are the purposes of interacting with the participants and ChatGPT during lesson planning?

iii. To what extent did the participants deal with the content provided by ChatGPT during the lesson plan preparation process and include it in the plan?

Method

Type of study

This study was conducted with a qualitative lens to examine the participants’ profiles during planning for teaching acids and bases with ChatGPT. Qualitative research methods aim to understand the participants' contexts, reasons, deep thoughts, behaviours, and reactions while addressing a particular issue (Cresswell, 2013). Specifically, this research is a multi-case study aimed at gaining insights into participants' interactions with ChatGPT as they prepare lesson plans for teaching acids and bases to middle school students. The case study design is strong in providing detailed and rich descriptions of cases (i.e., schools, teachers) (Merriam, 1998). Different types of data were collected to paint a rich picture of the AI interaction of the participants with varying levels of teaching experience. In this study, we use the term “profile” to refer to qualitatively developed characterizations of participants’ approaches to using ChatGPT in lesson planning, based on patterns observed in their interaction with the model. Contrary to previous studies that formed profiles through statistical clustering methods (e.g., LPA by Stojanov et al., 2024), in our case, qualitative inquiry was utilized to form profiles through interpretive analysis.

The participants

Two pre-service science teachers (PSTs), one induction-year science teacher, two experienced science teachers, and an experienced science teacher educator (STE) participated in this study. The details of the participants are presented in Table 1.

Table 1 The participants’ details

Participants	Gender	Degree	Professional experience	Teaching acids–bases	Perceived tech. competency (over 10)
(–) is used in the sense of absence and (✓) sign is used in the sense of presence.
Zoe	Female	PST	—	—	8
Nancy	Female	PST	—	—	6
Emily	Female	Graduate student	One semester	—	9
Teagan	Female	PhD candidate	Eight years	✓	8
Carter	Male	PhD candidate	Six years	✓	8
Asley	Female	Full professor	Twelve years	✓	9

---

Based on the related literature (Shulman, 1987; Stojanov et al., 2024; Wijaya et al., 2024), the participants' teaching experience, familiarity with AI tools, previous experience of teaching the acid–base topic (i.e., the subject-specificity of TPACK), and how competent they consider themselves in terms of technology use are important for this study.

Regarding familiarity with AI, all participants reported having experience with ChatGPT for general use and educational purposes (e.g., preparing projects or classroom activities). Based on the data collected via the participant information form, all participants previously used ChatGPT and other Large Language Models (LLMs) (e.g., GEMINI). Except for Zoe, the other participants did not have any formal training in using Gen AI tools. Zoe stated that she participated in basic AI training at the entrepreneurship centre for a short time. She had the opportunity to examine and discuss the purposes for which AI applications are used in education, as well as examples of studies on AI. Additionally, all other participants read research papers on AI or its use in education, except Nancy.

Regarding technology competency, Nancy gave herself a six out of ten, indicating that she did not consider herself technology competent. The other participants rated themselves between eight and nine out of ten, indicating that they were highly competent (Table 1).

Regarding the degrees held by participants and their teaching experience, Zoe and Nancy are senior PSTs enrolled in the four-year undergraduate program. They were in the seventh semester of the program (i.e., each semester has 14 weeks). They took general science content courses (e.g., general chemistry, physics), pedagogy courses (e.g., classroom management), technology courses (e.g., instructional technologies), and science-specific pedagogy courses (e.g., science curriculum, science teaching methods I&II). They did not have any teaching experience. Emily, the induction year teacher, graduated from the same undergraduate program that all the PSTs enrolled in and experienced teachers graduated from (i.e., a four-year undergraduate science teacher education program). Then, she enrolled in a graduate program. She was taking graduate courses at the time of this study and designing her thesis proposal on GenAI use in science education. She also started to work in a state middle school three months ago. However, she had not taught the acids and bases topic previously (Table 1). Teagan and Carter graduated from the same science teacher education program and then completed their graduate studies. They have over five years of experience in science teaching at middle schools and were PhD candidates at the time of the study. Both had taught the topic many times previously (Table 1). Finally, an experienced chemistry teacher educator, Asley graduated from secondary science teaching undergraduate and graduate programs. She had taught the topic many times before (Table 1).

The data collection process and the tools utilized

This study was conducted at the end of the Fall semester in 2024. The participant information form, the Content Representation (CoRe) (i.e., lesson plan format), a copy of dialogue with ChatGPT (i.e., conversation history with LLM), a reflection paper on details of the lesson planning with ChatGPT, and online focus group interview data were collected from all participants (Fig. 3).

Fig. 3 The details of the data collection process.

To receive the necessary details of the participants, the researchers prepared a ‘Participant Information Form’. An expert opinion was taken, and revisions were made in light of the experts’ suggestions. In the final form, there were questions about perceived technological competence, participation in training on AI use, familiarity with AI use in education, teaching experience in years, and experience teaching acid–base topics at the middle school level. All participants completed the form at the beginning of the research.

The CoRe (i.e., a lesson plan format) was developed by Loughran et al. (2012) as a matrix format for lesson planning (Fig. 4). The CoRe is both a research tool for accessing science teachers’ understandings of content and a way of representing this knowledge. The CoRe helps researchers codify teachers’ knowledge in the content area under investigation, thereby identifying important content features that science teachers notice and respond to. Each question in the CoRe aligns with a few key ideas identified by the teacher. Due to the matrix format, teachers planning lessons answer the eight questions for each big idea.

Fig. 4 The CoRe.

All the participants were requested to underline the parts taken from the AI's suggestions in the lesson plan submitted to the researchers. Therefore, the researchers could understand the extent to which, and for which components of instruction (e.g., instructional strategy, or learners’ alternative conception) were informed by ChatGPT. In other words, at this stage, we focused on the participants’ enacted TPACK, enriched by the use of ChatGPT.

Yet another data source was a copy of the dialogue with ChatGPT. All participants shared the link to the dialogue for further analysis (e.g., the prompts and the CoRe questions on which more support is needed). Additionally, after preparing the lesson plan (i.e., the CoRe), the participants were asked to write a short reflection paper on their experience preparing the CoRe on acids and bases topics with ChatGPT support. Specifically, we provided the participants with a short description of what we expect:

Please explain in a paragraph how you benefited from ChatGPT in preparing a lesson plan on acids and bases. You can specify how the process progressed, how your lesson plan preparation process with the AI started and continued, in which aspects you received more or less support, etc.

Finally, a focus group interview was conducted to collect rich data to determine at which stage, to what extent, and how the support was received from ChatGPT during the CoRe preparation process by the participants. The interview was then transcribed verbatim. The questions were prepared for the focus group interview and shared with an expert in science education and TPACK. A revised version of the questions was prepared in light of the feedback. This interview was conducted online after all participants prepared the CoRe. It lasted 90 minutes. The questions were:

1. When preparing the plan, the CoRe, on which question did ChatGPT help you more? What is the possible reason for this?

(i) Were there any questions where you used ChatGPT's suggestion without any changes? Can you give an example?

(2) Were there any steps where you did not get any help?

(3) How did you incorporate the answers into your CoRe?

(i) Did you use them directly, or did you make changes in the answers? Explain.

(4) While preparing the CoRe, were there any ChatGPT answers that surprised you positively or negatively? Can you give an example?

(5) Was there a situation where you realized that ChatGPT hallucinated or gave a biased answer? Please explain.

(6) How was the setup of the prompts you sent to ChatGPT while preparing CoRe? When the scope of the prompts changed, did the answers you received change?

(7) Did using ChatGPT on acids and bases contribute to you in terms of pedagogical content knowledge? Were you able to solve the difficulties in teaching this subject with the help of ChatGPT?

Data analysis

Both deductive and inductive coding were utilized to examine how the participants with different years of teaching experience interacted with ChatGPT to prepare lesson plans for teaching acids and bases (Patton, 2002). First, in the deductive analysis, the related literature on AI use in lesson planning informed us. The literature indicated important aspects, namely, to determine how the interaction proceeds, the details of the prompts, how the information provided by ChatGPT is dealt with, and the ethics and other dimensions (e.g., biased information).

In the first round, the researchers coded a randomly selected participant's lesson plan, dialogue with ChatGPT, and reflection paper on how the dialogue proceeded. Then, they came together to compare and contrast the coding. The intercoder agreement was calculated as Cohen's Kappa. In the first round, we obtained a value of 0.44, which falls into a medium agreement category (Landis and Koch, 1977). After coding and discussing the inconsistencies (i.e., the nature of the prompts that are either specific to the big ideas set or curriculum objectives or not, and the presence of ethical issues), we developed an alternative lesson plan. In the second round, the agreement was calculated as 0.84, which indicates a strong agreement (Landis and Koch, 1977). After addressing the minor inconsistencies, researchers coded the remaining data. At the end of this stage, we compiled a summary table for all participants and aspects (e.g., the process, prompts, and handling the content proposed by ChatGPT) (Table 2).

