Promoting students' ability to engage in scientific argumentation is a central goal of teaching in science (e.g., NGSS, 2013) and language domains (Keller et al., 2020). Argumentative skills encompass the ability to construct, evaluate, and engage with arguments from others (Clark & Sampson, 2008). These skills enable students to participate meaningfully as informed citizens in societal discussions, particularly concerning socio-scientific issues (SSIs) (Osborne & Pimentel, 2022). Effectively teaching argumentation skills requires high-quality instruction that helps students manage the substantial cognitive and motivational demands of argumentative discourse (Keller et al., 2020, 2024; Klaver et al., 2023; Osborne et al., 2013).
Learning Progressions (LPs) serve as valuable “road maps” describing the successively more sophisticated ways of thinking about argumentation (Smith, et al., 2006). They outline the lower and upper anchors of skill development and allow students' routes between them to be highly variable and rarely linear (Alonzo & Gotwals, 2012). LPs are useful for informing instruction (Berland & McNeill, 2010; Kuhn & Udell, 2003; Osborne et al., 2016), and the effectiveness of LP-based instruction has been empirically validated (e.g., Friedrichsen et al., 2016; Smit et al., 2025). However, implementing LP-based instruction in real classrooms is resource-intensive, which means LP-based instruction is rarely implemented in science classrooms (Graham et al., 2021; Newton et al., 1999). Technology offers promising avenues to address these implementation challenges (Bächtold et al., 2023; Chen et al., 2024; Friedrichsen et al., 2016; Mikeska et al., 2025). Recent advancements, particularly in artificial intelligence (AI), have enabled AI to deliver interventions that have traditionally been part of LP-based instruction. This includes providing personalized feedback, acting as an argumentation partner, or simulating dialogues (e.g., Banihashem et al., 2024; Lin & Hung, 2025; Wambsganss et al., 2021).
However, concrete examples demonstrating to educators how AI's diverse capabilities can support students' advancement along established learning progressions in argumentation are missing. This article demonstrates how tasks leveraging recognized AI capabilities can be purposefully designed to support distinct stages of students' learning progression in argumentation, aiming to bridge the gap between subject matter teaching and technological advancement. We offer a free online course and AI tutor (https://chatgpt.com/g/g-6856795be5788191872fcf33b1cbebf2-argumentation-trainer) related to this article to improve its usefulness in the classroom. The course offers practice tasks, explanatory videos for secondary students, and AI prompts for the practice task. It is available as an open educational resource under https://open.hpi.de/courses/chatgpt2025 .
Argumentation has been researched intensively, both in rather overarching areas like dialectics (Eemeren et al., 2015) and, more specifically, within science education (for an overview, see Clark & Sampson, 2008; Rönnebeck et al., 2016). Despite some ambiguities in terminology (Rönnebeck et al., 2016; Ryu & Sandoval, 2012), a central aspect is the importance given to the use of evidence, that is, the need to justify different kinds of claims with data (Clark & Sampson, 2007). Accordingly, scientific argumentation is often evaluated with regard to structural elements (e.g., the justification of claims with data and the use of warrants to connect them; Osborne et al., 2016). Other approaches (additionally or solely) focus on the conceptual quality of argumentation, i.e., whether an argument is substantive and knowledge of the content is required for its understanding (Clark & Sampson, 2007; Kelly et al., 1998). Here, the accuracy and adequacy of the argument's content serve as criteria for evaluating the argument's conceptual quality (Heitmann et al., 2014). In addition, complex problems often require consideration of multiple perspectives, and arguments must be managed through specific decision-making and opinion-forming strategies (Eggert & Bögeholz, 2010).
“Taken as a whole, a competency for scientific argumentation demands a complex orchestration of construction and critique of claims, warrants, and evidence in situations that require scientific knowledge to resolve” (Osborne et al., 2016, p. 826). Complex problems are often chosen in the form of so-called socio-scientific issues that are directly linked to social contexts, show relevance for learners and curricula, and stimulate them to critical reflection and argumentation (Zeidler et al., 2019). In consequence, scientific argumentation happens at the interface between understanding science (with aspects of the nature of science and scientific reasoning) and social relevance (by means of socio-scientific issues and benefits of context-based learning) and, thus, reflects many goals and demands of regular science courses at the task level. In addition, language demands are high as students need to process (often mainly written) materials and background information as well as engage in argumentative writing. These processes depend on students' capabilities in the language these materials are presented in (Heitmann et al., 2014).
