n Empirische Arbeit Dieser Beitrag steht open access online unter https: / / dx.doi.org/ 10.2378/ peu2026.art12d Psychologie in Erziehung und Unterricht, 2026, 73, 153 -169 DOI 10.2378/ peu2026.art12d © Ernst Reinhardt Verlag For Whom Does AI Feedback Support Writing Self-Efficacy? The Role of Students’ Achievement Goal Orientation and Writing Skills Thorben Jansen, Hannah Pünjer, Mira Tanz, Nils-Jonathan Schaller and Lars Höft Leibniz Institute for Science and Mathematics Education, Kiel Summary: Writing self-efficacy supports the development of key competencies in secondary school. Feedback from re-reading one’s writing or feedback from artificial intelligence can enhance selfefficacy. However, effects differ between students, raising questions about for whom AI feedback supports writing self-efficacy. We thus examined the moderating role of students’ achievement goal orientations and their writing skills. In a two-group experiment (n 1 = 400/ n 2 = 405), students wrote an argumentative text and received either AI-based feedback or were prompted to create their own feedback by re-reading their text. Students with stronger (weaker) performance goals reported lower (higher) self-efficacy in the AI feedback compared to self-feedback group; performance-avoidance goals showed the strongest moderation. The relation between mastery goal orientation and AI feedback effects was positive (negative) for students with low (high) writing skills. The feedback effects ranged from d = +0.4 to -0.2 between students, highlighting the need for personalized feedback based on students’ goals and skills. Keywords: Feedback, self-efficacy, achievement goals, argumentative writing, artificial intelligence Wessen Selbstwirksamkeit in Schreibaufgaben wird durch KI-Feedback gefördert? Die Rolle von Zielorientierungen und Schreibfähigkeiten Zusammenfassung: Sekundarschüler: innen erwerben mehr Schreibkompetenzen, wenn sie überzeugt sind, Schreibaufgaben bewältigen zu können. Feedback beeinflusst diese Selbstwirksamkeitsüberzeugung und kann durch das Lesen des eigenen Textes oder durch externe Rückmeldung, z. B. von künstlicher Intelligenz, generiert werden. Feedbackeffekte variieren zwischen Schüler: innen, unklar ist jedoch, wer von welchem Feedback profitiert. Die vorliegende Studie untersucht die moderierende Rolle von Zielorientierungen und Schreibfähigkeiten. In einem Zweigruppen- Experiment (n 1 = 400, n 2 = 405) schrieben Schüler: innen einen argumentativen Text und erhielten entweder KI-basiertes Feedback oder sollten sich selbst Feedback durch das Lesen ihrer Texte generieren. Schüler: innen mit stärkeren (schwächeren) Leistungszielen berichteten nach Erhalt des KI-Feedbacks von einer geringeren (höheren) Selbstwirksamkeit im Vergleich zum Selbstfeedback; Leistungs-Vermeidungsziele zeigten die stärkste Moderation. Der Zusammenhang zwischen Lernzielorientierung und KI-Feedbackeffekten war bei geringen Schreibfähigkeiten positiv, bei hohen Schreibfähigkeiten negativ. Die Effektstärken reichten von d = +0,4 bis d = -0,2 und legen nahe, Feedback in der Schule an die individuellen Ziele und Fähigkeiten der Schüler: innen anzupassen. Schlüsselbegriffe: Feedback, Selbstwirksamkeit, Leistungsziele, argumentatives Schreiben, künstliche Intelligenz 154 Thorben Jansen, Hannah Pünjer, Mira Tanz, Nils-Jonathan Schaller, Lars Höft Writing enables individuals to participate in society (UNESCO, 2016) and is essential for success across all school subjects (Graham, Kiuhara & MacKay, 2020). Within secondary school writing education, argumentative writing is a focus area in both language and STEM (science, technology, engineering, and mathematics) education because it combines multiple of the OECD’s 21 st Century Skills (e. g., communication, collaboration, and critical thinking; Ananiadou & Claro, 2009). Students engage more in writing when they see themselves as competent (i. e., having a high writing self-efficacy; Bandura, 1997). Accordingly, self-efficacy is central to writing theories (e. g., Graham, 2018; Zimmerman & Risemberg, 1997) and strongly relates to achievement (Jansen, Meyer, Hattie & Möller, 2024 a). Consequently, it is an educational aim to support students’ views of themselves as competent writers. Instructional support can strengthen students’ writing self-efficacy by helping them produce texts that come closer to their intended goals (Zimmerman & Risemberg, 1997). Feedback can be an effective instructional support (Graham et al., 2025 a) because it provides information that helps students learn how to close the gap between their current and target performance. During writing, students can create feedback themselves, for example, by re-reading their own text (Nicol, 2021) or receiving externally generated feedback (Fleckenstein, Liebenow & Meyer, 2023; Graham, Olson, Baker & Chung, 2025 b). Artificial intelligence (AI) can serve as a source of externally generated writing feedback in classrooms at scale, but effects differ for students (Jansen et al., 2025 a; c). To provide high-quality education for all, this difference raises the question: For whom does AI feedback support writing self-efficacy? Feedback effects arise from supporting students in using their writing skills to reach their own goals during writing. Hence, students’ goals and writing skills are assumed to explain why feedback affects learners differently (Bell & Kozlowski, 2002; Graham, Hebert & Harris, 2015). These assumptions are supported by empirical Achievement goal orientation Does the feedback support my development in writing? Does the feedback help to fade my weakness in writing? Does the feedback show I am a superior writer to others? Writing skills Goals × Writing skills AI feedback Self-efficacy Do I possess the skills to implement the feedback? AI feedback (vs. self-feedback) after drafting parts of an argumentation. Higher when feedback is perceived as supporting the process toward goals and reinforcing control. Lower when feedback is perceived as indicating unmet goals and reducing control. Figure 1: Assumed moderation of feedback effects on students’ writing self-efficacy Achievement Goals and Writing Skills Moderate Motivational AI Feedback Effects 155 evidence from teacher or peer feedback contexts (Fong & Schallert, 2023). However, as feedback effects depend on the social relationship between the feedback receiver and the feedback source (Daumiller & Meyer, 2025), it is unclear whether these assumptions generalize to AI feedback. Drawing on Zimmerman and Risemberg’s social cognitive model of self-regulated writing (1997), we test whether AI-generated feedback versus self-feedback differentially supports writing self-efficacy as a function of students’ achievement goal orientations and writing skills (Figure 1). The background proceeds in three steps. First, we describe writing as a self-regulated process in which self-efficacy supports engagement in iterative writing cycles. Second, we discuss feedback - both self-generated and externally provided, including AI - as an instructional support within these cycles. Third, we introduce achievement goals and writing skills as learner characteristics that may explain for whom AI feedback supports writing self-efficacy. Background Writing and the Role of Self-Efficacy Writing is a socially situated activity intended to communicate information and meaning to a community of readers (Graham, 2018). A focus area in secondary school is argumentative writing, because argumentation competencies enable students to formulate their own claims, justify them with evidence, and integrate complex ideas coherently (Common Core State Standards Initiative, 2010). Writing competencies develop through cycles of planning, translation, and revision (Graham et al., 2015) in a triadic reciprocity among personal factors (e. g., goals, skills), behavioral processes (e. g., self-monitoring), and environmental supports (e. g., feedback). In each cycle, writers are enacting a feedback loop to monitor whether strategies work to reach their goals and adapt based on observed performance (Zimmerman & Risemberg, 1997). Students engage more actively in these cycles when they perceive themselves as capable agents who can