Table 2 Example coding

Participants	The process	Interaction purposes	Dealing with ChatGPT's content offer	Ethics and other issues
Zoe	“First, I tried to fill in the relevant fields given in the blank lesson plan myself. Since I was still at the beginning of my profession and I had not taught the topic of acids and bases before, I thought that I could not answer the questions like a professional teacher. Then I got help from ChatGPT’’ (from reflection paper)	All the alternative conceptions listed and the instructional strategies planned to be used were taken directly from ChatGPT's offer (from the lesson plan, CoRe)	At the very beginning of the lesson, she takes the activity suggested by ChatGPT to attract attention: she pours an acid and base into purple cabbage juice. She asks the students which one is the acid and which one is the base based on the colour change. (From lesson plan) There is no way for them to know this, but they can guess.	Non-existent
		Seek ideas	She introduces the definitions of acid and base by providing a few examples and then asks a question taken from ChatGPT: “Why do you need to brush your teeth after eating candies?” The student does not know the concept of neutralization. She took those from ChatGPT; however, she was not able to critique ChatGPT's offer.
	Dialogue history also supports the reflection paper data
	Preparing a lesson plan with ChatGPT		Mostly unable to deal with ChatGPT's offers
Asley	“First of all, I prepared a lesson plan for three different big ideas. When this work was done, I had a question mark in my mind whether I was able to fully predict the alternative conceptions of 8th graders. Since I did not teach this subject at the 8th grade level, I felt the need to ask.” (from reflection paper) ChatGPT dialogue history also confirmed that.	“I had a prompt about the alternative conceptions. The offers were not at the level I wanted…. But in terms of assessment and evaluation, it saved me time. I mean the concept cartoon in terms of creating a visual.” To check her knowledge about the alternative conceptions she wrote on the plan & to save time and have professional-looking visuals (e.g., concept cartoons) (from focus group interview)	Asley: based on the 8th grade curriculum, can you prepare a branched tree diagnostic test including possible alternative conceptions about acids, bases and their general properties?	Existent
	Preparing own lesson plan and then asking ChatGPT		ChatGPT: sure! Here is a structured branched tree on acids and bases based on grade 8.	“Can you show the sources from where you get those alternative conceptions?” (From the dialogue with ChatGPT)
			Asley asked when she realized some alternative conceptions in this structure, such as acid–base forces, which were not focused on in Grade 8:
			Asley asked: “Are these at the 8th grade level?”
			Then ChatGPT replied: “You are right, the content may have been a bit verbose. At the 8th grade level, I should have presented a more simplified structure, in line with the knowledge level and age group of the students.”
			“I want a concept cartoon on the formulas of acids and bases.” (from dialogue with ChatGPT)
			Able to deal with ChatGPT's offers

---

In the second part of the analysis (i.e., inductive), based on the summary table we created (Table 3), we attempted to categorize the participants' interactions with ChatGPT during the lesson planning process. This part was inductive, and we formed the categories from the data (Patton, 2002).

Table 3 Profile categorization

Participants	Prompts	Dealing with what ChatGPT offers	Interaction purpose	Ethical issue	Profiles
Nancy	Specific to curricular objectives	She took all of ChatGPT's recommended activities without critically evaluating them	Planning by taking all the activities from ChatGPT	She did not address	AI-reliant planner
Zoe	Specific to curricular objectives	She took almost all of ChatGPT's recommended activities without critically evaluating them	Planning by taking almost all of the activities from ChatGPT	She did not address	AI-reliant planner
Emily	Specific to curricular objectives	She took all of ChatGPT's recommended activities without critically evaluating them	Planning by taking all the activities from ChatGPT	She did not address	AI-reliant planner
Teagen	Specific to curricular objectives, specific to the instructional strategy that she planned to implement	She dealt with some of ChatGPT's offers	Preparing her lesson plan and then dialoging with ChatGPT through the plan to address some missing points in the plan	She did not address	AI-collaborative planner
Carter	Specific to curricular objectives	He dealt with ChatGPT's offers	Preparing his own lesson plan and then asking ChatGPT to critique the plan	He did not address	AI-assisted plan refiner
Asley	Specific to alternative conception, specific to grade level	She dealt with ChatGPT's offers	Preparing her lesson plan and then asking ChatGPT questions on two specific points	She asked for the sources where ChatGPT took the alternative conceptions	AI-independent planner

---

In the category-defining first step we determined the endpoints of the spectrum. At the lowest level, a category was defined in which all suggestions from ChatGPT were accepted without question. The process continued by asking ChatGPT for a lesson plan, and there was no consideration of ethical implications. Respondents in this category made plans based on AI. They even had the AI make plans, which indicates AI-reliant Planners. Nancy, Zoe, and Emily were in this category. At the other end of the spectrum there is a profile of a user who asks ChatGPT questions on a few specific points (e.g., alternative conceptions common to middle school students), evaluates the content presented by AI, and even provides the original source articles of the alternative conceptions presented to her, which indicates an AI-independent Planner. Asley was in this category. Between the two, Carter and Teagan's prompts, dealing with offers from AI and the interaction purposes were not only different from the other categories but also different from each other. Additionally, they were neither AI-reliant nor completely independent. Hence, Teagan's and Carter's categories were named in light of their data, which indicated that Teagan was an AI-collaborative planner, whereas Carter was an AI-assisted plan refiner, as Carter did not plan with AI. On the contrary, he asked AI to critique his plan. In this category naming process, for consistency with the literature, we utilized the category names proposed by Guner and Er (2025) and Stojanov et al. (2024) to name these endpoints. They proposed great examples of profile names (e.g., AI-Collaborative Coders, AI-Assisted Code Reviewers, and AI-Independent Coders, all-around low-trusters, information seekers). We changed the ‘coder’ and ‘reviewer’ category names to ‘planner’ for our context.

Finally, to address trustworthiness, data triangulation (i.e., lesson plan, reflection paper, ChatGPT dialogue history, and focus group interview), calculation of intercoder agreement for two rounds, and member check were employed (Patton, 2002). Five available participants were informed about the data analysis results and the profile details. The authors presented the differences and similarities of the participants’ collaboration process with ChatGPT. Those five participants approved the profiles.

The necessary ethical permission was obtained from Van Yuzuncu Yil University, Social and Human Sciences Publication Ethics Committee, with the number 2025/39. All participants voluntarily participated in the study. They signed the consent form before the study started. To ensure anonymity, we utilized pseudonyms.

Findings

In this part, each profile will be presented in detail. A cross-case analysis will be conducted at the end of the findings.

Profile-1: AI-reliant planners

There were three participants in this profile: Nancy, Zoe, and Emily. Both PSTs and the induction-year teacher (Zoe) prepared the lesson plans by mostly relying on ChatGPT. Regarding the planning process, Zoe and Emily prepared their lesson plans with ChatGPT. In the reflection paper, Emily stated:

First, I tried to fill in the relevant fields in the blank lesson plan. Since I was still at the beginning of my profession and had not taught the topic of acids and bases before, I thought that I could not answer the questions like a professional teacher. Then, I got help from ChatGPT.

On the contrary, Nancy preferred to ask ChatGPT to prepare a plan, and then review it. She wrote in her reflection paper:

I needed to have a general plan in mind. I used ChatGPT for this. First, I created a general framework by asking ChatGPT the questions, “Can you prepare a lesson plan for 8th grade science on acids and bases?” and “What activity can I do on acids and bases?”

Second, regarding interaction purposes, AI-reliant planners seek ideas for learners’ alternative conceptions, difficulties in learning the topic, instructional strategies for teaching ‘acids and bases,’ and assessing middle school students’ understanding. The analysis of Zoe's CoRe revealed that all statements about learners’ possible alternative conceptions (i.e., the sixth question in the CoRe) related to the topic were generated by ChatGPT. In the ChatGPT conversation history, Zoe asked, “I will teach the acids and bases topic to my 8th-grade students. Which alternative conceptions are observed in the acids and bases, and pH?” In the CoRe, Zoe listed all the alternative conceptions offered by ChatGPT (e.g., ‘acids and bases are dangerous and should not be touched’, ‘students may think that all acids and bases are strong and dangerous’, ‘acids are always hot, bases are always cold because students may think that acids are hot because they burn, while bases feel cold’).