These complex demands in science and language domains come at a cost, as empirical findings consistently show that students from secondary school (e.g., Klaver et al., 2023; Schaller et al., 2024a) to university (e.g., Noroozi et al., 2023) have difficulty providing high-quality arguments; see Sadler, 2004; Noroozi et al., 2012 for reviews). Arguments often contain unjustified claims and lack a rational reasoning component (Kelly et al., 1998) while being largely intuitive and emotional (Dawson & Venville, 2009). Given these pervasive challenges, how can educators effectively develop students' argumentation skills?
Understanding students' developmental trajectory is essential to effectively guiding them in developing argumentation skills. For socio-scientific argumentation, student progression can be conceptualized and fostered across three key stages, as delineated by Berland and McNeill (2010): the instructional context, drafting of an argumentative product, and the argumentative process. At the instructional context stage, the key learning goals are ensuring that students understand the argumentative problem, its value to themselves, the possible standpoints in the argumentation, and the usable criteria for evaluating these standpoints. As a result of these steps, students are able to decide based on explicit criteria which standpoint they will take in the following argumentation. Interventions designed to foster advancement at this stage include building students' connection to the argumentative problem (e.g., Smit et al., 2025), providing explicit explanations of the data (e.g., Friedrichsen et al., 2016), or offering representations to aid data interpretation (e.g., Hsu et al., 2015).
Drafting an argumentative product involves constructing a high-quality argumentation, including strong arguments and an effective argumentative structure. An effective structure includes finding an introduction that establishes the topic with a clear thesis statement, and writing a conclusion that summarizes arguments and provides appropriate closure, for instance, through calls to action, advice, predictions, or compromises (Peltzer et al., 2024). Osborne and colleagues (2016) describe how students learn to articulate claims, evidence, reasoning (warrants), and rebuttals, progressing from a competence level in which students might only identify or critique isolated claims or evidence, to a level where they can construct explicit warrants linking claims and data. Interventions to support students in building an argumentative product mainly involve feedback on specific elements of their arguments (Peltzer et al., 2024).
The argumentative process emphasizes argumentation's social and dialogic aspects, which are critical for responding to and negotiating potentially diverging opinions in discussions (e.g., Kuhn & Moore, 2015; Larrain et al., 2019). Engaging in dialogic practice, where students question, critique, and build upon each other's ideas, is central for developing sophisticated argumentative discussions (e.g., Bouwer & van der Veen, 2024; Kuhn et al., 1997; Tavsanli et al., 2024; see Bächtold et al., 2023 for an overview). The goal is to move beyond simply stating one's own well-constructed argumentation and toward a dynamic process of listening, reflection, and adaptation in response to the arguments of others, which is necessary in societal discussions.
These learning progressions, from understanding the instructional context to building arguments and refining the argumentation through dialogue, have guided multiple effective programs, as exemplified by the work of Smit et al. (2025) and Friedrichsen et al. (2016). While these programs have demonstrated effective pathways for developing argumentation skills, their systematic implementation often demands considerable time and resources from educators. The detailed planning, individualized student support, and iterative feedback cycles inherent in fostering deep argumentative competencies across the instructional context, product, and process dimensions can be challenging to sustain in many classroom settings.
Artificial Intelligence (AI) presents compelling opportunities to make pedagogical approaches more accessible and scalable, yet there are also challenges regarding its implementation in practice to support student learning rather than replacing it (e.g., by cognitive offloading, Skulmowski, 2024). Emerging research and applications increasingly highlight AI's potential to deliver or enhance many of the critical interventions needed to support student progression in argumentation—acting, for instance, as an intelligent tutor to clarify content, a sophisticated feedback provider to refine arguments, or even a dynamic argumentation partner to foster critical engagement and reflection.
