How do students regulate their learning with a genAI chatbot?
Open accessAbstract
Self-regulated learning (SRL) is essential for effective learning, yet students often struggle to regulate in digital environments, including suboptimal learning tool use. The rapid integration of generative AI (e.g., ChatGPT) into learning settings raises questions about their role in supporting or hindering SRL. This exploratory study investigated how students’ SRL in a technology-enhanced learning environment with a genAI tool was associated with learning processes and performance. Thirty university students were tasked to read texts and write an essay within 45 minutes. Learning performance was measured using knowledge and transfer tests, and the essay. SRL processes were measured using trace-based SRL event analysis and learner-genAI interaction with chatbot log events and coded queries. Most students (73%) used the chatbot voluntarily, primarily for seeking information. Chatbot users achieved higher essay scores than non-users. Chatbot interaction frequencies correlated positively with high cognitive activities; durations of chatbot use and high cognition negatively correlated with reading time. Qualitative data indicated reliance on the chatbot to summarise and extract key points, suggesting offloading of learning. Findings highlight potential performance benefits but also risks of outsourcing critical SRL processes to genAI. Implications point to instructional and tool design that align technological advances with educational fit.
Integrity signals
Registry
Cites
Statistics
Claims the paper makes
6 of 16 cited sources checked · 5 supported · 0 flagged · accuracy not yet measured
- Premisecitation
A recurring challenge has been learners' suboptimal interactions with SRL support tools, such as when and how to use them to support learning.
“Yet, regardless of design, a recurring challenge has been learners' suboptimal interactions with these tools, such as when and how to use them to support learning.”
Quote in the paperSubstantial claim1 of 1 sources checked- cites Saint et al., 2024 source
- Findingown data
Students had sufficient metacognitive strategy knowledge (M = 75.80%, SD = 11.10%).
“and sufficient metacognitive strategy knowledge (M = 75.80%,SD = 11.10%).”
Quote in the paperNot a substantial claim - Findingown data
The majority of students voluntarily used the chatbot (73%; n = 22).
“The majority of the students voluntarily used the chatbot (73%; n = 22).”
Quote in the paperNot a substantial claim - Findingown data
All students learned significantly from the task (pre-post learning gain), t(29) = 4.90, p < .001, d = .89.
“All students learned significantly from the task (pre-post learning gain), 𝑡(29) = 4.90, 𝑝 < .001,𝑑 = .89.”
Quote in the paperNot a substantial claim - Findingown data
Students used the chatbot on average 11.90% of the learning time.
“Students used the chatbot on average 11.90% of the learning time.”
Quote in the paperNot a substantial claim
- Findingown data
Higher frequencies of chatbot interactions were linked to higher cognitive activities, such as writing the essay (chatbot events: rs = .70, padj < .05; chatbot queries: rs = .62, padj < .05).
“Results revealed that higher frequencies of chatbot interactions were linked to higher cognitive activities, such as writing the essay (chatbot events: rs = .70,padj < .05;chatbot queries: rs = .62,padj < .05).”
Quote in the paperSubstantial claim - Findingown data
Chatbot users showed higher rereading frequencies than non-users.
“Interesting differences between the groups are that chatbot users showed higher rereading frequencies”
Quote in the paperNot a substantial claim - Findingown data
High cognitive activities were not significantly associated with rereading activities after p-value adjustment (rs = .49, padj > .05).
“High cognitive activities were not significantly associated with rereading activities after p-value adjustment (rs = .49,padj > .05).”
Quote in the paperNot a substantial claim - Findingown data
Students often did not engage in deeper processing of the learning materials on their own, but rather asked the chatbot to highlight the key points in the text.
“Looking into students' chatbot queries, we found that students often did not engage in deeper processing of the learning materials on their own, but rather, asked the chatbot to highlight the key points in the text.”
Quote in the paperSubstantial claim - Interpretationcitation
The findings reflect a well-documented phenomenon where students possess sufficient SRL knowledge and skills but fail to spontaneously engage in SRL during their learning (i.e., production deficit).
“The findings reflect a welldocumented phenomenon where students possess sufficient SRL knowledge and skills but fail to spontaneously engage in SRL during their learning (i.e., production deficit; Veenman et al., 2005).”
Quote in the paperSubstantial claim0 of 1 sources checked- cites Veenman et al., 2005 source
- Premisecitation
Learners who make productive use of support tools through active and consistent engagement with them demonstrated higher learning outcomes.
“learners who make productive use of support tools through active and consistent engagement with them demonstrated higher learning outcomes (Bannert, Sonnenberg, et al., 2015;Zheng et al., 2023).”