While preparing the lesson plan, Zoe used ChatGPT to identify the alternative conceptions that students may have. Since Zoe did not verify whether the alternative concepts presented by LLM ChatGPT were already in the literature, the authors did so. The literature review showed that some of them had equivalents in the literature. For example, statements such as “acids and bases are dangerous, so they should not be touched”, “acids are always harmful, bases are not harmful”, and “everything that tastes sour is acid” were reported by Ross and Munby (1991). Likewise, “cleaners such as soap and detergents are only basic” (Nahadi et al., 2022), and the idea that “pH value is just a number” (Mubarokah et al., 2018) exist in the literature. However, “the colour change of the reagents is irreversible”, “acids are always hot, while bases are cold”, and “pH measurement can only be done with substances such as litmus or phenolphthalein” are not reported as alternative conceptions in the literature.

In the focus group, Zoe stated:

I mostly benefited from ChatGPT in terms of the alternative conceptions and typical difficulties that students might have. I needed this because I thought that I would not be able to identify them since I had no previous experience teaching the acid–base topic.

In addition to alternative conceptions, the analysis of their CoRes revealed that they took almost all the instructional activities written on the CoRe from ChatGPT. For instance, Emily aimed to teach the students how to compare acids and bases, classify substances using the concept of pH, and the similar properties of acids and bases and their uses in daily life (i.e., three big ideas Emily set). To achieve her goals, she planned to discuss the properties of acids and bases using a two-column table, focusing on pH measurement and the daily life uses of acids and bases, incorporating simulations and videos. Likewise, Nancy stated in the focus group interview that she received the most help finding activities suitable for the topic.

The part where I used the AI most was instructional strategy. For example, I noted that these pH-related activities are somewhat abstract. I asked if we could show students a simulation about it [to make the topic more concrete]. ChatGPT recommended a website called PhET Simulations… It was great to have a simulation idea.

Similarly, in the assessment part of the CoRe (i.e., the last question), Emily utilized all assessment strategies provided by ChatGPT. She took the idea of asking learners to prepare a poster or class presentation about acids and bases, supported with visual material, make a list of substances containing acids and bases that they use at home, and explain the properties of these substances. Additionally, she copied ChatGPT's offer: “Learners can be given self-assessment questionnaires that include questions such as “Which concepts of acids and bases did I find difficult?” and “At which point am I still undecided?”. To conclude, the role that the AI-reliant planners assigned to AI was knowledge authority.

Finally, regarding the integration of information provided by ChatGPT, the AI-reliant planners encountered issues with incorporating ChatGPT's suggestions into their plan. For instance, in her CoRe, Nancy set the third big idea as “learners will be able to classify these substances by the use of pH value to understand the differences and similarities between acids and bases, and generalize those substances by creating a pattern.” To achieve this goal, Nancy stated that she would use the PhET simulation proposed by ChatGPT. However, when we examine the flow in the lesson plan, it is evident that she plans to use this simulation suggestion without even defining pH. In other words, when integrating ChatGPT's suggestion, she could not focus on the flow of the lesson. Similarly, ChatGPT proposed an alternative conception that 'acids are always hot, and bases are cold.' Zoe took it without inquiry and wrote it to the CoRe. Since her experience and reading of the literature were insufficient, she took it directly. When we reviewed the conversation history, we noticed that she did not send any follow-up prompts to question ChatGPT's offers. Additionally, there are no instructional activities to address the alternative conceptions identified.

In Emily's CoRe, similar to Nancy's CoRe, she took the idea of preparing solutions of varying concentrations and forming a pH spectrum, which is consistent with her 2nd big idea (i.e., classifying substances by their use of pH). However, for the 1st big idea (i.e., comparison of acids and bases), she took ChatGPT's offer, which was “Prepare a table with two columns (acids and bases). Ask the students to write down the properties appropriate to each group. Then discuss with the whole class, emphasizing the common characteristics.” Although this idea may work, it seems more like a method for diagnosing learners’ prior knowledge about the topic rather than a teaching strategy. Without any supportive hands-on activity or video, only taking learners’ ideas, it seems that she did not criticise the suggestions made by ChatGPT.

Profile-2: AI collaborative planner

Teagan's planning with the AI was categorized in this profile. First, in the reflection paper, Teagan wrote that she considered each aspect of the lesson plan (e.g., alternative conceptions, assessment), and then she engaged in a dialogue with the model. Then, she synthesized her ideas and the model's suggestions.

Second, she aimed to secure the model's offer to complement her own. In the focus group interview, she stated:

In the alternative conception section, I outlined some possible options based on my experience. I also found ChatGPT's suggestions useful. Likewise, in terms of typical difficulties, it offered different perspectives in addition to the situations I experienced, and I blended them into my plan.

Teagan stated that the alternative conceptions written in the plan were not based on her own classroom observations and experiences. She received some support from ChatGPT in this process. She added the alternative conceptions suggested by AI. Similar to previous ones, Teagan also did not check the ChatGPT's output for students’ alternative conceptions. The authors checked all of them. For instance, “all acids are harmful” and “acids and bases are opposite substances, they have no common properties” overlap with the ones previously identified in the literature (e.g., Ross and Munby, 1991; Mubarokah et al., 2018). Likewise, “thinking that all acids are strong and all bases are weak” was revealed in the literature (Ross and Munby, 1991; Mubarokah et al., 2018; Nahadi et al., 2022). On the other hand, “pH of strong acids is close to 1, pH of weak acids is close to 7; pH of strong bases is close to 14, pH of weak bases is close to 7” were not found in the literature. Similarly, “thinking that acids always show acidic properties when they react with water” and “focusing only on the properties of acids and bases against water” were not found in the literature.

In this interaction, Teagan used ChatGPT as if she were exchanging information with a colleague to complete the missing points of the plan in her mind (i.e., complementary role). In her CoRe plan, Teagan, who planned to use a design-based STEM activity, asked ChatGPT to write a scenario for daily life related to this activity. In other words, she had a plan with a specific aim and flow. She only needed a daily-life problem to make the STEM activity more interesting and attractive for students, which was copied below.

You are on the R&D team of a Baby Care Products company. Recently, the company received many complaints that the baby shampoo irritates babies’ eyes. The company officials are asking your team to design a shampoo formula that does not harm or irritate babies' eyes.

In the focus group interview, Teagan stated:

I integrated the support I received from ChatGPT into my plan. I used ChatGPT's suggestions in some question examples and scenarios that I would use in the lesson because it made my job easier… I have taught acids and bases many times in my classes. However, ChatGPT has different ideas. It puts everything packaged in front of you. There will be some things you take from these. There are some situations that you add to your repertoire.

Finally, when she issued ChatGPT's offer, she filtered it through her pedagogical content knowledge, which she gained through eight years of teaching experience at middle school. In the reflection paper, Teagan wrote, “I blended my eight years of teaching experience at middle school and ChatGPT's suggestions.”

She described how she dealt with ChatGPT's suggestion in the focus group interview.

I did not want ChatGPT to intervene too much. I found it [ChatGPT's suggestions] superficial. In terms of alternative conceptions, for example, I found it practical to consider its suggestions and adopted the three or four alternative conceptions it proposed. ChatGPT offered seven alternative conceptions. When I checked them, I noticed some things I observed in the classroom. Some of them were unreasonable; I did not take them. ChatGPT can assist us in presenting packaged information. In this sense, ChatGPT really saves time. Can it be adapted to the classroom? It can definitely be adapted, but of course, it should be done to the extent that the teacher knows his/her students.

Likewise, she took some of the items written in the assessment part of the plan. The question taken from ChatGPT is copied below.

She took three questions offered by ChatGPT, which were a good match with the activity she planned. In the CoRe, Teagan planned to ask students to observe the colour changes by dropping black cabbage juice on different solutions prepared (e.g., vinegar). pH strips were introduced, and students were asked to measure and note the pH.

In a chemistry laboratory, a group of students work with four different liquid samples, measuring the pH levels of different liquids. The students conduct a test using litmus paper to measure the pH of each liquid. At the end of each trial, they would note the colour of the litmus paper and determine whether the liquids were acidic or basic. The students obtained the following results:

Lemon juice: Red

Soapy water: Blue

Vinegar: Red

Carbonate solution: Blue

Based on these data, which liquid is acidic and basic? Explain in which ranges the pH values of these liquids can be found and in which products these liquids can be found in daily life.

Profile-3: AI-assisted plan refiner

Carter's planning experience with ChatGPT was in this profile. His plan with ChatGPT was unique. He wrote in the reflection paper:

I started my ChatGPT dialog with a prompt like “I prepared a 5-hour lesson plan on acids and bases for 8th grade in a middle school with 45 students. Can you provide suggestions and criticisms for the plan?” However, ChatGPT created a lesson plan directly without criticizing my own. At this point, I changed my prompt. After I restated that I had prepared a plan and wanted to move forward with it, the conversation became productive.