improve through effort. Self-efficacy summarizes students’ beliefs about being such an agent in a future task, essentially asking: Can I do this? (e. g., Wigfield & Eccles, 2002). Given the importance of self-efficacy in the writing process, many studies investigate instructional methods to support students in developing their self-efficacy (see Graham et al., 2025 a, for a recent review). Generally, the literature indicates that if a student perceives an instruction as supporting their goals (i. e., has a mastery or vicarious experience, Pajares, 2003), their self-efficacy is reinforced, and they maintain the behavior; if they perceive failure, they feel the need to modify their approach, and their selfefficacy decreases. In this process, feedback helps students interpret their performance, make sense of their progress, and identify a concrete next step (e. g., Jansen et al., 2024 b). Feedback as Instructional Support to Develop Self-Efficacy Feedback is defined as information provided by an agent aiming to close the gap between a student’s actual performance and a desired target performance (Hattie & Timperley, 2007). To effectively support learning and motivation, feedback must address three core questions students ask themselves to reach their goals: “Where am I going? ”; “How am I going? ”; and “Where to next? ” (Hattie & Timperley, 2007). Feedback can influence students’ writing self-efficacy by shaping how they interpret their progress toward their goals during writing. In the writing feedback loop, students compare their current text with their task goals and internal standards and appraise the discrepancy as either manageable or difficult to overcome (Zimmerman & Risemberg, 1997). External feedback becomes relevant for self-efficacy when students internalize it as information about what they can improve through their own effort and skill in order to reach their goals. Feedback 156 Thorben Jansen, Hannah Pünjer, Mira Tanz, Nils-Jonathan Schaller, Lars Höft supports self-efficacy when it makes improvement visible, identifies a feasible next step, and allows students to experience successful revision as evidence of competence. In contrast, feedback can undermine self-efficacy when it is experienced primarily as a signal of failure or as evidence that the gap cannot be closed (Schunk & Swartz, 1993). In writing tasks, much of the feedback comes from internal self-feedback. Often described as a “conversation with oneself ” (Murray, 1982), students re-read and compare their text with the demands of the task, their intentions, or internal standards (Graham, 2018; Flower & Hayes, 1981). This internal loop already echoes the three feedback questions: assessing task alignment, monitoring progress, and identifying necessary revisions. Self-feedback is central in the writing-within-community model (Graham, 2018) and contributes to the development of self-efficacy (Pajares, 2003) because when students experience the gap as solvable given their capabilities in a revision, self-feedback can strengthen self-efficacy; when they experience the gap as unsolvable, the self-feedback leads to diminishing self-efficacy. Yet many students struggle to generate feedback that is specific enough to guide revision (Carless & Boud, 2018). Advances in natural language processing have turned AI-generated feedback into a complement to this internal process. Systems based on machine learning algorithms (Meyer, Jansen, Fleckenstein, Keller & Köller, 2020) or large language models (Meyer et al., 2024) can analyze writing in process and produce feedback about argument structure, evidence, or alignment with learning progressions. In this way, AI feedback can support self-efficacy when it provides information that helps students interpret the discrepancy between their current text and the task goal as manageable, identify a feasible next step, and experience successful revision. However, AI feedback may also undermine self-efficacy when it is interpreted primarily as an external signal of failure or as information that is too difficult to use. Empirical studies investigating selfand external feedback, show, on average, positive effects on students’ writing self-efficacy (Graham et al., 2025 a). However, the 95 % prediction interval (-0.56, 1.16) implies that feedback can support, but also undermine students’ beliefs in their writing competence (Graham et al., 2025 a). Aiming to explain this heterogeneity, contemporary feedback models emphasize the role of individual learner characteristics in shaping feedback effects (e. g., Daumiller & Meyer, 2025). Given that self-efficacy in writing develops through students’ interpretations of achievable progress toward personally relevant goals, achievement goal orientations and writing skills are particularly plausible moderators to address the question: For whom does AI feedback support writing self-efficacy? Self-efficacy, Feedback, Achievement Goals, and Students’ Skills Achievement goal orientations (AGO) describe motivational dispositions that shape perceptions of personal relevance and action toward challenges (Dweck & Leggett, 1988). Achievement goal theory distinguishes mastery goals, which prioritize developing competence, from performance goals, which prioritize demonstrating competence (Dweck & Leggett, 1988). Performance goals are further subdivided into approach and avoidance dimensions (Elliot & Church, 1997). Lastly, work-avoidance goals reflect a tendency to complete tasks with as little effort as possible (Nolen, 1988). Mastery-oriented students view feedback as diagnostic information for improvement (Park, Schmidt, Scheu & DeShon, 2007) and use it to facilitate conceptual change (Miller, Fassett & Palmer, 2021). Conversely, students strongly orientated to performance-approach goals may view corrective feedback as a threat to their demonstrated competence (Park et al., 2007), while students with a performance-avoidance orientation likely interpret feedback as confirmation of inadequacy, leading to disengagement (Elliot & Church, 1997). For students Achievement Goals and Writing Skills Moderate Motivational AI Feedback Effects 157 with strong work-avoidance orientations, feedback is perceived as less useful (Keller, Dresel & Daumiller, 2024), likely because engaging with it requires additional effort. Writing skills, along with goal orientations, are also associated with feedback effects because they determine how effectively students can decode, evaluate, and apply feedback. Stronger writers generate richer self-feedback (Zhou, Yu, Liu & Jiang, 2022), view discrepancies as actionable rather than threatening, and engage more deeply with external feedback (Meyer et al., 2025). In contrast, weaker writers must invest more cognitive effort to interpret feedback and may experience evaluative comments as overwhelming or ambiguous, hindering acceptance and uptake (Winstone, Balloo & Carless, 2022). Achievement goals and writing skills interact to shape feedback processing (Bell & Kozlowski, 2002). For weaker writers, AI offers a scaffold to achieve goals they could not reach on their own, potentially boosting self-efficacy. However, this benefit risks being undermined if students attribute the success to the AI rather than their own ability (Jansen, Meyer, Fleckenstein, Wigfield & Möller, 2025 b). Conversely, stronger writers possess the ‘will and skill’ to convert feedback into mastery experiences (Kanfer & Ackerman, 1989), yet they may face the expertise reversal effect (Kalyuga, 2007). For them, external AI guidance may become redundant or intrusive, interfering with their already effective internal feedback loops. The Present Study We investigate for whom AI-generated feedback on students’ argumentative writing supports or undermines their writing self-efficacy. We test moderators of feedback effects building on achievement goal theory (Dweck & Leggett, 1988) and the assumption that performance is a function of skills and motivation (Kanfer & Ackerman, 1989). Specifically, we examine the moderating role of mastery, performance-approach, performance-avoidance, and work-avoidance goals in combination with students’ writing skills. We test four hypotheses regarding the interaction between feedback type (AI vs. self-feedback) and goal orientation (see model of change in Figure 1, above). Additionally, we analyse if this interaction is further moderated by students’ writing skills (Feedback × Performance- Approach × Skill) as an exploratory analysis, given limited prior evidence on how ability shapes the motivational meaning of externally provided corrections. Our hypotheses are: 1. Mastery goals. We expect a positive twoway interaction (Mastery Goal × Feedback), such that students higher in mastery goals would report higher post-task self-efficacy after receiving AI feedback compared to selffeedback, because they would interpret corrective information as useful feed-forward for improvement. 