AI as a tutor can shape, clarify, and adapt the instructional context for argumentation (Berland & McNeill, 2010), tailoring the learning environment and the presentation of tasks (Wambsganss et al., 2021). Learning with AI tutors has been shown to enhance critical thinking (Romero Ariza et al., 2024), a vital skill for understanding the demands of an argumentative task within a given context. Furthermore, AI tutors helped clarify the structural expectations of arguments and enabled students to reflect on and improve their work according to task requirements, for example, by visualizing argumentation schemes (Hoffmann & Lingle, 2015; Nussbaum et al., 2007).
Beyond establishing a supportive instructional context, AI excels as a feedback provider, a role crucial for helping students develop the quality and constituent components of their argumentative product (Berland & McNeill, 2010; Osborne et al., 2016). AI can offer timely and personalized critiques on the claims, evidence, warrants, and rebuttals students construct. It can effectively scaffold the creation of these argumentative components (Lin & Hung, 2025), guiding students towards more coherent and well-supported written or verbal arguments. This feedback can be delivered through different methods; AI might act as a peer (Banihashem et al., 2024) or simulate the feedback a teacher might give, for example, on the argument's structure and content (Meyer et al., 2024; Ruwe & Mayweg-Paus, 2022). Furthermore, AI systems can assist students in refining their argumentative product by assessing the relevance and weighting of arguments concerning a specific topic (Huang et al., 2015), prompting students to select stronger evidence or justify their reasoning more explicitly, thus enhancing the overall quality of their product.
To support the dynamic and interactive argumentative process, AI can transcend static instruction and feedback to act as a persona that is engaged in an argumentation (e.g., Guo et al., 2023). The effect of such practice is that it amplified student reflections during or after an argumentative exchange (Naik et al., 2025), a practice shown to be supportive of argumentation development (Bächtold et al., 2023). Another effect, when used as an argumentative chatbot, allows students to practice responding to differing viewpoints, critiquing opposing claims, and refining their positions in a simulated but responsive discursive environment, thereby actively participating in the argumentative process.
However, while the potential benefits of AI in these roles are offering tailored support for the instructional context, argumentative product, and argumentative process, it is also crucial to acknowledge potential drawbacks. The integration of AI into learning processes is not without its challenges. For example, there are concerns that over-reliance on AI tools could inadvertently reduce human agency and diminish deep engagement with the learning content itself (Darvishi et al., 2024). This underscores the need for careful implementation and a balanced perspective, ensuring AI serves as a tool to augment, not replace, critical human cognitive functions and active student participation in all dimensions of argumentation.
This paper aims to provide a framework for integrating AI into argumentative writing instruction, assisting educators in implementing LP-based methods that improve students' argumentative skills. At each stage of the LP, we propose a function for AI support. Throughout the course, AI offers assistance as a tutor, feedback provider, and argumentation partner. Educators can use the units as they are or adapt the proposed course structure to different topics or students' needs.
The initial stage of the LP, the instructional context, leverages AI to create a supportive and tailored learning environment. The progression begins with students building personal connections to the topic, a process AI can facilitate by helping brainstorm relevant scenarios (Smit et al., 2025). To ensure students grasp the debatable nature of the issue, an AI tutor can present multiple plausible standpoints (Friedrichsen et al., 2016; Wambsganss et al., 2021). As students explore background materials, the AI acts as an on-demand resource, providing simplified explanations for technical terms (Friedrichsen et al., 2016; Smit et al., 2025). This stage culminates in students analyzing and weighing criteria to form a standpoint, where AI can present comparative data through interactive graphics and help students assess how their personal ranking of criteria aligns with a particular solution (Hsu et al., 2015).
The second stage focuses on the argumentative product, where AI functions as a writing coach and feedback provider. After students plan the structure of their argument (Peltzer et al., 2024), AI provides guidance on the essential components of a compelling introduction. A key AI intervention is offering feedback on the clarity and argumentative power of a student's claim (Lin & Hung, 2025) and providing templates to map the claim-evidence structure (Hoffmann & Lingle, 2015; Nussbaum et al., 2007). As students write their warrants to connect evidence to claims (Osborne et al., 2016), AI can simulate a reader's perspective to test for clarity. For developing rebuttals, AI can generate relevant counterarguments for students to address or provide holistic feedback on a paragraph that integrates a refutation (Osborne et al., 2016; Meyer et al., 2024). Finally, AI can assist in crafting a strong conclusion and provide a digital rubric for self-assessment of the complete essay (Peltzer et al., 2024).