The process consisted of the AI reviewing the existing lesson plan and providing suggestions.

Second, as the flow revealed, the dialogue with ChatGPT aimed to ask the AI to critically examine the plan with a well-defined context (i.e., the number of students, the length, grade level, and the topic). Carter gave ChatGPT a critic and, to some extent, a complementary role. For instance, he mentioned an interesting case in the focus group interview.

… There was a point where I said how I could have forgotten it. I do not know why I did not put it in my plan. There was a point where ChatGPT made a great point; it noted that we need to assess students' readiness levels. Immediately afterward, I asked it to explain this more. It stated that 8th-grade students generally struggle to understand abstract concepts. Therefore, it said to be careful to make a plan that includes visuals, experiment, activities, and videos.

Finally, regarding dealing with ChatGPT's offer, due to the difference in his interaction with ChatGPT, Carter mainly dealt with the model's critics. For example, Carter's plan included a long section on the properties of acids and bases. At this point, the model suggested that he divide the section and include activities to ensure the student's active participation and attention. When we examined Carter's CoRe, we realized that he divided the related section into sub-sections between which he included videos focusing on acids and bases properties, a game (i.e., in which the properties are read to determine whether the substance belongs to the acid or base class), and brainstorming (i.e., to attract students’ attention). Likewise, ChatGPT criticized a point that made him delete the part from the plan. In the focus group interview, he explained it:

In my plan, there was a section mentioning the taste of acids and bases. ChatGPT issued a serious warning about this. It said, “Don't use this example [the point about tasting soap]; take this example out.”

Regarding the complementary role, in the reflection paper, Carter wrote that although he did not write a specific prompt for students’ alternative conceptions of the topic, ChatGPT provided possible ones. In his CoRe, in addition to his statements of the possible alternative conceptions (e.g., increasing acidic properties as pH increases, acids and bases always cause colour change in litmus paper), he also added ChatGPT's offer (e.g., bases are harmless, acid with a pH of 5 is twice as acidic as an acid with a pH of 6). In the focus group interview, Carter stated that he received some ideas related to alternative conceptions; however, he also wrote many other possible ones based on his experience in teaching. He added:

In the process, I benefited from ChatGPT because its suggestions were efficient, paying attention to the student's readiness and highlighting key aspects in a specific part of the plan.

Carter did not check the existence of the alternative conceptions. When we checked them, we found that the first one (i.e., bases are harmless) was reported by Bučková and Prokša (2021). The second alternative conception related to pH was not found in the literature as presented here, but it was stated that students had difficulty understanding the concept of pH and could not fully comprehend what the mathematical value of pH meant (Nakhleh, 1990; Sheppard, 2006). In our search, we discovered that Sheppard (2006) employed a similar task. “Students were shown beakers with colorless solutions marked ‘pH 3’,’ pH 5’ and ‘pH 11’, and were asked to explain their sub-microscopic composition using drawings.” (p. 34). Sheppard (2006) reported:

Only two students defined pH as pH = −log[H⁺] and only one of these was able to explain correctly the hundred-fold difference in H⁺ concentration between the pH three and pH five solutions, despite all students having had instruction that had emphasized the use of the equation, including several simple calculations of pH values from H⁺ concentrations. For most students, pH was a linear scale (p. 36).

Although not exactly the same, an alternative concept with a similar meaning was presented by AI. Carter received ChatGPT's suggestions by filtering them through some criteria (i.e., students’ readiness). However, those are mainly based on the pedagogical aspect rather than PCK. The former focuses on pedagogy that is not specific to the content of acids and bases. The latter is bifocal with both pedagogy (i.e., teachers’ knowledge about 8th-grade students’ characteristics, such as difficulties in abstract thinking in general) and content (i.e., teachers’ knowledge about 8th-grade students’ readiness, difficulties, and alternative conceptions related to acids and bases for 8th-grade students).

Profile-4: AI-independent planner

Asley is the only participant in this profile, and her process was also unique. First, she prepared a plan with three big ideas (i.e., the definition of acids and bases, the formulas and properties of acids and bases, and the acid/base pattern with pH values). Later, she was unsure whether she could fully predict the alternative conceptions of 8th graders since her expertise was in high school chemistry education and chemistry teacher education.

Second, in terms of the purpose of the interaction with the AI, analysis revealed two main purposes. The AI-independent planner's first interaction purpose was to check her knowledge about the alternative conceptions of middle school students, a different group from high school students, with which she is familiar.

The second purpose was to create professional-looking visuals (i.e., concept cartoon and diagnostic branched tree with possible alternative conceptions), thus saving time. In the focus group interview, Asley said:

I am unable to create a concept cartoon by myself. I use ChatGPT Pro, especially when creating visuals like these. For example, I get help when I need to draw at a particle level or when I need to create a concept map. Drawing on a computer is very difficult and time-consuming, and the result often appears amateurish when compared to drawing by hand. In this sense, I think it will help me with measurement and evaluation.

In her dialogue with ChatGPT, first, she wrote a prompt: “Can you draw me a concept cartoon on acids, bases, and pH at an 8th-grade level, including alternative conceptions?” The concept cartoons offered by ChatGPT did not satisfy Asley, so she wrote, “I want a concept cartoon with alternative conceptions related to the formulas of acids and bases.” She integrated the concept cartoon into her plan (Fig. 5).

Fig. 5 The concept cartoon Asley took from ChatGPT.

In the reflection paper, Asley also wrote:

For the assessment section, I also asked ChatGPT to form a diagnostic branched tree to assess the extent to which the lesson I planned addressed the learners’ alternative conceptions. However, I did not use the suggestion that ChatGPT provided because it was not visual, but rather in a statement form. So, I drew it by myself.

Asley's CoRe analysis revealed that she accepted ChatGPT's offer only for the assessment part. As mentioned above, she planned to use the concept cartoon that ChatGPT drew. However, the use was not directly taken from the cartoon. On the contrary, she planned to revise it. In the focus group interview, Asley stated:

This was not the concept cartoon I had envisioned, but I had indicated in my plan that I would use it with some revisions. In the speech bubble, ‘I am an acid and my formula starts with H’. In the cartoon, the child also has N and H atoms on his sweatshirt. I will revise it by making it NH₃. The child will ask the question, ‘Then is NH₃ also an acid? My students will answer that child. There is an alternative conception among the students: “Acids are chemicals containing hydrogen. Bases contain OH.” … I envisioned using it by making such an arrangement.

In other parts of the CoRe, she relied on her repertoire, which appeared to be rich. For example, ‘identifying acidity/basicity pattern with pH value’ is a big idea she set; Asley planned to use an activity that allowed students to measure the pH of chemicals from daily life and create their pH scales. The students then use those pH values to create their pH scales by colouring them with crayons and exhibiting them in the laboratory or classroom. Asley also proposed a similar activity for a school without a laboratory environment. She proposed an option that included PhET Interactive Simulations (i.e., the pH Scale simulation, which lets students measure daily life solutions such as milk, coffee, vinegar, orange juice, vomit, blood, soap, and drain cleaner). In the plan, Asley emphasized that the visuals of ions in the acidic environment should also be shown. In this way, she stated that she planned to address the alternative conception of “there are no OH⁻ ions in acidic solutions or no H⁺ ions in basic solutions” mentioned in the previous alternative conception dimension.

Finally, regarding the information provided by ChatGPT, although Asley used AI in a limited way, based on the dialogue history, she always inquired about the information presented to her in this dialogue. When she asked for a diagnosed branched tree, the following dialogue took place.

Asley: Can you prepare a diagnostic branched tree with possible alternative conceptions about acids, bases, and general properties based on the 8th-grade Turkish curriculum?

ChatGPT: [ChatGPT provided True and False statements] …. [F] Neutralization always occurs at pH 7…… The pH after neutralization depends on the strength of the acid and the base [T]. Strong bases can also be harmful [T]……

Asley: Are these for 8th-grade level students?

ChatGPT: You are right; the content may have been detailed. At the 8th grade level, I should have presented a more simplified structure in accordance with the students' knowledge level and age group.

When we reviewed the related literature on the existence of the “neutralization always occurs at pH 7” alternative conceptions, we found references to Schmidt (1995). In the reflection paper, Asley stated that although she specifically asked for 8th-grade students, ChatGPT's offer included information not taught at middle school (e.g., acid strength and neutralization of weak and strong acids and bases).