2. Performance-approach goals. We expect a negative two-way interaction (Performance- Approach × Feedback), such that students higher in performance-approach goals would report lower post-task self-efficacy after receiving AI feedback compared to self-feedback, because they are more likely to construe corrective AI comments as evaluative signals about competence. 3. Performance-avoidance goals. We expect a negative two-way interaction for performance-avoidance (Performance-Avoidance × Feedback), such that students higher in performance-avoidance goals would show lower post-task self-efficacy under AI feedback than under self-feedback due to heightened threat and anticipated failure. 4. Work-avoidance goals. We expect a negative two-way interaction (Work-Avoidance × Feedback), such that students higher in work-avoidance goals would report lower post-task self-efficacy after receiving AI feedback compared to self-feedback. We assume that AI feedback would make revision demands more salient and thereby increase the perceived effort required to improve the text. 158 Thorben Jansen, Hannah Pünjer, Mira Tanz, Nils-Jonathan Schaller, Lars Höft Method Transparency and Openness All data and analysis scripts are available in the Supplementary Material at https: / / osf.io/ nte2h. The hypotheses were not pre-registered. During the preparation of this work, the authors used GPT-5 and Gemini 2.5 Pro to generate ideas and improve language. Any use of generative AI in this manuscript adhered to ethical guidelines for the use and acknowledgment of generative AI in academic research (Porsdam-Mann et al., 2024). Sample The study was conducted in secondary schools in northern Germany. In total, 805 students from 47 classes in grades 10 and 11 participated. Recruitment began in September 2024 via telephone and by sending letters to the school principals. Participation was voluntary and required written parental consent. Participants were between 14 and 21 years old (M = 16.48, SD = 0.96). Of these, 339 identified as female (54.8 %), 242 as male (39.1 %), 12 as non-binary (1.9 %), and seven selected “diverse/ other” (1.1 %). 19 students (3.1 %) did not provide gender information. The majority of participants (81.3 %) reported speaking German as their primary home language. The sample comprised students from academic-track schools (74.6 %), non-academic-track schools (18.2 %), and vocational schools (7.2 %). The sample was approximately evenly distributed between conditions (Selffeedback prompt: n = 400; AI feedback: n = 405). Variables Independent Variable: Feedback Feedback was varied at two levels (AI feedback versus self-feedback). Students in the AI feedback condition received five discrete feedback messages, each targeting specific components of the text (e. g., major claim, arguments) based on a learning progression for argumentative writing (Jansen, Pünjer, Schaller, Bahr & Höft, 2025 d; Osborne et al., 2016). The feedback was structured around three guiding questions: “Where am I going? ”, “How am I going? ”, and “Where to next? ” (Hattie & Timperley, 2007). First, it made the task goal explicit by restating what a successful response should address. Second, it indicated how well the current text met this goal by identifying both included and missing or underdeveloped aspects. Third, it provided strategic guidance for revision through concrete suggestions for improvement, such as adding omitted content, strengthening comparisons, or clarifying the final recommendation. The feedback was designed to be constructive by helping students address knowledge gaps while reducing purely evaluative elements (Jansen et al., 2024 a). We operationalized the feedback using Chat GPT-4o with a prompt engineering strategy (Persona, Chain-of-Thought, and Few-Shot; Meincke, Mollick, Mollick & Shapiro, 2025) to ensure alignment with the learning progression (see Supplementary Material for the prompt and all generated feedback). At each event, the AI received only the current part of the argumentative text (e. g., one pro argument); it did not have access to prior blocks or earlier feedback, and prior AI messages were not fed back into the model. Students in the self-feedback group received prompts to re-read their text that was intended to trigger the generation of self-feedback (“Now take the time to revise your text.”; similar to Meyer, Jansen & Fleckenstein, 2024). Such prompts are a typical control group condition (e. g., Shih & Alexander, 2000; see Graham et al., 2025 a for a review showing 64 % of studies included self-feedback) and has been shown to encourage students to evaluate their own work (see Nielsen, 2014, for the conceptual framework). Aside from differences in information in the AI feedback and prompt, the timing, interface, and revision checkpoints within the groups were the same. The self-feedback group also had five total revision checkpoints, identical in timing to those of the experimental group. So any variation in results can be attributed to the content and structure of AI feedback. Dependent Variable: Self-Efficacy Students’ self-efficacy for writing was assessed with four items adapted from Pintrich, Smith, Garcia & Mckeachie (1993), following the implementations used in Keller et al. (2024). We relied on a general writing scale rather than an argument-specific scale because the general form shows comparable correlations with writing performance (Wang, Graham, Kim & Steiss, 2025) while being easier to apply across tasks. The scale measured students’ confidence to write effective German texts (e. g., “I can write a good text in German”, “I can structure a German text well”, “I can link my ideas coherently”). Responses were given on a five-point Likert scale (1 = strongly disagree, 5 = strongly agree). We measured students’ baseline self-efficacy for writing at the beginning of the study and again as Achievement Goals and Writing Skills Moderate Motivational AI Feedback Effects 159 a dependent variable after the intervention. Reliability was good for the first measurement and excellent for the second ( α = .86 and α = .92, respectively). Moderator: Achievement Goals Students’ achievement goals were measured using twelve self-report items adapted from Daumiller, Dickhäuser & Dresel (2019). The items assessed four goal orientations relevant to how students respond to AI feedback. All items referred to the current writing task and were rated on an eight-point Likert scale (1 = strongly disagree, 8 = strongly agree). The scales were mean-centered and entered as moderators in the structural models to test whether students’ goal orientations shaped their responses to AI-generated feedback and their post-intervention self-efficacy. Mastery goals captured students’ intentions to develop competence and learn from the task (e. g., “In today’s task, it is important to me to learn something new.”). Reliability was excellent ( α = .96). Performance-approach goals reflected the desire to demonstrate competence to others (e. g., “In today’s task, I want to be perceived as competent.”), α = .93. Performance-avoidance goals captured concerns about appearing incompetent (e. g., “In today’s task, I want to avoid being perceived as incompetent.”), α = .95. Work-avoidance goals captured students’ intentions to minimize effort and get through the task with as little work as possible (e. g., “In today’s task, it is important to me to put in as little effort as possible.”), α = .95. Moderator: Writing Skills