Quote in the paperSubstantial claim0 of 2 sources checked - Interpretationown data
The chatbot benefits performance.
“Chatbot users had higher essay scores than non-users, which suggests that the tool benefits performance.”
Quote in the paperSubstantial claim- Rests on an assumption: The difference in essay scores between chatbot users and non-users is attributable to chatbot use rather than pre-existing differences or other confounders.
- Interpretationown data
For some students, the chatbot potentially replaced reading, which is a core learning activity to build an initial knowledge base.
“Hence, for some students, the chatbot potentially replaced reading, which is a core learning activity to build an initial knowledge base.”
Quote in the paperSubstantial claim - Interpretationown data
Students adopted strategies driven by a performance-driven approach in order to complete the task efficiently.
“We observed that students adopted strategies driven by a performance-driven approach in order to complete the task efficiently.”
Quote in the paperSubstantial claim - Interpretationown data
Students prioritised task completion over deeper learning with the provided texts.
“the findings suggest that students prioritised task completion over deeper learning with the provided texts.”
Quote in the paperSubstantial claim
- Classified as probably not a substantial claim. 14
- May combine two claims. 13
Classified as probably not a substantial claim. May combine two claims.
PremisecitationResearchers have raised concerns about learners treating genAI as a crutch and forgoing critical skills, e.g., excessive cognitive offloading leading to lesser depth of cognitive processing and outsourcing of metacognitive activities hampering productive SRL enactment.
“Researchers have raised concerns about learners treating genAI as a crutch and forgoing critical skills-e.g., excessive cognitive offloading of tasks, leading to lesser depth of cognitive processing (Stadler et al., 2024) and outsourcing of metacognitive activities, hampering productive SRL enactment (Fan et al., 2024).”
Quote in the paperSubstantial claim2 of 2 sources checkedClassified as probably not a substantial claim.
Study design and conductcitationSRL processes were measured via trace data with Fan et al.'s (2022) approach, using an SRL process library that parsed raw data into sequences of learning actions representing SRL activities.
“We measured SRL processes via trace data with Fan et al.'s (2022) approach, using an SRL process library (see Appendix A) that parsed raw data into sequences of learning actions representing SRL activities.”
Quote in the paperSubstantial claim1 of 1 sources checked- cites Fan et al.'s (2022) source
Probably not a substantial claim: may not pass S3 (load-bearing).
PremisecitationgenAI provides novel affordances previously limited by technological capabilities, such as adaptive and interactive feedback.
“genAI provides novel affordances previously limited by technological capabilities, such as adaptive and interactive feedback (Bauer, Richters, et al., 2025;Neshaei et al., 2025).”
Quote in the paperNot a substantial claim1 of 2 sources checkedMay combine two claims.
PremisecitationLearners often struggle with effective learning regulation, particularly in digital settings, leading to missing learning outcome effects.
“learners have been found to often struggle with effective learning regulation, particularly in digital settings, leading to missing learning outcome effects (e.g., Azevedo & Feyzi-Behnagh, 2011;Lim et al., 2021).”
Quote in the paperSubstantial claim1 of 2 sources checked- cites Azevedo & Feyzi-Behnagh, 2011 source
- cites Lim et al., 2021 source
The Introduction states that students have difficulty regulating their learning in digital or online settings and that self-regulated learning is associated with better learning outcomes. It does not establish that those difficulties cause “missing learning outcome effects”; the Results instead report a significant learning gain in the study’s online setting. (LLM, conf=0.85)
Experimental — accuracy not yet measured
May combine two claims.
Findingown dataQueries for seeking information constituted the largest proportion (53%), followed by language support (25%), generating ideas (13.18%), and correction/repetition (8.53%).
“Queries for seeking information constituted the largest proportion (53%), followed by language support (25%), generating ideas (13.18%), and lastly, correction/repetition (8.53%).”
Quote in the paperNot a substantial claimProbably not a substantial claim: may not pass S3 (load-bearing).
Findingown dataStudents had moderate AI literacy (M = 49.50%, SD = 14.30%).
“Results indicated that students had moderate AI literacy (M = 49.50%,SD = 14.30%)”
Quote in the paperNot a substantial claimMay combine two claims.
Findingown dataNegative correlations were found between duration proportions of reading and high cognition (rs = -.74, padj < .05), as well as chatbot use and reading (rs = -.61, padj < .05).
“Negative correlations were found between duration proportions of reading and high cognition (rs = -.74,padj < .05),as well as chatbot use and reading (rs = -.61,padj < .05).”
Quote in the paperSubstantial claimMay combine two claims.
Findingown dataChatbot users had higher essay scores than non-users, trending towards better performance in the essay (t(28) = 2.14, p < .05, d = .88).