In another dialog, Asley asked ChatGPT to draw a concept cartoon with an alternative concept about acids, bases, and pH at the 8th-grade level. After ChatGPT's suggestion, Asley asked, “Which alternative concept is this about? Can you explain what the students think is wrong here?” ChatGPT replied: “This cartoon targets an alternative conception about the relationship between litmus paper and pH. The alternative conception is: The idea that bases (e.g., soap) will also turn litmus paper red.” When we reviewed the literature, we could not find it. As can be seen, her dialogue is based on questioning what the model offers. Asley's knowledge of the learner and curriculum enabled her to filter the suggestions presented to her.

Finally, as observed in Asley's case, the dialogue history showed that Asley deeply evaluated ChatGPT's offer. Asley asked the model to give citations for the possible alternative conceptions provided. ChatGPT replied to her:

My information about alternative conceptions is based on common findings in general education research and science teaching. However, for direct references, I cannot say at this time that this information comes from specific academic studies or teaching guides because it is based on general pedagogy and science education literature (from dialogue history).

In the following prompt, Asley wrote:

I would like information on which alternative conceptions 8th-grade students have about the general properties of acids and bases. Provide article tags for direct quotations from current academic articles or guides related to science education.

ChatGPT stated that it scanned five websites and provided five resources related to learners’ alternative conceptions of acids and bases (e.g., Mubarokah et al., 2018) (from dialogue history).

Cross-case analysis

To make the cross-case analysis more effective, we formed Fig. 6.

Fig. 6 Cross-case analysis with all aspects.

To conclude, profound variances were observed between cases with limited experience and cases with long-term and varied experience (i.e., both teaching and researching acids and bases) in terms of the flow of using the AI, the purpose of use, and the rate at which the model suggestions were reviewed and incorporated into the lesson plan.

Discussion and conclusions

In this era, in which AI tools have rapidly entered our lives including in scientific studies and education, it is necessary to examine the use of these tools for educational purposes by learners and teachers. Specifically, “a qualitative investigation of how teachers choose and arrange content, instructional strategies, and particularly ChatGPT functions is crucial” (Lee and Zhai, 2024, p. 1652). To address the gap in the literature, this multiple-case study aimed to examine the lesson planning processes of pre-service teachers (PSTs), teachers, and a science/chemistry teacher educator with different experience levels in teaching acids and bases topics to Grade 8 students through ChatGPT interaction. To achieve this aim, a qualitative analysis of multiple data collection tools focused on how participants with different years of experience differed in the lesson planning process, the purposes of interaction, and their approach to dealing with content provided by ChatGPT. As a result of the analysis, the existence of different profiles and the details of the lesson planning process for participants in these profiles, as facilitated by ChatGPT, were summarized.

In our cases, two PSTs and Emily (i.e., a teacher in her induction year) adhered to ChatGPT while preparing lesson plans for teaching acids and bases. They also directly incorporated ChatGPT suggestions into the lesson plan. Additionally, two PSTs and Emily viewed ChatGPT as a source of information, with their intended use being to seek the model's suggestions about instructional strategies, learners’ difficulties, and alternative conceptions related to acids and bases. However, they did not need assistance in setting curricular goals and objectives. Likewise, Lee and Zhai (2024) reported that PSTs have a high dependency on ChatGPT in lesson planning. Additionally, Lee and Zhai (2024) stated that participating PSTs did not need support in every component when preparing a lesson plan. Similarly, in their study, the PSTs were more proficient in curriculum goals and instructional strategies than in integrating those aspects for effective planning.

The literature on AI use by teachers emphasizes the necessity of professional knowledge to prepare a successful lesson plan to help students learn a concept using AI tools (Van den Berg and du Plessis, 2023; Zhai, 2023). The first aspect of this knowledge is technological knowledge (TK), which encompasses the nature of knowledge produced by AI tools (e.g., biased knowledge) and how to effectively deal with and integrate that knowledge into planning (Zhai, 2023). Based on the participants’ self-reports, only Nancy rated herself 6 out of 10 on technology competency, meaning she considers herself above average. On the other hand, Zoe (8) and Emily (9) consider themselves reasonably competent. Moreover, Zoe stated that she had previously participated in basic AI training. Other experienced participants also consider themselves reasonably competent in this sense. Since we did not measure TK and technology proficiency, which is one of the limitations of this study, we have to rely on participants' self-reports. When we consider all the data we collected, the participants have standard features, except for Zoe, who did not receive training on AI tools and is familiar with ChatGPT. Therefore, we argue that a different dimension should explain these profile differences: the purposes of the AI use, the nature of the interaction (e.g., seeing this technology as a source of information), and PCK.

The second aspect of the necessary professional knowledge is PCK and, with the integration of AI technology, TPACK (Van den Berg and du Plessis, 2023; Zhai, 2023).

If teachers simply focus on just integrating cutting-edge GenAI technologies such as ChatGPT into their pedagogical practice without careful consideration of how to best use them, it can be a waste of resources that fails to have a meaningful impact on learning outcomes (Lee and Zhai, 2024, p. 1652).

Careful consideration that Lee and Zhai (2024) indicated can be possible with a solid PCK and TPACK. The studies in the teacher education literature revealed that with experience, teachers could coordinate multiple knowledge types (e.g., content knowledge [CK], pedagogical knowledge [PK], and technology knowledge [TK]) and PCK components (e.g., learner, curriculum, and assessment knowledge) for solid instructional planning and decision-making (Gess-Newsome, 1999; Akin and Uzuntiryaki-Kondakci, 2018). However, PSTs have limited CK, PCK, and TPACK, and require PST education programs to enrich their development (Jong et al., 2005; Akin and Uzuntiryaki-Kondakci, 2018; Tondeur et al., 2020; Ekiz-Kiran et al., 2021). Abell et al. (2009) argued that at the beginning of their careers, PSTs are observers unaware of how the classroom works. In the professional continuum, structured training and courses (Aydin et al., 2013), as well as experience, are valuable sources for PCK (Grossman, 1990) and TPACK (Tondeur et al., 2020) development. The experienced teachers in this study and the teacher educator had teaching experience in addition to other areas (e.g., research on acids and bases, alternative conceptions, and experience).

Regarding the influence of prior knowledge on AI use profiles, limited studies have examined the interaction process between prior knowledge and AI in the fields of chemistry education (e.g., Clark et al., 2024) and programming learning with AI (e.g., Guner and Er, 2025). Clark et al. (2024) revealed that chemistry professors’ prior knowledge about historical experiments (Thomson, Millikan, and Rutherford's experiments) influenced how the professors engaged with AI tools. Likewise, Guner and Er (2025) reported that the participants with limited prior programming knowledge showed a superficial use of the tool. In contrast, those with strong knowledge “prefer to first try coding the task themselves, then they ask ChatGPT for revision or refinement” (p. 20). Likewise, in this study, experienced science teachers’ dependence on ChatGPT was less than that of PSTs. Furthermore, experienced teachers viewed ChatGPT as a colleague, so their process was more akin to exchanging information with a colleague or critiquing a lesson plan. Furthermore, Teagan, Carter, and Asley incorporated a filter role by evaluating the model's suggestions. However, PSTs with limited or almost no experience, for example Emily, could not play that role, which is in line with Zhai's (2023) point. “Teachers must also be able to evaluate the quality and relevance of the information provided by ChatGPT and make informed decisions about its use in the classroom.” (p. 45). As stated above, we argue that years of observing their students and seeing the students' difficulties and mistakes in their exam papers may have served as this filter. The results showed they need limited assistance, especially with learners’ difficulties and alternative conceptions.

Another important point regarding the difference between the profiles was paying attention to ethics and other related issues (e.g., bias, hallucination). Due to the non-questioning nature of the participants except for Asley, the authors checked these alternative concepts presented by AI from the literature. Some of the alternative conceptions offered by the model are found in the literature, while others are not. In fact, in a few cases, alternative conceptions that are not exactly the same but may have a close meaning have been encountered. Moreover, the concept cartoon (Fig. 4) created by the AI also had some issues regarding ammonia's formula (i.e., the formula was given as NH, but it should be NH₃). Likewise, the formula of sodium hydroxide was also problematic (i.e., on the blue bottle the formula was written as Naoh, but it should be NaOH). The results showed that content AI provides should be filtered and checked, which is one of the important suggestions that AI users should adhere to (Ivanova, 2025). Because ChatGPT can generate information that lacks a scientific basis or has not been proven accurate, it can exhibit artificial intelligence-induced hallucinations (Tyson, 2023). Only Asley's data had an example of this aspect. She prompted ChatGPT regarding the references to the alternative conceptions that ChatGPT listed. Although other participants, except Nancy, mentioned reading papers about AI use in education, they did not address these issues in their dialogue with ChatGPT. It seems that an important dimension of the TPACK-AI has not yet taken a significant place in the participants' minds. Reading papers and participating in a brief training process did not yield results for the participants. More emphasis and longer training are needed, as noted by Feldman-Maggor et al. (2025).