Students wrote an argumentation text in the beginning of the study as a baseline measure using a standardized task (see Bahr, Höft, Meyer & Jansen, 2024; Jansen et al., 2024 c; 2025 d; Höft Meyer, Bernholt & Jansen, 2025). Texts were evaluated using AI algorithms trained explicitly for this writing task (Schaller, Ding, Horbach, Meyer & Jansen, 2024 b), based on over 5000 students’ texts (Schaller et al., 2024 a). The algorithms first scored the structural and content qualities of the text individually using analytical rubrics. To obtain a psychometrically sound proficiency estimate that accounts for differences in item difficulty, the component scores were integrated using a partial credit model (PCM) estimated in the R package TAM (Robitzsch, Kiefer & Wu, 2022). Individual baseline proficiency scores were then extracted as weighted likelihood estimates (WLEs; Warm, 1989) from this model and used as the moderator of writing skills in the subsequent analyses. Control Variables We conducted robustness checks, including pre-treatment covariates to adjust for individual differences unrelated to the intervention. The control variables included gender, language background, and grades. Gender was coded as a binary indicator (female vs. other) and included because of established differences in using AI feedback for writing tasks (Meyer et al., 2025). Language background was coded dichotomously to distinguish students with German as the primary language at home from those with other language backgrounds; this covariate captures potential differences based on migration background. Grades were represented by self-reported school grades in German, and Mathematics, entered on their native scale and treated as continuous predictors. Procedure We used a two-group between-subjects design with individual randomization within classes. Allocation was concealed in the interface. Each class completed the study in a single 90-minute session on the same set of tablet computers with external keyboards. Figure 2 summarizes the session flow. Before the lesson began, the research team prepared tablet computers equipped with external keyboards. After a welcome from the research team, students completed a brief questionnaire on their achievement goals and covariates. Next, they wrote a 15-minute pre-test essay (Task 1) piloted at the class level and in the context (Schaller et al., 2024 a). In the task, students were first asked to decide which energy source should be funded if diesel engines were banned (accumulator, e-fuels, or hydrogen, see Supplementary Material for the tasks). The writing performance in the task served as a measure of prior performance. In the main writing and revision task (Task 2), participants argued whether a solar farm, a wind farm, or a hydroelectric power plant should be funded to supply renewable energy to their local district. Task 2 comprised five drafting blocks in a fixed order (introduction argument 1, argument 2, argument 3, conclusion, see Figure 2) - reflecting prior studies indicating that students have difficulties incorporating all elements of a robust argumentation (Schaller et al., 2024 a). After students stated that they had finished, exactly one feedback event occurred, yielding a total of five feedback messages per student. Trained student assistants supervised all sessions. 160 Thorben Jansen, Hannah Pünjer, Mira Tanz, Nils-Jonathan Schaller, Lars Höft Statistical Analyses All analyses were conducted using R (see supplementary material for the analyses code). We estimated linear mixed-effects models with random classroom intercepts to account for the clustered structure of the data. The design effect factor was 1.64, indicating that clustering was present and should be considered in statistical inference (Luo et al., 2021). The dependent variable was students’ post-task writing self-efficacy. For each achievement goal orientation, we estimated a separate model including feedback condition (0 = self-feedback, 1 = AI feedback), baseline writing skill, baseline writing self-efficacy, the respective goal orientation, and all two-way and three-way interactions among feedback condition, writing skill, and goal orientation. Missing data were handled using multilevel multiple imputation with the R package mdmb, following a sequential modeling approach based on Bayesian estimation (Grund, Lüdtke & Robitzsch, 2021). The imputed datasets were analyzed using the same mixed-effects specification, and parameter estimates were pooled across imputations using Rubin’s rules. No cases were excluded through listwise deletion. We did not exclude data points. The model specification followed the equation: SelfEfficacy post = β 0 + β 1 (SelfEfficacy baseline ) + β 2 (WritingSkill) + β 3 (GoalOrientation) + β 4 (FeedbackCondition) + β 5 (WritingSkill × GoalOrientation) + β 6 (WritingSkill × FeedbackCondition) + β 7 (GoalOrientation×FeedbackCondition) + β 8 (WritingSkill × GoalOrientation × FeedbackCondition) + u class + e Significant interactions were probed using two visualization techniques to facilitate interpretation: We studied the Goal × Feedback interaction across three different levels of baseline writing skill: Low (-1 SD), Mean (0 SD), and High (+1 SD). For each level, we plotted the predicted self-efficacy slopes and drew 95 % confidence bands to show where the AI versus No-AI conditions differed. To give a complete overview of significant regions, we created contour heatmaps. Draft Introduction Draft Argument 1 Draft Argument 2 Draft Argument 3 Draft Conclusion AI / Self-feedback AI / Self-feedback AI / Self-feedback AI / Self-feedback AI / Self-feedback Revision Introduction Revision Argument 1 Revision Argument 2 Revision Argument 3 Revision Conclusion Questionnaire of receptivity and control variables (5 min) Pre-Test Writing (15 min) Writing and Revision Figure 2: Study procedure Achievement Goals and Writing Skills Moderate Motivational AI Feedback Effects 161 These showed the significance of the AI treatment effect across the entire continuous range of both Goal Orientation (x-axis) and Baseline Writing Skill (y-axis), pinpointing the exact combinations of motivation and skill where AI feedback had a positive or negative impact on self-efficacy. As a robustness check to ensure our findings were not artifacts of the estimation method, assumptions, or missing data handling, we replicated the analysis using multiple approaches proposed by Edelsbrunner et al. (2025). The approaches were a person-centered approach, Bayesian multilevel modeling, and additive regression models (see Supplementary Material for analysis code and results). Results Across all models, the conditional effects varied substantially (d = 0.40 to about d = -0.20; see heatmaps in the supplementary material) depending on students’ achievement goals and writing skills. The average treatment effect was zero, indicating that AI feedback did not increase or decrease students’ self-efficacy averaged across students (Bs = 0.01 to -0.01, ps ≥ .89). Figures 3, 4, 5, and 6 visualize the conditional effects. For mastery goals, there was no evidence of a two-way interaction with feedback condition (B = 0.012, p = .706), indicating that mastery orientation alone did not predict differential responses to AI feedback versus self-feedback. However, there was a significant three-way interaction among feedback condition, baseline writing skill, and mastery goals (B = -0.081, p = .046). This pattern indicates that the relation between mastery goals and feedback effects depended on students’ writing skills. More specifically, AI feedback was less favorable relative to self-feedback for students who were both higher in mastery orientation and higher in baseline writing skill, whereas students with lower writing skill showed a flatter or somewhat more favorable pattern. Robustness analyses were mixed: the covariate-adjusted model reproduced the negative three-way interaction with a similar effect size, whereas the CR2 clusterrobust analysis yielded the same coefficient but no longer reached statistical significance.Together with the visualization in Figure 3, these results, with caution, point to a conditional role of mastery goals that