“chatbot users showed higher rereading frequencies, and trended towards better performance in the essay, t(28) = 2.14, 𝑝 < .05,𝑑 = .88.”
- Also stated in: § Discussion and conclusion · plain
“Chatbot users had higher essay scores than non-users”
Quote in the paperSubstantial claim- Also stated in: § Discussion and conclusion · plain
May combine two claims.
Findingown dataThere was a total of 381 interaction events, about one-third of which were queries posed by students to the chatbot (nqueries = 129).
“There was a total of 381 interaction events, about one-third of which were queries posed by students to the chatbot (nqueries = 129).”
Quote in the paperNot a substantial claimProbably not a substantial claim: may not pass S3 (load-bearing).
Findingown dataNo other notable between-group trends were found.
“No other notable between-group trends were found.”
Quote in the paperNot a substantial claimMay combine two claims.
Findingown dataThere was a declining trend of queries posed over time, with students seeking information from the chatbot more intensively in the first half of the learning session.
“Figure 2 shows the declining trend of queries posed over time, with students seeking information from the chatbot more intensively in the first half of the learning session.”
Quote in the paperNot a substantial claimProbably not a substantial claim: may not pass S2 (checkable).
PremisecitationSelf-regulated learning (SRL) is a dynamic process in which learners actively monitor and control their learning activities to pursue their goals.
“Self-regulated learning (SRL) is a dynamic process in which learners actively monitor and control their learning activities to pursue their goals (Winne & Hadwin, 1998)”
Quote in the paperNot a substantial claim0 of 1 sources checked- cites Winne & Hadwin, 1998
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
Probably not a substantial claim: may not pass S3 (load-bearing).
Study design and conductnone givenThe study was approved by the university ethics committee.
“The study was approved by the university ethics committee.”
Quote in the paperNot a substantial claimClassified as probably not a substantial claim.
Study design and conductreasoningGiven the study's exploratory nature and small sample size, causal interpretations should be avoided.
“However, given the study's exploratory nature and small sample size, causal interpretations should be avoided.”
Quote in the paperSubstantial claimMay combine two claims.
Interpretationown dataA genAI-integrated environment potentially supports the learning task, but it could also hamper productive SRL processes, particularly metacognitive processes (i.e., low enactment of planning and monitoring).
“Therefore, the results suggest that a genAI-integrated environment potentially supports the learning task, but it could also hamper productive SRL processes, particularly metacognitive processes (i.e., low enactment of planning and monitoring).”
Quote in the paperSubstantial claim- Rests on an assumption: The low enactment of planning and monitoring observed is attributable to the genAI-integrated environment rather than to the task or sample characteristics.
Classified as probably not a substantial claim. May combine two claims.
Study design and conductnone givenLearning performance was measured via pre-post domain knowledge tests (Cronbach's α = 0.78), a transfer test (Cronbach's α = 0.58), and the essay (Cohen's κ, inter-rater reliability = 0.88).
“Learning performance was measured via pre-post domain knowledge tests (Cronbach's α = 0.78), a transfer test (Cronbach's α = 0.58), and the essay (Cohen's κ, inter-rater reliability = 0.88).”
Quote in the paperSubstantial claimClassified as probably not a substantial claim.
Study design and conductcitationTo control for multiple comparisons, the Benjamini-Hochberg false discovery rate correction was applied.
“To control for multiple comparisons, we applied the Benjamini-Hochberg false discovery rate correction (Benjamini & Hochberg, 1995) using the p.adjust function with method =”
Quote in the paperSubstantial claim0 of 2 sources checked- cites Benjamini & Hochberg, 1995 source
- cites R Core Team, 2024
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
May combine two claims.
Interpretationown dataStudents offloaded the more superficial learning activity (reading) for a deeper learning activity (high cognition), but they outsourced comprehension and synthesis activities to the chatbot.
“Although students offloaded the more superficial learning activity (reading) for a deeper learning activity (high cognition), they outsourced comprehension and synthesis activities to the chatbot (e.g., identifying and synthesising relevant parts of the reading text).”
Quote in the paperSubstantial claimClassified as probably not a substantial claim.
Study design and conductown dataStudents' chatbot queries were coded post hoc (κ = 0.72) into the categories: Generating Idea (GI), Seeking Information (SI), Language Support (LS), and Correction/Repetition (CR).
“Students' chatbot queries were coded post hoc (κ = 0.72) into the categories: Generating Idea (GI), Seeking Information (SI), Language Support (LS), and Correction/Repetition (CR)”
Quote in the paperSubstantial claimMay combine two claims.
Interpretationown dataStudents relied on the chatbot to make important regulatory decisions, such as evaluating the relevance of the text, deciding what to read and why, and integrating new information with prior knowledge.