Yet another point is that, contrary to the PSTs and induction year teachers’ inconsistent decisions in the plan flow, the experienced teachers' and teacher educator's plans with ChatGPT were consistent throughout the lessons. In other words, two experienced teachers and the teacher educator could successfully synthesize their knowledge and experience with ChatGPT's recommendations in a logical sequence. Regarding the issue, Van den Berg and du Plessis (2023) stated that one of the participants in their study (i.e., an experienced teacher) made an excellent analogy for the use of ChatGPT to plan a lesson, that is, “a recipe which still needed a chef.” (p. 6). Based on this analogy and in light of the findings, as teachers gain experience in their careers, they become more capable of critically analysing the lesson plans presented to them. It appears that this critical view is based on several variables, including the level of the students, curricular gains, and the time and flow.

Given that the experienced teachers had master's degrees, these graduate courses may also have enriched their PCK and TPACK. In other words, those courses may have strengthened the filter for the model's suggestions evaluation, in addition to providing teaching experience. As Abell et al. (2009) argued, over time and through various roles (i.e., observer as a PST, apprentice as a graduate student, and independent instructor after PhD), PCK is enriched. Likewise, this aspect could explain the experienced teachers’ (i.e., Teagan and Carter) and teacher educator's (i.e., Asley) profiles. Asley's graduate and PhD education, through which she took advanced teaching and assessment methods courses, her research, and others, most likely informed her about many aspects of the lesson plan. Additionally, considering her experiences as a PhD candidate, her contributions to the planning and teaching of courses, her observations, and her own teaching experiences as an instructor, it is understandable that she only wants to see the model's suggestions on specific points (e.g., 8th-graders' possible alternative conceptions related to the topic). Similarly, Clark et al. (2024) argued that “for instructors seeking to improve their content knowledge, a conversation with ChatGPT can be useful. Conversely, instructors with extensive prior knowledge may have difficulty setting aside their expertise when communicating information to a novice.” (p. 1995).

Finally, teachers' critical approach to AI when using AI-supported systems carries an important responsibility, not only in terms of their pedagogical competencies but also in terms of learning effectiveness and cognitive load. On the other hand, the ethical aspects of algorithms are also another issue. When discussing the ethical aspects of algorithms, Mittelstadt et al. (2016) argue that technical features of systems, such as transparency and auditability, alone do not solve ethical problems. AI systems in education can make seemingly unbiased and objective decisions, but these decisions can create unpredictable inequalities in students' learning processes. In this context, teachers need to be active actors in utilizing the outputs of AI systems while questioning their ethical and pedagogical implications. Finally, the concept of the “help dilemma,” defined by Koedinger and Aleven (2007), emphasizes the importance of finding the right balance between when and how to offer help from a pedagogical perspective in AI-supported instructional systems. As our findings show, the PSTs (Zoe and Nancy) and induction year teacher (Emily) displayed AI-reliant profiles. While this dependency is understandable given their lack of teaching experience, it is our responsibility as chemistry teacher educators to consider how we can reduce this dependency among our pre-service teachers, who are the teachers of the future, and to take action. If we do not reduce this level of reliance, there is a risk (Dodge et al., 2022) that PSTs will not be able to develop their thinking skills sufficiently and will not be able to use different PCK and TPACK components in a way that informs each other. Implications of this will be suggested in the following sections.

Implications

As the results showed, inexperienced participants (i.e., PSTs and induction-year teachers) tended to accept AI-generated content more readily, while experienced ones were more critical. This distinction necessitates the design of training programs with distinct content for pre-service teachers, teachers, and teacher educators at varying levels of experience. For inexperienced ones, the training on AI use in chemistry education courses should be planned and provided. During the course, chemistry teacher educators should critically evaluate the content of AI-generated plans or activities. When PSTs are faced with inaccurate or pedagogically inappropriate content generated by AI, they begin to think that the content should be critically evaluated. Additionally, in the course, the biases (gender, language) and limitations reported in the related articles (e.g., Feldman-Maggor et al., 2025) may be utilized. When the PSTs read articles with biased examples, they would come to understand the necessity of critically examining AI-created plans. Furthermore, for the pedagogical adaptation of AI-generated content, mentoring support (Aydin et al., 2013; Koh, 2020; Baser et al., 2021) has potential. With an experienced teacher educator (i.e., the mentor), experiencing how to interact with AI, how to elaborate on their offers for richer planning, and how to write quality prompts to achieve their goals would be useful for PSTs.

For experienced in-service teachers, elaboration on how to advance integration strategies, focusing on how AI can enrich existing plans, and collaborative lesson planning with AI to develop individual and group PCK (pPCK and cPCK) (Carlson and Daehler, 2019; Forsler et al., 2024) would be valuable. Collaborative lesson planning with AI research should be conducted on different topics with distinct natures. Acids and bases are among the primary topics in chemistry education, with daily-life examples, calculations, and particulate-level representations (Drechsler and Van Driel, 2008; Cetin-Dindar and Geban, 2017). More research on different topics with nature of science links (e.g., periodic table), abstract concepts (e.g., orbitals and hybridization), and more calculations (e.g., physical chemistry topics), and graphics (e.g., enthalpy of reactions) should be conducted to enrich the chemistry education literature with AI use.

Finally, teacher educators and other instructors require training on utilizing AI to enhance chemistry instruction at the tertiary level (Koh, 2020). The limited aspect of AI tools, ethical issues, and their strengths should be introduced to the teachers and instructors (van den Berg and du Plessis, 2023; Zhai, 2023). That introduction should be domain-specific (Feldman-Maggor et al., 2025) because TPACK and PCK are.

Limitations

This study is limited to six participants with varying teaching and academic experience. In future studies, it would be beneficial to include more participants from each level of teaching experience to gain a more detailed understanding of lesson planning with AI. Although the TPACK framework informed this study, the revised TPACK framework may not be sufficient in terms of AI use and the integration of AI into science and chemistry education. The ethical issues that AI brings with it, as well as the competencies required for its effective use, are the key areas that need to be studied. Hence, the use of different frameworks in future studies may lead to different perspectives and interpretations. Moreover, all the participants preferred to utilize ChatGPT. Diverse LLMs (e.g., Copilot, Gemini, Teachbot) should be utilized, and their dialogue and suggestions should also be compared. The study focused on the acids and bases unit. Research in other units may provide different results. Additionally, the use of qualitative data sources and their qualitative analysis. Although qualitative data provides rich and in-depth data, it may be influenced by the researchers' perspectives during analysis. In future studies, we recommend mixed-design studies that include quantitative data collection tools to address subjectivity. Furthermore, the dialogue with ChatGPT was in Turkish. More research can be conducted in other languages, as the literature is richer in English and German on chemistry education, which may result in more comprehensive lesson plans. Finally, when the cognitive abilities of the LLMs are assessed, Ivanova (2025) stated, “Do evaluate models under conditions similar to humans when comparing their performance… If a human needs instructions to perform well on a test, the model should also receive those instructions.’’ (p. 232). Based on the statement, when planning a lesson, teachers conduct research, reflect on their previous experiences, and consult chemistry textbooks. In this research, the participants were given complete freedom in the planning process. We only provided the topic, grade level, and lesson plan format. We did not provide any direction for participants to train LLMs for better results. If LLM had been trained by providing examples multiple times, it might have been possible to give more accurate answers and utilize existing articles.

Author contributions

Sevgi AYDIN GUNBATAR contributed to the conceptualization, data curation, data analysis, validation, and writing, as well as reviewing, revising, and editing the manuscript. Gizem TEZCAN SIRIN contributed to data curation, data analysis, writing, and revising the manuscript. Onur Can ILKYAZ contributed to manuscript writing, data curation, and data analysis. Yusuf MUTLU participated in data curation, analysis, and manuscript writing.

Conflicts of interest

There are no conflicts to declare.

Data availability

The data are not available for public release. However, the authors will share the data (i.e., in Turkish) upon reasonable request.

Appendices

Appendix 1: Focused learning outcomes for grade 8 in Turkish Middle School Science Curricula

Science [S].8.5.4.1. Compares the general properties of acids and bases.

(a) Identifies the properties of acids and bases.

(b) Lists the similar properties of acids and bases.