emerges only when accounting for students’ baseline writing skill. Figure 3: Three-way interaction of mastery approach goals, writing skills, and feedback Note: Solid lines (self-feedback) and dashed lines (AI feedback) show predicted self-efficacy, with shaded 95 % confidence intervals. Self-Efficacy 5 4 3 2 1 Low Ability (-1 SD) Mean Ability High Ability (+1 SD) 2 4 6 8 2 4 6 8 2 4 6 8 Mastery (Learning-Approach) AI Feedback Self-Feedback 162 Thorben Jansen, Hannah Pünjer, Mira Tanz, Nils-Jonathan Schaller, Lars Höft Performance-approach goals moderated the effect of AI feedback on students’ self-efficacy. Students higher in performance-approach goals reported lower self-efficacy in the AI feedback condition than in the self-feedback condition, as indicated by a significant Feedback × Performance-Approach interaction (B = -0.059, p = .034). The three-way interaction with baseline writing skill was not significant (B = -0.024, p = .515), indicating that this negative moderation effect held across skill levels. The robustness analysis replicated the two-way interaction with a comparable magnitude and again showed no evidence of skill involvement. Taken together, the results suggest that students striving to demonstrate competence responded less positively to the corrective, externally evaluative nature of AI feedback than to self-generated feedback, regardless of their initial writing ability. Performance-avoidance goals also moderated the effect of AI feedback. Students with stronger avoidance concerns reported lower self-efficacy in the AI-feedback than in the self-feedback group, as reflected in a significant two-way interaction (Performance-Avoidance × Feedback: B = -0.087, p < .001). The three-way interaction with baseline writing skill was not significant (B = -0.015, p = .673). Yet, the negative coefficient suggests that higher-skilled students with strong avoidance motives may be particularly sensitive. The robustness analysis with covariates replicated the pattern. Taken together, the results indicate that students motivated to avoid failure consistently experienced AI feedback as more threatening and less conducive to sustaining selfefficacy than self-feedback, with initial skill potentially amplifying this vulnerability. Work-avoidance goals did not moderate the effect of AI feedback on students’ self-efficacy. There was no evidence of a two-way interaction between work-avoidance goals and the feedback groups (Work-Avoidance × Feedback: B = -0.032, p = .297), indicating that students higher in work-avoidance did not respond differently to AI feedback than to self-feedback. The three-way interaction with baseline writing skill was also not significant (B = -0.008, p = .824), suggesting that this null pattern held Self-Efficacy 5 4 3 2 1 Low Ability (-1 SD) Mean Ability High Ability (+1 SD) 2 4 6 8 2 4 6 8 2 4 6 8 Performance-Approach AI Feedback Self-Feedback Figure 4: Three-way interaction of performance approach goals, writing skills, and feedback Note: Solid lines (self-feedback) and dashed lines (AI feedback) show predicted self-efficacy, with shaded 95 % confidence intervals. The dashed line marks the Johnson-Neyman point at which the effect of feedback type becomes statistically significant. Achievement Goals and Writing Skills Moderate Motivational AI Feedback Effects 163 across levels of initial writing ability. Taken together, these results indicate that, in contrast to performance-approach and performanceavoidance goals, work-avoidance goals did not explain for whom AI feedback supported or undermined writing self-efficacy. Self-Efficacy 5 4 3 2 1 Low Ability (-1 SD) Mean Ability High Ability (+1 SD) 2 4 6 8 2 4 6 8 2 4 6 8 Performance-Avoidance AI Feedback Self-Feedback Figure 5: Three-way interaction of performance avoidance goals, writing skills and feedback Note: Solid lines (self-feedback) and dashed lines (AI feedback) show predicted self-efficacy, with shaded 95 % confidence intervals. The dashed line marks the Johnson-Neyman point at which the effect of feedback type becomes statistically significant. Self-Efficacy 5 4 3 2 1 Low Ability (-1 SD) Mean Ability High Ability (+1 SD) 2 4 6 8 2 4 6 8 2 4 6 8 Work-Avoidance AI Feedback Self-Feedback Figure 6: Three-way interaction of work-avoidance goals, writing skills and feedback Note: Solid lines (self-feedback) and dashed lines (AI feedback) show predicted self-efficacy, with shaded 95 % confidence intervals. 164 Thorben Jansen, Hannah Pünjer, Mira Tanz, Nils-Jonathan Schaller, Lars Höft Discussion The present study examined for which students AI-generated feedback versus self-feedback supports secondary students’ self-efficacy in argumentative writing. The results show that some students benefited more from AI feedback, while others benefited more from self-feedback regarding their self-efficacy. The pattern could be explained by students’ achievement goals and writing skills (in line with Bell & Kozlowski, 2002; Kanfer & Ackerman, 1989; Meyer et al., 2025). Our findings add to recent accounts of feedback as an interaction between the message and the learner (e. g., Daumiller & Meyer, 2025) and to AI’s influence on students’ self-perception (Jansen et al., 2025 b). For school practice, our study points to the need to personalize feedback based on students’ goals and skills to provide the best self-efficacy support for all. Regarding our specific hypotheses, we found, contrary to expectation, no positive two-way interaction for mastery goals. Instead, mastery goals explained differences in feedback effects only in combination with writing skills. In the self-feedback condition, the relation between mastery orientation and self-efficacy is more positive the more the writing skills increase. A plausible explanation is that skilled students with a strong desire to learn were especially able to use the self-feedback prompt productively. They were motivated to re-read their text, detect discrepancies between their draft and the task goal, and create useful feedback for revision. Their writing skills likely enabled them to translate this internally generated feedback into successful revisions, which in turn strengthened self-efficacy by making progress feel controllable and attributable to their own competence. In the AI-feedback condition, this pattern was reversed: the positive relation between mastery goals and feedback effects weakened as writing skills increased. This suggests that low-skilled students with strong mastery goals particularly benefited from AI feedback because it provided the diagnostic information they were motivated to use but could not easily generate themselves. For these students, AI feedback may have made the gap between their writing and goals more understandable, highlighted a feasible path forward, enabled successful revisions, and thereby supported self-efficacy. For highly skilled students with strong mastery goals, however, externally generated feedback may have added little diagnostic value in helping them reach their goal, as they could already regulate their revision effectively on their own. In this case, AI feedback may have become redundant or even mildly disruptive by replacing part of an already well-functioning internal feedback process, consistent with expertise-reversal logic (Kalyuga, 2007). As hypothesized, performance-approach goals showed a negative two-way interaction on self-efficacy (consistent with Gjerde, Skinner & Padgett, 2022), which was stable across levels of writing skill: students with a strong (weak) performance-approach orientation reported lower (higher) self-efficacy receiving AI feedback than creating self-feedback. Students with high performance orientation aim to demonstrate superior competencies relative to others. Feedback indicating that revision is still needed may therefore signal that others perform better. AI feedback could not resolve these students’ concerns: it specified how to improve a text, but provided no signal about one’s standing relative to peers. The feedback, therefore, addressed a dimension of performance that is largely irrelevant to what these students care about most, while simultaneously highlighting shortcomings that can be read as indicators