The final stage, the argumentative process, uses AI to facilitate social and dialogic practice. Here, AI can model constructive comments for peer-review activities and even check peer feedback for quality (Banihashem et al., 2024). To move beyond static text, AI can engage students as a chatbot or dialogue partner, playing the role of a “concerned citizen” or “expert” to challenge students' arguments and prompt them to defend and adapt their positions in a responsive environment (Guo et al., 2023; Kuhn & Moore, 2015). In a final synthesis step, students can even discuss their emerging group conclusions with an AI to reflect on their collaborative process (Naik et al., 2025; Tavsanli et al., 2024).
The following Table 1 offers a detailed breakdown of each step in this learning progression, aligning the specific student activities, AI support functions, and intended learning outcomes.
| No | Step in LP | Activity | AI Support | Learning Outcome: Students can... |
|---|---|---|---|---|
| Instructional context | ||||
| 1 | Engaging with the problem | Write down personal reasons why the skill of argumentation is important for the future (Smit et al., 2025). | Brainstorm additional reasons for the importance of argumentation. | ... explain the personal relevance and value of argumentation skills. |
| 2 | Understanding the problem & potential solutions | Identify debatable questions that have multiple possible standpoints (Friedrichsen et al., 2016; Smit et al., 2025). | Tutor shows plausible standpoints (Wambsganss et al., 2021). | ...identify and formulate debatable questions relevant to the topic. |
| 3 | Data exploration for each potential solution | Explore background information for an overview of standpoints' pros/cons (Friedrichsen et al., 2016; Smit et al., 2025). | Provides explanations for the information and technical terms. | ...access, interpret, and assess the relevance of initial information about the argumentation topic. |
| 4 | Analyzing criteria & limitations | Identify evaluation criteria from the provided background information; discuss their importance (Smit et al., 2025). | Helps categorize information into criteria; presents comparative data using interactive graphics (Hsu et al., 2015). | ...identify and understand different criteria for evaluating. |
| 5 | Weighing criteria & forming a standpoint | Create a personal ranking of the most important criteria for making the final decision. | Helps students compare/contrast criteria. | ...select and justify the set of key criteria for evaluating the options for action. Decide own standpoint based on criteria. |
| Drafting an Argumentation product | ||||
| 6 | Structure of the argumentation | Plan the three main parts of the argument: introduction, main body, and conclusion. (Peltzer et al, 2024). | Checks if students' plans are in the correct part (i.e., all arguments in the main part, standpoint in the introduction). | …recognize and plan the structure of the argumentation. |
| 7 | Writing an Introduction | writing introductions that include: topic, thesis statement, transition (Peltzer et al, 2024). | Guidance on introduction components; provides sample transitions; offers feedback on drafts. | … write compelling introductions that clearly establish their position on energy choices and preview their main arguments. |
| 8 | Formulating a claim | Formulate a clear, arguable claim for the chosen option, citing one key criterion (Smit et al., 2025). | Feedback on claim clarity, arguability, and link to criterion (Lin & Hung, 2025). | ...formulate clear, arguable claims about the prioritization of the option for action, based on criteria. |
| 9 | Finding evidence | Select strong evidence from background information for the claim & criterion; map out claim-evidence structure (Nussbaum et al., 2007). | Helps map argument components (claim, evidence) using a template (Hoffmann & Lingle, 2015; Nussbaum et al., 2007). | ...select strong evidence and structure it effectively with their claim. |
| 10 | Connecting claim and evidence | Write paragraphs explaining why a criterion is relevant & how selected evidence supports the claim via that criterion (Osborne et al., 2016). | Feedback on clarity/logic of the warrant/relevance explanation (Lin & Hung, 2025). | ...write clear explanations (warrants) establishing the relevance of their chosen criteria and evidence for their claims. |
| 11 | Summarising the argumentation | Summarise arguments and position (Peltzer et al, 2024). | Helps students find strongest arguments; rephrase summaries; develop closing sentence. | …write conclusions that summarize their argument and energy choice recommendation and end with an appropriate closing sentence. |
| 12 | Avoiding logical fallacies | Identify common logical fallacies in sample arguments; review own arguments for logical gaps. | Presents examples of logical fallacies in debates about the topic and provides corrective explanations. | ...identify and learn to avoid common logical fallacies in argumentation. |
| 13 | Developing counterarguments and refutations | Draft rebuttals to specific counterarguments (AI-generated or from list); incorporate into a more developed argument paragraph (Osborne et al., 2016). | Students submit paragraph with rebuttal. AI holistic feedback on structure/components (Meyer et al., 2024). | ...construct effective rebuttals and integrate them into a coherent argumentative paragraph. |