(c) Lists the different properties of acids and bases.

S.8.5.4.3. Uses “pH” values related to the acidity and basicity of substances to reason inductively.

(a) Finds patterns related to the “pH” values of substances.

(b) Makes generalizations using the “pH” values of substances (National Ministry of Education, 2024, p. 213).

Appendix 2: Alternative conceptions revealed by the previous studies focused on acids and bases

Alternative conceptions reported related to the acids and bases topic	The study reported the alternative conception
✓ Acids and bases are dangerous, so they should not be touched	Ross and Munby (1991)
✓ Acids are always harmful; bases are not harmful
✓ Everything that tastes sour is acidic
✓ All acids are harmful
Cleaners such as soap and detergents are only basic	Nahadi et al. (2022)
✓ pH value is just a number	Mubarokah et al. (2018)
✓ Acids and bases are opposite substances; they have no common properties
All acids are strong, and all bases are weak
Bases are harmless	Bučková and Prokša (2021)
✓ pH is a linear scale	Sheppard (2006)
✓ pH value is just a number
Neutralization always occurs at pH 7	Schmidt (1995)

References

1. Abell S. K., Rogers M. A. P., Hanuscin D. L., Lee M. H. and Gagnon M. J., (2009), Preparing the next generation of science teacher educators: a model for developing PCK for teaching science teachers, J. Sci. Teacher Educ., 20(1), 77–93.
2. Akin F. N. and Uzuntiryaki-Kondakci E., (2018), The nature of the interplay among components of pedagogical content knowledge in reaction rate and chemical equilibrium topics of novice and experienced chemistry teachers, Chem. Educ. Res. Pract., 19(1), 80–105.
3. AlBadarin Y., Tukiainen M., Saqr M. and Pope N., (2023), A Systematic Literature Review of Empirical Research on ChatGPT in Education, Discover Educ., 3(1), 60 DOI:10.1007/s44217-024-00138-2.
4. Angeli C. and Valanides N., (2009), Epistemological and methodological issues for the conceptualization, development, and assessment of ICT–TPCK: advances in technological pedagogical content knowledge (TPCK), Comput. Educ., 52(1), 154–168.
5. Aydin S., Demirdogen B., Tarkin A., Kutucu S., Ekiz B., Akin F. N., Tuysuz M. and Uzuntiryaki E., (2013), Providing a set of research-based practices to support preservice teachers’ long-term professional development as learners of science teaching, Sci. Educ., 97(6), 903–935.
6. Aydin-Gunbatar S., Ekiz-Kiran B. and Oztay E. S., (2020), Pre-service chemistry teachers’ pedagogical content knowledge for integrated STEM development with LESMeR model, Chem. Educ. Res. Pract., 21(4), 1063–1082.
7. Baser D., Akkus R., Akayoglu S., Top E. and Gurer M. D., (2021), Training in-service teachers through individualized technology-related mentorship, Educ. Technol. Res. Dev., 69, 3131–3151.
8. Bučková A. and Prokša M., (2021), The persistence of primary school students’ initial ideas about acids and bases in the mental models of adults, Chem. Educ. Res. Pract., 22(1), 164–174.
9. Cao S. and Zhong L., (2023), Exploring the effectiveness of ChatGPT-based feedback compared with teacher feedback and self-feedback: evidence from Chinese to English translation, arXiv, preprint, arXiv:2309.01645 DOI:10.48550/arXiv.2309.01645.
10. Carlson J. and Daehler K. R., (2019), The refined consensus model of pedagogical content knowledge in science education, in Hume A., Cooper R. and Borowski A. A., (ed.) Repositioning pedagogical content knowledge in teachers’ knowledge for teaching science, Springer, pp. 77–92.
11. Celik I., (2023), Towards intelligent-TPACK: an empirical study on teachers’ professional knowledge to ethically integrate artificial intelligence (AI)-based tools into education, Comput. Human Behav., 138, 107468 DOI:10.1016/j.chb.2022.107468.
12. Cetin-Dindar A. and Geban O., (2017), Conceptual understanding of acids and bases concepts and motivation to learn chemistry, J. Educ. Res., 110(1), 85–97.
13. Chai C. S., Koh J. H. L. and Tsai C. C., (2013), A review of technological pedagogical content knowledge, Educ. Technol. Soc., 16(2), 31–51.
14. Clark T. M., Fhaner M., Stoltzfus M. and Queen M. S., (2024), Using ChatGPT to Support Lesson Planning for the Historical Experiments of Thomson, Millikan, and Rutherford, J. Chem. Educ., 101(5), 1992–1999 DOI:10.1021/acs.jchemed.4c00200.
15. Corp A. and Revelle C., (2023), ChatGPT is here to stay: using ChatGPT with student teachers for lesson planning, Texas Forum Teacher Educ., 14, 116–124.
16. Cresswell J., (2013), Qualitative inquiry & research design: Choosing among five approaches, Sage.
17. Dodge J., Prewitt T., Tachet Des Combes R., Odmark E., Schwartz R., Strubell E., Luccioni A. S., Smith N. A., DeCario N. and Buchanan W., (2022), Measuring the carbon intensity of AI in cloud instances, Proceedings of the 2022 ACM Conference on Fairness, Accountability, and Transparency (FAccT '22), Association for Computing Machinery, pp. 1573–1583 DOI:10.1145/3531146.3533234.
18. Drechsler M. and Van Driel J., (2008), Experienced teachers’ pedagogical content knowledge of teaching acid–base chemistry, Res. Sci. Educ., 38, 611–631.
19. Edwards C., Edwards A., Spence P. R. and Lin X., (2018), I, teacher: using artificial intelligence (AI) and social robots in communication and instruction, Commun. Educ., 67(4), 473–480 DOI:10.1080/03634523.2018.1502459.
20. Ekiz-Kiran B., Boz Y. and Oztay E. S., (2021), Development of pre-service teachers’ pedagogical content knowledge through a PCK-based school experience course, Chem. Educ. Res. Pract., 22(2), 415–430.
21. Faikhamta C., (2013), The development of in-service science teachers’ understandings of and orientations to teaching the nature of science within a PCK-based NOS course, Res. Sci. Educ., 43, 847–869.
22. Feldman-Maggor Y., Blonder R. and Alexandron G., (2025), Perspectives of generative AI in chemistry education within the TPACK framework, J. Sci. Educ. Technol., 34(1), 1–12 DOI:10.1007/s10956-024-10147-3.
23. Forsler A., Nilsson P. and Walan S., (2024), Collective pedagogical content knowledge for teaching sustainable development, Int. J. Sci. Math. Educ., 22(6), 1197–1214.
24. Gess-Newsome J., (1999), Pedagogical content knowledge: an introduction and orientation, Examining pedagogical content knowledge: The construct and its implications for science education, Springer, pp. 3–17.
25. Goodman J., Handa V., Wilson R. E. and Bradbury L. U., (2024), Promises and pitfalls: using an AI chatbot as a tool in 5E lesson planning, Innovations in Science Teacher Education, 9(1), 1–13, Retrieved from https://innovations.theaste.org/promises-and-pitfalls-using-an-ai-chatbot-as-a-tool-in-5e-lesson-planning/.
26. Grossman P., (1990), The Making of a Teacher, Teacher's College Press.
27. Guner H. and Er E., (2025), AI in the classroom: exploring students’ interaction with ChatGPT in programming learning, Educ. Inf. Technol., 30, 12681–12707 DOI:10.1007/s10639-025-13337-7.
28. Hanuscin D. L., Lee M. H. and Akerson V. L., (2011), Elementary teachers' pedagogical content knowledge for teaching the nature of science, Sci. Educ., 95(1), 145–167.
29. Hasriadi H. and Nurul N., (2023), Analysis of Pedagogic Content Knowledge Technology (TPCK) Capabilities of Teachers in Senior High School. Didaktika: Jurnal Kependidikan, 12(4), 1029–1040.
30. Ivanova A. A., (2025), How to evaluate the cognitive abilities of LLMs, Nat. Human Behav., 1–4 DOI:10.1038/s41562-024-02096-z.
31. Jong O. D., Van Driel J. H. and Verloop N., (2005), Preservice teachers' pedagogical content knowledge of using particle models in teaching chemistry, J. Res. Sci. Teach., 42(8), 947–964.
32. Karataş F. and Aksu Ataç B., (2024), When TPACK meets artificial intelligence: analyzing TPACK and AI-TPACK components through structural equation modelling, Educ. Inf. Technol., 1–26.