of relative inferiority and therefore lowering self-efficacy. In the self-feedback condition, by contrast, students retained greater control over which discrepancies they noticed and how they framed them. This may have reduced the evaluative weight of identified gaps, making it easier to interpret them as manageable rather than as evidence of falling short of others. When students construct their own feedback, they can frame limitations in ways that preserve a sense of competence. More broadly, this finding suggests that AI feedback is not processed as purely neutral task information. Achievement Goals and Writing Skills Moderate Motivational AI Feedback Effects 165 Supporting our hypothesis, students higher (lower) in performance-avoidance showed lower (higher) self-efficacy after AI feedback compared to self-feedback. Similar to the pattern observed for performance-approach goals, the self-feedback condition may have allowed students to regulate revision with less threat to the self. In contrast, externally provided feedback may have readily signaled the very signs of incompetence these students seek to avoid (Muis, Ranellucci, Franco & Crippen, 2013). In the AI-feedback condition, students likely processed the feedback as an external competence signal that highlighted a discrepancy between their current performance and the standard, they wished not to fall short of. Rather than interpreting this discrepancy as a manageable next step for revision, they may have experienced it as threatening evidence that incompetence had become visible. This likely reduced their sense of control over successful revision and, in turn, lowered self-efficacy. Work-avoidance goals did not moderate the effects of feedback. In the classroom setting of our study, where students completed a single writing task, the goal of minimizing effort may not have been salient enough to shape how they processed AI feedback. Unlike performanceoriented goals, work-avoidance is less directly tied to whether feedback signals competence, failure, or progress. Instead, it primarily concerns the costs of engaging further. In this context, AI feedback may therefore not have altered selfefficacy through effort-avoidance motives, but through how students interpreted the feedback as information about their competence and the manageability of revision. Implications for Theory, Empirical Research, and School Practice The results align with self-regulated writing models (e. g., Graham, 2018), confirming that feedback supports self-efficacy only when interpreted as diagnostic information supporting reaching one’s own goals rather than evaluative threat for one’s own competencies (Schunk & Schwartz, 1993). That AI can be a threat, especially for students with strong performance goals, is interesting for developing feedback theory in AI supported environments: Although AI is not a social actor in the same way as peers or teachers, the importance of performance goals in shaping the effects of feedback indicates that students did not process AI feedback merely as neutral information about text quality, but also as evaluative information with social meaning. Thus, aligning with assumptions of the situated expectancy value theory for AI supported environments (Jansen et al., 2025 b), the social effects of AI feedback should not be neglected. For the empirical literature, the results imply that studies should treat individual differences as a core design feature and follow the idea feedback cycles work conditionally, depending on the learner’s motivational-cognitive profile (Kluger & DeNisi, 1996). Many prior evaluations focus on average effects (e. g., Meyer et al., 2024), which can obscure opposing subgroup effects and lead to premature conclusions about effectiveness. For school practice, the message is that AI feedback is not a universal good. Average effects can look neutral because benefits for some students are offset by costs for others, a pattern confirmed by meta-analyses showing non-significant differences but high heterogeneity in outcomes when comparing AI and human feedback (Kaliisa, Misiejuk, López-Pernas & Saqr, 2025). This challenges the common approach to deciding whether to deploy AI writing tools at the class level. Instead, the relevant question becomes for which students, under which motivational conditions, and at which skill level does AI feedback help rather than harm. More concretely, that AI feedback was more beneficial for students with higher writing capabilities and weaker performance goal orientations suggests that teachers may want to use AI feedback preferentially in contexts where students already acquired the competencies to interpret and translate diagnostic comments into concrete revisions, and where the classroom climate minimizes social comparison (e. g., individual work phases). 166 Thorben Jansen, Hannah Pünjer, Mira Tanz, Nils-Jonathan Schaller, Lars Höft Limitations Five limitations need special consideration. The writing process was intentionally constrained to five blocks (introduction argument 1, argument 2, argument 3, conclusion) to follow students’ learning progression in argumentative writing (Jansen et al., 2025 d). It is plausible that the structured nature of the AI guidance, while beneficial for those who need structure, may have inadvertently imposed a performance ceiling for students who already possess strong self-regulatory skills, preventing them from fully leveraging their abilities. Second, our evaluation captured a single 90-minute session with immediate revision outcomes (see Meyer et al., 2024; Jansen et al., 2024 b for similar designs). There are indications that the effect of performance feedback changes with time as students reflect on it (Duijnhouwer, Prins & Stokking, 2012). We do not know whether effects persist or change after some reflection. A third limitation is that it remains unclear whether all students remained fully unaware of the differences in the conditions. We chose to conduct the study in students’ regular classroom environment to increase ecological validity and reflect authentic learning conditions; however, this means students may have noticed or inferred the alternative condition (e. g., by glancing at peers’ screens), which could have attenuated the contrast between conditions. The fourth limitation concerns the interpretation of the self-feedback condition. Although the revision prompt was intended to trigger students’ generation of self-feedback, we did not directly assess whether participants in this condition actually reread their text, generated internal feedback, or used it during revision. Accordingly, the present findings should be interpreted as comparing AI-generated feedback with a condition designed to prompt self-feedback, rather than with directly observed self-feedback itself. A final limitation is that the mastery-related three-way interaction was sensitive to the inferential specification. Although it appeared in the primary multilevel MI models, it was no longer statistically significant under CR2 cluster-robust inference. This pattern may indicate lower power under the more conservative specification, but it may also suggest that clusterlevel dependencies contributed to the original effect. The result should therefore be interpreted cautiously. Conclusion Overall, the study suggests that AI feedback should not be treated as a single, uniformly beneficial instructional tool for supporting selfefficacy. Whether it is more effective than prompting students to reread their texts depends on who receives it: the same feedback can be interpreted as evidence of progress and a clear next step, or as an evaluative signal that threatens competence. Our study thus supports feedback theories that place the learner at the center of the feedback cycle (e. g., Daumiller & Meyer, 2025) and motivation theories that position AI as a trigger of social comparison (e. g., Jansen et al., 2025 b). For school practice, we conclude that effective school implementation of feedback will require targeted, adaptive use, tailored to students’ skills and goals rather than a class-level, one-size-fits-all approach. References Ananiadou, K. & Claro, M. (2009). 