| 14 | Writing argumentation | Write a short argumentative essay including introduction and conclusion using multiple criteria, evidence, warrants, and rebuttals (Smit et al., 2025, Lesson 6). | Provides digital rubric for self/peer assessment of the essay (Peltzer et al., 2024). | ...assemble a complete argumentation addressing, effectively using multiple criteria and argument components. |
| Argumentation process | ||||
| 15 | Peer-Review | Provide structured feedback for an argumentation written by a peer (Banihashem et al., 2024). | Checks comments for constructiveness (Banihashem et al., 2024). | ...provide useful peer feedback on written arguments, identifying areas for improvement. |
| 16 | Revising | Prompt AI feedback and use it to revise argumentation (Banihashem et al., 2024). | Provides targeted feedback on argumentative aspects. | ... implement feedback to improve their arguments. |
| 17 | Evaluating arguments | Assess and critique the structure and content of others' arguments (Kuhn & Moore, 2015). | Helps identifying strengths and weaknesses in arguments. | ... assess peers' arguments objectively and identify diverse perspectives on the topic. |
| 18 | Audience awareness | Refine arguments for a specific audience (Hubbart, 2025). | Chatbot as critical reader, suggesting improvements and clarifications. | ... adapt and strengthen arguments by addressing audience needs. |
| 19 | Final argumentation | Review and refine final argumentation. | Helps check structure, content, and counterarguments. | ... confidently present clear, well-written argumentation. |
The AI support functions described in this learning progression can be executed through carefully engineered text prompts within widely available large-context reasoning models. We adopted a structured instruction-plus-role-based prompting strategy, drawing on established techniques in the prompt-engineering literature (Sahoo et al., 2024). To map the prompts to the LP, we embedded the full description of the learning progression in the contextual block of each system prompt so that the model could condition its responses on the intended argumentative stage. After drafting the initial prompts, we refined them using an LLM-based prompt optimizer, consistent with evidence that iterative prompt revision improves model performance (Yang et al., 2024).
A custom AI model, including the prompts, is available under https://chatgpt.com/g/g-6856795be5788191872fcf33b1cbebf2-argumentation-trainer. In the supplementary material, we provide example tasks and AI prompts for each step of the following learning progression. The supplementary material provides a model prompt for every learning progression step, each constructed in accordance with current vendor guidance (Anthropic, 2024b; Meta, 2024b; Mistral, 2024b; OpenAI, 2024b). The following schema illustrates this structure so that educators can readily adapt the prompts to their own tasks and contexts.:
<role and pedagogical goal> This first component defines the specific persona of the AI (e.g., “supportive tutor,” “feedback provider,” “discussion partner”) and the immediate learning objective of the exchange. This critical step aligns the tone and function of the AI with the pedagogical intent of the activity, guiding it to act as a facilitator of learning rather than merely providing information. <role and pedagogical goal>
<students activity> This segment contains the task that students are asked to do. <students activity>
<students input> This block contains the student's specific output that the AI is meant to analyze, such as a draft sentence, a brainstormed list, or a formulated question. Grounding the interaction in the student's own work makes the AI's feedback personal, relevant, and immediately actionable. <students input>
<task> This defines the concrete action the AI should perform. For each step in the learning progression, the educators would specify the unique task for the AI, such as “Your task is to provide feedback on the clarity of the student's warrant” or “Your task is to act as a critical politician and ask a challenging question about this argument.” <task>
< relevant pedagogical context> For all prompts, this final segment should contain the full text of this article. Including the complete description of the learning progression provides the AI with the necessary pedagogical framework to understand its specific role, the learning goals of each stage, and how best to support the student's development. < relevant pedagogical context>
This paper has presented a detailed, theoretically grounded framework for integrating AI into a learning progression for secondary school argumentation. While research has established the value of both argumentation LPs for many subjects (e.g., Berland & McNeill, 2010; Osborne et al., 2016) and the potential of AI as an educational tool (e.g., Wambsganss et al., 2021), a detailed synthesis mapping specific AI functions to specific LP stages has been missing so far. Our framework fills this gap by detailing three core instructional phases—Instructional Context, Argumentative Product, and Argumentative Process—and specifying how AI can function as a tutor, feedback provider, or argumentation partner within each.