33. Koedinger K. R. and Aleven V., (2007), Exploring the assistance dilemma in experiments with cognitive tutors, Educ. Psychol. Rev., 19(3), 239–264 DOI:10.1007/s10648-007-9049-0.
34. Koehler M. and Mishra P., (2009), What is technological pedagogical content knowledge (TPACK)? Contemp. Issues Technol. Teacher Educ., 9(1), 60–70.
35. Koh J. H. L., (2020), Three approaches for supporting faculty technological pedagogical content knowledge (TPACK) creation through instructional consultation, Br. J. Educ. Technol., 51(6), 2529–2543.
36. Landis J. R. and Koch G. G., (1977), The measurement of observer agreement for categorical data, Biometrics, 33(1), 159–174 DOI:10.2307/2529310.
37. Lee I., Park H. and Yoon H. D., (2022), Science teachers’ theory-based teaching: connecting a learning cycle model to a lesson plan, J. Baltic Sci. Educ., 21(3), 462–480 DOI:10.33225/jbse/22.21.462.
38. Lee G. G. and Zhai X., (2024), Using ChatGPT for Science Learning: A Study on Pre-service Teachers' Lesson Planning, IEEE Trans. Learn. Technol., 17, 1643–1660.
39. Li B., Bonk C. J. and Kou X., (2023), Embracing the disrupted language teaching and learning field: analyzing YouTube content creation related to ChatGPT, Languages, 8(3), 197 DOI:10.3390/languages8030197.
40. Li S., Liu Y. and Su Y.-S., (2022), Differential Analysis of Teachers’ Technological Pedagogical Content Knowledge (TPACK) Abilities According to Teaching Stages and Educational Levels, Sustainability, 14(12), 12 DOI:10.3390/su14127176.
41. Lorenz U. and Romeike R., (2023), What is AI-PACK? –Outline of AI competencies for teaching with DPACK, International Conference on Informatics in Schools: Situation, Evolution, and Perspectives, Springer Nature, pp. 13–25.
42. Loughran J., Berry A. and Mulhall P., (2012), Understanding and developing science teachers’ pedagogical content knowledge, Springer Science & Business Media, vol. 12.
43. Merriam S. B., (1998), Qualitative Research and Case Study Applications in Education, Sage.
44. Mishra P. and Koehler M. J., (2006), Technological pedagogical content knowledge: a framework for teacher knowledge, Teachers College Rec., 108(6), 1017–1054 DOI:10.1111/j.1467-9620.2006.00684.x.
45. Mishra P. and Koehler M. J., (2008), Introducing technological pedagogical content knowledge, annual meeting of the American Educational Research Association, vol. 1, p. 16.
46. Mishra P., Koehler M. J. and Henriksen D., (2010), The 7 transdisciplinary habits of mind: extending the TPACK framework towards 21st century learning, Educ. Technol., 51(2), 22–28, https://www.jstor.org/stable/44429913.
47. Mittelstadt B. D., Allo P., Taddeo M., Wachter S. and Floridi L., (2016), The ethics of algorithms: mapping the debate, Big Data Soc., 3(2), 1–21 DOI:10.1177/2053951716679679.
48. Mubarokah F. D., Mulyanı S. and Indrıyantı N. Y., (2018), Identifying students’ misconceptions of acid-base concepts using a three-tier diagnostic test: a case of Indonesia and Thailand, J. Turkish Sci. Educ., 15, 51–58.
49. Nahadi N., Siswaningsih W., Firman H., Dewi E. P., Lestari T. and Rahmawati T., (2022), Development and application Of a two-tier acid-base misconception diagnostic test based on pictorial to identifying student misconceptions in chemistry, J. Eng. Sci. Technol., 207–223.
50. Nakhleh M. B., (1990), A study of students' thought processes and understanding of acid/base concepts during the performance of instrument-based titrations, College Park: University of Maryland.
51. Nakhleh M. B., (1994), Student's models of matter in the context of acid-base chemistry, J. Chem. Educ., 71(6), 495.
52. Ministry of National Education, (2024), Turkiye's Century Education Model, Science Curriculum(3rd, 4th, 5th, 6th, 7th, and 8th Grades).
53. Ning Y., Zhang C., Xu B., Zhou Y. and Wijaya T. T., (2024), Teachers’ AI-TPACK: exploring the relationship between knowledge elements, Sustainability, 16(3), 978.
54. Park S., (2005), A study of PCK of science teachers for gifted secondary students going through the national board certification process, Unpublished Dissertation, University of Georgia.
55. Patton M. Q., (2002), Qualitative Research & Evaluation Methods, Sage.
56. Powell W. and Courchesne S., (2024), Opportunities and risks involved in using ChatGPT to create first grade science lesson plans. PLoS One, 19(6): e0305337 DOI:10.1371/0305337.
57. Ross B. and Munby H., (1991), Concept mapping and misconceptions: a study of high-school students’ understandings of acids and bases, Int. J. Sci. Educ., 13(1), 11–23.
58. Schmidt H.-J., (1995), Applying the concept of conjugation to the Brønsted theory of acid-base reactions by senior high school students from Germany, Int. J. Sci. Educ., 17, 733–741.
59. Scholl A. and Kiesler N., (2024), How Novice Programmers Use and Experience ChatGPT when Solving Programming Exercises in an Introductory Course, arXiv, preprint, arXiv:2407.20792, 10.48550/arXiv.2407.20792.
60. Sesen B. A. and Tarhan L., (2011), Active-learning versus teacher-centered instruction for learning acids and bases, Res. Sci. Technol. Educ., 29(2), 205–226.
61. Sheppard K., (2006), High school students’ understanding of titrations and related acid-base phenomena, Chem. Educ. Res. Pract., 7(1), 32–45.
62. Shulman L., (1986), Those who understand: knowledge growth in teaching, Educ. Res., 15(2), 4–14 DOI:10.3102/0013189X015002004.
63. Shulman L., (1987), Knowledge and teaching: foundations of the new reform, Harvard Educ. Rev., 57(1), 1–23.
64. Stojanov A., Liu Q. and Koh J. H. L., (2024), University students’ self-reported reliance on ChatGPT for learning: a latent profile analysis, Comput. Educ.: Artif. Intell., 6, 100243.
65. The Institute for Ethical AI in Education, (2022), The Ethical Framework for AI in Education, The-Institute-for-Ethical-AI-in-Education-The-Ethical-Framework-for-AI-in-Education.pdf.
66. Tondeur J., Scherer R., Siddiq F. and Baran E., (2020), Enhancing pre-service teachers’ technological pedagogical content knowledge (TPACK): a mixed-method study, Educ. Technol. Res. Dev., 68(1), 319–343.
67. Tyson J., (2023), Shortcomings of ChatGPT. J. Chem. Educ., 100(8),3098–3101 DOI:10.1021/acs.jchemed.3c00361.
68. Van den Berg G., (2024), Generative AI and Educators: Partnering in Using Open Digital Content for Transforming Education, Open Praxis, 16(2), 130–141 DOI:10.55982/openpraxis.16.2.640.
69. van den Berg G. and du Plessis E., (2023), ChatGPT and generative AI: possibilities for its contribution to lesson planning, critical thinking and openness in teacher education, Educ. Sci., 13(10), 998.
70. Vorsah R. E. and Oppong F., (2024), Leveraging AI to enhance active learning strategies in science classrooms: implications for teacher professional development, World J. Adv. Res. Rev., 24(02), 1355–1370 DOI:10.30574/wjarr.2024.24.2.3499.
71. Wang X., Liu Q., Pang H., Tan S. C., Lei J., Wallace M. P. and Li L., (2023), What matters in AI-supported learning: a study of human-AI interactions in language learning using cluster analysis and epistemic network analysis, Comput. Educ., 194, 1–17 DOI:10.1016/j.compedu.2022.104703.
72. Wijaya T. T., Yu Q., Cao Y., He Y. and Leung F. K. S., (2024), Latent Profile Analysis of AI Literacy and Trust in Mathematics Teachers and Their Relations with AI Dependency and 21st-Century Skills, Behav. Sci., 14(11), 1008 DOI:10.3390/bs14111008.
73. Xu L., (2020), The Dilemma and countermeasures of AI in educational application, 2020 4th international conference on computer science and artificial intelligence, pp. 289–294.
74. Yerdelen-Damar S., Boz Y. and Aydın-Günbatar S., (2017), Mediated effects of technology competencies and experiences on relations among attitudes towards technology use, technology ownership, and self efficacy about technological pedagogical content knowledge, J. Sci. Educ. Technol., 26, 394–405.
75. Zhai X., (2023), ChatGPT for next generation science learning, XRDS: Crossroads, ACM Mag. Stud., 29(3), 42–46 DOI:10.1145/3589649.

---