21st century skills and competences for new millennium learners in OECD countries. Bahr, J. L., Höft, L., Meyer, J. & Jansen, T. (2024). Receptivity to Instructional Computer-Based Feedback (receptivity. European Journal of Psychological Assessment. https: / / doi.org/ 10.1027/ 1015-5759/ a000841 Bandura, A. (1997). The anatomy of stages of change. American Journal of Health Promotion: AJHP, 12 (1), 8. Bell, B. S. & Kozlowski, W. J. (2002). Goal orientation and ability: Interactive effects on self-efficacy, performance, and knowledge. Journal of Applied Psychology, 87 (3), 497 - 505. https: / / doi.org/ 10.1037/ 0021-9010.87.3. 497 Carless, D. & Boud, D. (2018). The development of student feedback literacy: Enabling uptake of feedback. Assessment & Evaluation in Higher Education, 43 (8), 1315 - 1325. https: / / doi.org/ 10.1080/ 02602938.2018. 1463354 Common Core State Standards Initiative (2010). Common Core State Standards Initiative. Achievement Goals and Writing Skills Moderate Motivational AI Feedback Effects 167 Daumiller, M., Dickhäuser, O. & Dresel, M. (2019). University instructors’ achievement goals for teaching. Journal of Educational Psychology, 111 (1), 131 - 148. https: / / doi.org/ 10.1037/ edu0000271 Daumiller, M. & Meyer, J. (2025). Advancing feedback research in educational psychology: Insights into feedback processes and determinants of effectiveness. Contemporary Educational Psychology, 102390. https: / / doi. org/ 10.1016/ j.cedpsych.2025.102390 Duijnhouwer, H., Prins, F. J. & Stokking, K. M. (2012). Feedback providing improvement strategies and reflection on feedback use: Effects on students’ writing motivation, process, and performance. Learning and Instruction, 22 (3), 171 - 184. https: / / doi.org/ 10.1016/ j.learn instruc.2011.10.003 Dweck, C. S. & Leggett, E. L. (1988). A social-cognitive approach to motivation and personality. Psychological Review, 95 (2), 256 - 273. https: / / doi.org/ 10.1037/ 00 33-295X.95.2.256 Edelsbrunner, P. A., Tetzlaff, L., Bach, K. M., Dumas, D., Hofer, S. I., Köhler, C., Kozlova, Z., Moeller, J., Reinhold, F., Roberts, G. J., Sengewald, M. & Bichler, S. (2025). Beyond linear regression: Statistically modeling aptitude-treatment interactions and the differential effectiveness of educational interventions. Learning and Individual Differences, 124, 102812. https: / / doi. org/ 10.1016/ j.lindif.2025.102812 Elliot, A. J. & Church, M. A. (1997). A hierarchical model of approach and avoidance achievement motivation. Journal of Personality and Social Psychology, 72 (1), 218. Fleckenstein, J., Liebenow, L. W. & Meyer, J. (2023). Automated feedback and writing: A multi-level meta-analysis of effects on students’ performance. Frontiers in Artificial Intelligence, 6, 1162454. https: / / doi.org/ 10. 3389/ frai.2023.1162454 Flower, L. & Hayes, J. R. (1981). A cognitive process theory of writing. College Composition and Communication, 32 (4), 365 - 387. https: / / doi.org/ 10.2307/ 356600 Fong, C. J. & Schallert, D. L. (2023). “Feedback to the future”: Advancing motivational and emotional perspectives in feedback research. Educational Psychologist, 1 - 16. https: / / doi.org/ 10.1080/ 00461520.2022.2134135 Gjerde, K. P., Skinner, D. & Padgett, M. Y. (2022). Importance of Goal and Feedback Orientation in Determining Feedback Effectiveness. Journal of the Scholarship of Teaching and Learning, 22 (3). https: / / doi.org/ 10. 14434/ josotl.v22i3.31866 Graham, S., Hebert, M. & Harris, K. R. (2015). Formative Assessment and Writing. The Elementary School Journal, 115 (4), 523 - 547. https: / / doi.org/ 10.1086/ 681947 Graham, S. (2018). A revised writer (s)-within-community model of writing. Educational Psychologist, 53 (4), 258 - 279. Graham, S., Kiuhara, S. A. & MacKay, M. (2020). The Effects of Writing on Learning in Science, Social Studies, and Mathematics: A Meta-Analysis. Review of Educational Research, 0034654320914744. https: / / doi.org/ 10.3102/ 0034654320914744 Graham, S., Ng, C., Hebert, M., Santangelo, T., Aitken, A. A., Camping, A. & Nusrat, A. (2025 a). Can teaching writing enhance students’ writing self-efficacy: A metaanalysis. Journal of Educational Psycholog y, 117 (4), 559 - 581. https: / / doi.org/ 10.1037/ edu0000923 Graham, S., Olson, C. B., Baker, T. & Chung, H. (2025 b). The future of writing instruction and research at the secondary level: Recommendations from the WRITE Center summit. Reading and Writing. https: / / doi.org/ 10.1007/ s11145-025-10727-8 Grund, S., Lüdtke, O. & Robitzsch, A. (2021). Multiple imputation of missing data in multilevel models with the R package mdmb: a flexible sequential modeling approach. Behavior Research Methods, 53 (6), 2631 - 2649. https: / / doi.org/ 10.3758/ s13428-020-01530-0 Hattie, J. & Timperley, H. (2007). The power of feedback. Review of Educational Research, 77 (1), 81 - 112. https: / / doi.org/ 10.3102/ 003465430298487 Höft, L., Meyer, J., Bernholt, S. & Jansen, T. (2025). A situated perspective on CONIC: Evidence of compensatory effects of conscientiousness and situational interest on the task level. Contemporary Educational Psychology, https: / / doi.org/ 10.1016/ j.cedpsych.2025.102375 Jansen, T., Meyer, J., Hattie, J. & Möller, J. (2024 a). Who am I? A second-order meta-analytic review of correlates of the self in childhood and adolescence. Psychological Bulletin, 150 (11), 1287 - 1317. https: / / doi.org/ 10.10 37/ bul0000449 Jansen, T., Meyer, J., Fleckenstein, J., Horbach, A., Keller, S. & Möller, J. (2024 b). Individualizing goal-setting interventions using automated writing evaluation to support secondary school students’ text revisions. Learning and Instruction, 89, 101847. https: / / doi.org/ 10. 1016/ j.learninstruc.2023.101847 Jansen, T., Höft, L., Bahr, L., Fleckenstein, J., Möller, J., Köller, O. & Meyer, J. (2024 c). Comparing Generative AI and Expert Feedback to Students’ Writing: Insights from Student Teachers. Psychologie in Erziehung und Unterricht, 71 (2), 80 - 92. Jansen, T., Höft, L., Bahr, L., Kuklick, L. & Meyer, J. (2025 a). Constructive feedback can function as a reward: Students’ emotional profiles in reaction to feedback perception mediate associations with task interest. Learning and Instruction, 95, 102030. https: / / doi.org/ 10.1016/ j.learninstruc.2024.102030 Jansen, T., Meyer, J., Fleckenstein, J., Wigfield, A. & Möller, J. (2025 b). “Can (A)I do this task? ” The role of AI as a socializer of students’ self-beliefs of their abilities. Learning and Individual Differences, 122, 102731. https: / / doi.org/ 10.1016/ j.lindif.2025.102731 Jansen, T., Horbach, A. & Meyer, J. (2025 c). Feedback from Generative AI: Correlates of Student Engagement in Text Revision from 655 Classes from Primary and Secondary School. Proceedings of the 15th International Learning Analytics and Knowledge Conference, 831 - 836. https: / / doi.org/ 10.1145/ 3706468.3706494 Jansen, T., Pünjer, H., Schaller, N.-J., Bahr, L. & Höft, L. (2025 d). Accelerating Secondary Students’ Learning Progression in Scientific Argumentation using Artificial Intelligence: An Educator’s Guide and Online Course. Research in subject matter teaching and learning, 8 (2), 1 - 26. https: / / doi.org/ 10.2478/ ristal-2025-0004 Kaliisa, R., Misiejuk, K., López-Pernas, S. & Saqr, M. (2026). How does artificial intelligence compare to human feedback? A meta-analysis of performance, feedback perception, and learning dispositions. Educational Psychology, 46 (1), 80 - 111. https: / / doi.org/ 10.1080/ 01443410.20 25.2553639 168 Thorben Jansen, Hannah Pünjer, Mira Tanz, Nils-Jonathan Schaller, Lars Höft Kalyuga, S. (2007). Expertise Reversal Effect and Its Implications for Learner-Tailored Instruction. Educational Psychology Review, 19 (4), 509 - 539. https: / / doi.org/ 10.1007/ s10648-007-9054-3 Kanfer, R. & Ackerman, P. L. (1989). Motivation and cognitive abilities: An integrative/ aptitude-treatment interaction approach to skill acquisition. Journal of Applied Psychology, 74 (4), 657 - 690. https: / / doi.org/ 10.1037/ 0021-9010.74.4.657 Keller, M. V., Dresel, M. & Daumiller, M. (2024). Do achievement goals and self-efficacy matter for feedback use? Learning and Instruction, 93, 101948. https: / / doi. org/ 10.1016/ j.learninstruc.2024.101948 Kluger, A. N. & DeNisi, A. (1996). The effects of feedback interventions on performance: A historical review, a meta-analysis, and a preliminary feedback intervention theory. Psychological Bulletin, 119 (2), 254 - 284. https: / / doi.org/ 10.1037/ 0033-2909.119.2.254 Luo, W., Li, H., Baek, E., Chen, S., Lam, K. H. & Semma, B. (2021). Reporting practice in multilevel modeling: A revisit after 10 years. Review Of Educational Research, 91 (3), 311 - 355. https: / / doi.org/ 10.3102/ 003465432 1991229 Meincke, L., Mollick, E., Mollick, L. & Shapiro, D. (2025). Prompting Science Report 1: Prompt Engineering is Complicated and Contingent (Version 1). arXiv. https: / / doi.org/ 10.48550/ ARXIV.2503.04818 Meyer, J., Jansen, T. & Fleckenstein, J. (2025). Nonengagement and unsuccessful engagement with feedback in lower secondary education: The role of student characteristics. Contemporary Educational Psycholog y, 81, 102363. https: / / doi.org/ 10.1016/ j.cedpsych.2025.10 2363 Meyer, J., Jansen, T., Schiller, R., Liebenow, L.W., Steinbach, M., Horbach, A. & Fleckenstein, J. (2024). Using LLMs to bring evidence-based feedback into the classroom: AI-generated feedback increases secondary students’ text revision, motivation, and positive emotions. Computers and Education: Artificial Intelligence, 6, 100199. Meyer, J., Jansen, T., Fleckenstein, J., Keller, S. & Köller, O. (2020). Machine learning im Bildungskontext: Evidenz für die Genauigkeit der automatisierten Beurteilung von Essays im Fach Englisch. Zeitschrift für Pädagogische Psychologie. https: / / doi.org/ 10.1024/ 1010- 0652/ a000296 Miller, A. L., Fassett, K.T. & Palmer, D. L. (2021). Achievement goal orientation: A predictor of student engagement in higher education. Motivation and Emotion, 45 (3), 327 - 344. https: / / doi.org/ 10.1007/ s11031-021-098 81-7 Muis, K. R., Ranellucci, J., Franco, G. M. & Crippen, K. J. (2013). The Interactive Effects of Personal Achievement Goals and Performance Feedback in an Undergraduate Science Class. The Journal of Experimental Education, 81 (4), 556 - 578. https: / / doi.org/ 10.1080/ 00220973.2012.738257 Murray, D. M. (1982). Learning by teaching. Montclair, NJ: Boynton/ Cook. Nicol, D. (2021). The power of internal feedback: Exploiting natural comparison processes. Assessment & Evaluation in Higher Education, 46 (5), 756 - 778. https: / / doi.org/ 10.1080/ 02602938.2020.1823314 Nielsen, K. (2014). Self-assessment methods in writing instruction: A conceptual framework, successful practices and essential strategies. Journal of Research in Reading, 37 (1), 1 - 16. https: / / doi.org/ 10.1111/ j.1467-9817.20 12.01533.x Nolen, S. B. (1988). Reasons for Studying: Motivational Orientations and Study Strategies, Cognition and Instruction, 5 (4), 269 - 287, https: / / doi.org/ 10.1207/ s1532690xci0504_2 Osborne, J. F., Henderson, J. B., MacPherson, A., Szu, E., Wild, A. & Yao, S.-Y. (2016). The development and validation of a learning progression for argumentation in science. Journal of Research in Science Teaching, 53 (6), 821 - 846. https: / / doi.org/ 10.1002/ tea.21316 Pajares, F. (2003). Self-efficacy beliefs, motivation, and achievement in writing: A review of the literature. Reading & Writing Quarterly, 19 (2), 139 - 158. https: / / doi.org/ 10.1080/ 10573560308222 Park, G., Schmidt, A. M., Scheu, C. & DeShon, R. P. (2007). A Process Model of Goal Orientation and Feedback Seeking. Human Performance, 20 (2), 119 - 145. https: / / doi.org/ 10.1080/ 08959280701332042 Pintrich, P. R., Smith, D. A. F., Garcia, T. & Mckeachie, W. J. (1993). Reliability and Predictive Validity of the Motivated Strategies for Learning Questionnaire (Mslq). Educational and Psychological Measurement, 53 (3), 801 - 813. https: / / doi.org/ 10.1177/ 00131644930530 03024 Porsdam-Mann, S., Vazirani, A. A., Aboy, M., Earp, B. D., Minssen, T., Cohen, I. G. & Savulescu, J. (2024). Guidelines for ethical use and acknowledgement of large language models in academic writing. Nature Machine Intelligence, 6 (11), 1272 - 1274. Robitzsch, A., Kiefer, T. & Wu, M. (2022). TAM: Test Analysis Modules (4.1-4) Computer software. https: / / cran.rproject.org/ web/ packages/ TAM/ index.html Schaller, N. J., Horbach, A., Höft, L. I., Ding, Y., Bahr, J. L., Meyer, J. & Jansen, T. (2024 a). DARIUS: A Comprehensive Learner Corpus for Argument Mining in German-Language Essays. Proceedings of the 2024 Joint International Conference on Computational Linguistics, Language Resources and Evaluation (LREC-COLING) 2024, 4356 - 4367. Schaller, N. J., Ding, Y., Horbach, A., Meyer, J. & Jansen, T. (2024 b). Fairness in Automated Essay Scoring: A Comparative Analysis of Algorithms on German Learner Essays from Secondary Education. Proceedings of the 19th Workshop on Innovative Use of NLP for Building Educational Applications (BEA) 2024, 210 - 221. Schunk, D. H. & Swartz, C. W. (1993). Goals and Progress Feedback: Effects on Self-Efficacy and Writing Achievement. Contemporary Educational Psycholog y, 18 (3), 337 - 354. https: / / doi.org/ 10.1006/ ceps.1993. 1024 Shih, S.-S. & Alexander, J. M. (2000). Interacting effects of goal setting and selfor other-referenced feedback on children’s development of self-efficacy and cognitive skill within the Taiwanese classroom. Journal of Educational Psychology, 92 (3), 536 - 543. https: / / doi.org/ 10.1037/ 0022-0663.92.3.536 UNESCO (2016). Education 2030: Incheon Declaration and Framework for Action for the implementation of Sustainable Development Goal 4. Ensure inclusive and Achievement Goals and Writing Skills Moderate Motivational AI Feedback Effects 169 equitable quality education and promote lifelong learning opportunities for all. Paris: UNESCO. Retrieved from https: / / unesdoc.unesco.org/ ark: / 48223/ pf0000 245656 Wang, J., Graham, S., Kim, Y.-S. G. & Steiss, J. (2025). Zooming into two measurement issues in writing selfefficacy: Revision as a distinct dimension and the generality hypothesis in argumentative writing. Reading and Writing. https: / / doi.org/ 10.1007/ s11145-025-10679-z Warm, T. A. (1989). Weighted Likelihood Estimation of Ability in Item Response Theory. Psychometrika, 54 (3), 427 - 450. https: / / doi.org/ 10.1007/ BF02294627 Wigfield, A. & Eccles, J. S. (2002). The Development of Competence Beliefs, Expectancies for Success, and Achievement Values from Childhood through Adolescence. In: Development of Achievement Motivation (pp. 91 - 120). Elsevier. https: / / doi.org/ 10.1016/ B978- 012750053-9/ 50006-1 Winstone, N. E., Balloo, K. & Carless, D. (2022). Discipline-specific feedback literacies: A framework for curriculum design. Higher Education, 83 (1), 57 - 77. https: / / doi.org/ 10.1007/ s10734-020-00632-0 Zhou, Y., Yu, S., Liu, B. & Jiang, L. (2022). Examining the role of writing proficiency in students’ feedback literacy development. Applied Linguistics Review, 15(2), 671-697. https: / / doi.org/ 10.1515/ applirev-2021-0133 Zimmerman, B. J. & Risemberg, R. (1997). Becoming a Self-Regulated Writer: A Social Cognitive Perspective. Contemporary Educational Psychology, 22 (1), 73 - 101. https: / / doi.org/ 10.1006/ ceps.1997.0919 Thorben Jansen https: / / orcid.org/ 0000-0001-9714-6505 E-Mail: tjansen@leibniz-ipn.de Hannah Pünjer https: / / orcid.org/ 0009-0008-6817-0694 E-Mail: puenjer@leibniz-ipn.de Mira Tanz E-Mail: tanz@leibniz-ipn.de Nils-Jonathan Schaller https: / / orcid.org/ 0000-0002-2604-5325 E-Mail: schaller@leibniz-ipn.de Lars Höft https: / / orcid.org/ 0000-0002-9040-8995 E-Mail: hoeft@leibniz-ipn.de Correspondence concerning this article should be addressed to Thorben Jansen Leibniz Institute for Science and Mathematics Education Olshausenstraße 62 24118, Kiel, Germany E-Mail: tjansen@leibniz-ipn.de Inklusion mit AVWS Dieses Buch gibt grundlegende Informationen zur Auditiven Verarbeitungs- und Wahrnehmungsstörung (AVWS), einer Störung, bei der akustische Reize gehört, aber nicht adäquat verarbeitet werden können. Konkrete und leicht anwendbare Praxishinweise unterstützen die Unterrichtsgestaltung und -durchführung bei der Inklusion von Schülerinnen und Schülern mit AVWS. 2025. 109 Seiten. 48 Abb. 7 Tab. Innenteil farbig. DIN A 4. (978-3-497-03298-3) kt a www.reinhardt-verlag.de