Supporting the implementation in the classroom, we offer corresponding AI prompts (see supplementary material) and an online course that includes videos and established example tasks as an Open Educational Resource (OER; Bahr et al., 2024; Jansen et al., 2024; 2025; Höft et al., 2025; Schaller et al., 2024a; b). A core principle guiding this design is that the AI support functions are always responsive in nature. In every case, the student must first produce an initial artifact (e.g., a drafted claim, a brainstormed list, or a question) before the AI intervenes. This “student-first” model is a deliberate pedagogical choice intended to prevent the premature offloading of thinking (Skulmowski, 2024). It ensures that the AI serves to scaffold and enhance the student's cognitive effort rather than replacing it.
While this paper presents a comprehensive and theoretically grounded framework for integrating AI into argumentation instruction, several limitations warrant consideration. The primary limitation is that the framework has not yet been subjected to large-scale empirical testing to measure student learning gains against a control group. The effectiveness of the framework is also dependent on the quality of rapidly-evolving commercial AI models and on the digital literacy and implementation fidelity of the educator.
Second, the LP plan, though detailed, is an illustrative model. Its successful implementation in diverse secondary school contexts will invariably require adaptation by educators. Successful implementation depends in part on teachers' technological, pedagogical, and content knowledge, as the willingness to adopt AI is shaped by their confidence and capability in using such tools (Yang et al., 2025). School conditions also matter: many classrooms face constraints related to device availability, internet stability, and the need to provide students and teachers with secure, data-protected access to AI chatbots.
Third, the prompts and AI support described here reflect current model behavior and may change as large language models evolve. Prompting techniques continue to develop, and recent work shows that strategies do not always transfer reliably across model versions (Meincke et al., 2025). Before implementation, educators should revise and update the prompts using an LLM-based prompt optimizer to ensure they remain aligned with the capabilities of the models available to them (Yang et al., 2024).
Educators can directly use the online course and argumentation trainer based on our framework in their science classrooms as is, or they can leverage the material as starting points to develop their own session plans, illustrative AI tasks, and suggested prompts for either a comprehensive unit or specific activities to incorporate into their existing curriculum. The AI learning progression can also support instruction in subjects like civic education, language learning, or mathematical argumentation because the main argumentative strategies are common across these fields. To adapt the framework, educators only need to make small adjustments, such as swapping science-based examples with scenarios relevant to the subject, replacing discipline-specific evidence sources, or framing AI prompts around questions students usually face in that area. Using the framework across multiple disciplines aims to make AI-supported teaching more accessible for teachers, helping them provide focused, manageable, and personalized argumentation support across various subjects.
In this way, the framework seeks to empower teachers, demystifying AI and equipping them with vetted strategies to make the teaching of sophisticated argumentation more manageable, individualized, and effective.
From a researcher's perspective, this framework offers a meso-level design pattern for integrating AI in complex, process-oriented skills instruction. It acts as a bridge between established learning theories regarding learning progressions on the one hand, and the technical practice of prompt engineering on the other. It demonstrates how to constrain and guide AI to fulfill specific pedagogical functions rather than merely functioning as an information-retrieval system. Future research should empirically validate this framework using a quasi-experimental design. A study could compare student outcomes (e.g., quality of written arguments as measured by established rubrics, measures of self-efficacy) in a classroom using the AI-supported LP against a control group using an equivalent traditional LP-based curriculum without AI support.
Combining the strengths of evidence-based pedagogy with the evolving capabilities of AI, researchers and educators can create more dynamic, supportive, and effective learning environments. If implemented within the appropriate contexts, it will better equip students to master the complex skills of argumentation, preparing them not only for academic success but for informed and engaged citizenship in knowledge-based societies.