“the qualitative analyses indicated that students relied on the chatbot to make important regulatory decisions, such as evaluating the relevance of the text, deciding what to read and why, and integrating new information with prior knowledge.”
Quote in the paperSubstantial claimMay combine two claims.
Interpretationown dataStudents who focused on reading the text used the chatbot less, and at the same time, engaged in high cognitive activities to a lesser degree.
“Conversely, students who focused on reading the text used the chatbot less, and at the same time, engaged in high cognitive activities to a lesser degree.”
Quote in the paperSubstantial claimProbably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.
Study design and conductnone givenParticipation was voluntary, and all students provided active consent prior to data collection.
“Participation was voluntary, and all students provided active consent prior to data collection.”
Quote in the paperNot a substantial claimProbably not a substantial claim: may not pass S3 (load-bearing).
Study design and conductcitationMetacognitive strategy knowledge was assessed with a questionnaire with Cronbach's α = 0.79.
“and a metacognitive strategy knowledge questionnaire (Bannert, Pieger, & Sonnenberg, 2015; Cronbach's α = 0.79).”
Quote in the paperNot a substantial claim0 of 1 sources checked- cites Bannert, Pieger, & Sonnenberg, 2015 source
Probably not a substantial claim: may not pass S3 (load-bearing).
Study design and conductcitationAI literacy was assessed with a test with internal consistency Cronbach's α = 0.70.
“We assessed learners' prerequisites with an AI literacy test (Hornberger et al., 2023; Cronbach's α, internal consistency = 0.70)”
Quote in the paperNot a substantial claim0 of 1 sources checked- cites Hornberger et al., 2023 source
The paper relies on this without support.
The difference in essay scores between chatbot users and non-users is attributable to chatbot use rather than pre-existing differences or other confounders.
“Chatbot users had higher essay scores than non-users, which suggests that the tool benefits performance.”
The paper relies on this without support.
The low enactment of planning and monitoring observed is attributable to the genAI-integrated environment rather than to the task or sample characteristics.
“the results suggest that a genAI-integrated environment potentially supports the learning task, but it could also hamper productive SRL processes, particularly metacognitive processes (i.e., low enactment of planning and monitoring).”
The paper relies on this without support.
The chatbot's custom prompt successfully prevented it from writing the essay for students.
“We pre-configured the chatbot with a custom prompt: 1) The chatbot should not write the essay for students, 2) Responses should be succinct, 3) Suggestions should be based on provided information.”
Statistics recomputed
Every reported test result, recomputed from its test statistic and degrees of freedom. A result shows as not adding up only once an independent check has confirmed it.
- Consistentt(29) = 4.9reported p < .001 · recomputed p < .001
“t(29) = 4.90, p < .001”
- Consistentt(28) = 2.14reported p < .05 · recomputed p = .041
“t(28) = 2.14, p < .05”
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
Self-check results appear here once an independent check has confirmed them.
Cites
The publisher hasn’t deposited this paper’s reference list.
Research works printed but not linked: 1
- [20]20. Schunk, D. H., & Greene, J. A. (2017). Handbook of self-regulation of learning and performance (2nd ed.). Routledge.
- Authors
- Lim, L., & Bannert, M.
- Affiliations
- FernUniversität in Hagen (DE)
- Technical University of Munich (DE)
- Venue
- Learning Letters
- Publisher
- Adelaide University
- Published
- 2026-02-23
- Volume
- 8
- Pages
- 61
- ISSN
- 2981-877X (ISSN-L)
- OpenAlex
- W7132820087
- Semantic Scholar
- c2207832eede0afb3c28cf7f06d895ae47c0907f
- Cited by
- 0
- References
- 0
- FWCI
- 0.00
- Type
- journal article
- Language
- en
- Open access
- diamond
- doi.org
- Locations
- Learning Letters (publishedVersion)
- Topics
- Social Sciences › Psychology › Developmental and Educational Psychology › Innovative Teaching and Learning Methods
- Physical Sciences › Computer Science › Information Systems › Artificial Intelligence in Education
- Physical Sciences › Computer Science › Artificial Intelligence › Intelligent Tutoring Systems and Adaptive Learning
- Keywords
- student self-regulation, generative AI chatbots, ChatGPT, essay writing, knowledge transfer, cognitive load, offloading, technology-enhanced learning, university students, learning performance
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
Claims read by deepseek-v4.1-flash under definition version 3 (extract-v3) on 2026-10-11.
Classified by jev-1.13-20260917 (gold-guideline-v0.2) on 2026-10-11.
Cited sources are checked by gpt-6-sol.
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
