CoChat
Sign Up
Cited by 47Educational Strategies and EpistemologiesArtificial Intelligence in EducationInnovative Teaching and Learning Methods

“ ChatGPT can make mistakes. Check important info.” Epistemic beliefs and metacognitive accuracy in students' integration of ChatGPT content into academic writing

Open access
Urban, Marek, Brom, Cyril, Lukavský, Jiří, Děchtěrenko, Filip, Hein, Veronika, Svacha, Filip, Kmoníčková, Petra, Urban, Kamila·2025·British Journal of Educational Technology
DOI 10.1111/bjet.13591
Open this paper in CoChat — free accountOpen PDF

Abstract

Recent studies have conceptualized ChatGPT as an epistemic authority; however, no research has yet examined how epistemic beliefs and metacognitive accuracy affect students' actual use of ChatGPT‐generated content, which often contains factual inaccuracies. Therefore, the present experimental study aimed to examine how university students integrate correct and incorrect information from expert‐written and ChatGPT‐generated articles when writing independently ( N = 49) or with ChatGPT assistance ( N = 49). Students working with ChatGPT‐4o integrated more correct information from both expert‐written ( d = 0.64) and ChatGPT‐generated articles ( d = 0.95), but ChatGPT‐assisted writing did not affect the amount of incorrect information sourced from the ChatGPT‐generated article. Regardless of the condition, hierarchical regressions revealed that lower metacognitive bias was moderately associated with increased inclusion of correct information from the expert‐written article ( R 2 = 12%). Conversely, a higher metacognitive bias ( R 2 = 10%) and epistemic beliefs ( R 2 = 12%) were moderately related to the inclusion of incorrect information from ChatGPT‐generated articles. These findings suggest that while ChatGPT assistance enhances the integration of correct human‐ and AI‐generated content, metacognitive skills remain essential to mitigate the risks of incorporating incorrect AI‐generated information. Practitioner notes: What is already known about this topic Generative AI tools, such as ChatGPT, are increasingly regarded as epistemic authorities due to their authoritative tone and human‐like interaction. ChatGPT has demonstrated utility in providing correct information and improving productivity in educational and professional contexts, but it is also prone to inaccuracies, hallucinations and misleading content. Students' epistemic beliefs and metacognitive skills predict their ability to critically evaluate and integrate conflicting information from multiple resources, particularly when searching for information on the Internet. What this paper adds This study experimentally examines how students integrate correct and incorrect information from expert‐written and ChatGPT‐generated articles when writing independently or with ChatGPT's assistance. The findings show that ChatGPT assistance improves the inclusion of correct information but does not significantly reduce or increase the inclusion of incorrect ChatGPT‐generated content. Metacognitive accuracy and epistemic beliefs are key factors in mitigating the inclusion of incorrect information, regardless of whether students work independently or with ChatGPT. Implications for practice and/or policy Generative AI tools can outperform human experts in specific scenarios, requiring little to no evaluation. However, in situations where these tools generate misleading or incorrect content, the application of metacognitive skills and epistemic beliefs becomes essential to discern reliable information and avoid the integration of errors. Educational interventions should include activities requiring justification of knowledge, evaluation of resources and reflection upon human‐generated and AI‐generated texts to enhance students' ability to discern accurate from inaccurate information. Interventions focused on metacognitive accuracy and epistemic awareness can empower individuals to critically evaluate and differentiate between reliable and erroneous information, enhancing their recognition of misinformation.

Integrity signals

Registry

Crossref · 10.1111/bjet.13591

Cites

76 of 76 references verified0 citation errors · 0 not checkable

Statistics

11 tests recomputed · 0 with the reported and recomputed p on opposite sides of .05

Claims the paper makes

4 of 36 cited sources checked · 1 supported · 0 flagged · accuracy not yet measured

  • Findingown data

    Experimental condition (ChatGPT-assisted writing) moderately predicted the inclusion of correct information from the expert-written article, explaining 9% of variance.

    “In Model 0 , the experimental condition (independent vs. ChatGPT-assisted writing) moderately predicted the inclusion of correct information, β = 0.37, p = 0.002, explaining R 2 = 9% of the variance.”
    Quote in the paperSubstantial claim
  • Findingown data

    Lower metacognitive bias significantly predicted higher inclusion of correct information from the expert-written article, increasing explained variance by 12%.

    “In Model 1 , lower metacognitive bias significantly predicted higher inclusion of correct information from the expert-written article, β = -0.34,p < 0.001, increasing the explained variance by ΔR 2 = 12%.”
    Quote in the paperSubstantial claim
  • Findingown data

    Article length was not associated with the inclusion of incorrect ChatGPT-generated information.

    “but not with the inclusion of incorrect ChatGPT-generated information (r = 0.03, p = 0.76).”
    Quote in the paperNot a substantial claim
  • Findingown data

    ChatGPT-assisted writing led to greater inclusion of correct information from the expert-written article compared to independent writing.

    “For the inclusion of correct information sourced from the expert-written article, the participants working with ChatGPT included moderately more correct information, F(1,93) = 11.43,p < 0.001, η p 2 = 0.11.”
    Quote in the paperSubstantial claim
  • Findingown data

    ChatGPT-assisted writing led to greater inclusion of correct information from the ChatGPT-generated article compared to independent writing.

    “For correct information sourced from the ChatGPT-generated article, participants working with ChatGPT included significantly more correct information, F(1,93) = 20.41,p < 0.001, with a large effect size, η p 2 = 0.18.”
    Quote in the paperSubstantial claim
  • May combine two claims. 13
  • Classified as probably not a substantial claim. 21
  • The quote only partly states this claim. 3

  • May combine two claims.

    Premisecitation

    Trust in generative AI tools can lead to overreliance, discouraging users from exploring alternative perspectives and reducing variability of ideas, resulting in more homogeneous outputs.

    “Trust in these tools can lead to overreliance, discouraging users from exploring alternative perspectives and reducing the variability of ideas, which results in more homogeneous outputs (Ding et al., 2023;Doshi & Hauser, 2024).”
    Quote in the paperSubstantial claim1 of 2 sources checked
    • cites Ding et al., 2023 source

      The Implications section reports that many students blindly trusted ChatGPT’s answers and heavily relied on them, supporting the overreliance part of the claim. The Accuracy and accountability of AI section discusses algorithms excluding alternative perspectives, but does not link that to users’ trust. The paper does not report reduced variability of ideas or more homogeneous outputs. (LLM, conf=0.88)

      Experimental — accuracy not yet measured
    • cites Doshi & Hauser, 2024

      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).

    Premisecitation

    Students with more sophisticated epistemic beliefs are more likely to activate and utilise metacognitive strategies such as planning, monitoring, regulation and evaluation when encountering conflicting information.

    “Bråten et al. (2011) proposed that students with more sophisticated epistemic beliefs are more likely to activate and utilise metacognitive strategies, such as planning, monitoring, regulation and evaluation, when encountering conflicting information.”
    Quote in the paperNot a substantial claim1 of 1 sources checked
    • cites Bråten et al. (2011) source

  • May combine two claims.

    Premisecitation

    Generative AI tools produce outputs that often appear highly credible but are prone to factual errors, fabricated sources and misleading content.

    “Generative AI tools produce outputs that often appear highly credible, but are simultaneously prone to factual errors, fabricated sources and misleading content (Chelli et al., 2024;Ferrario et al., 2024).”
    Quote in the paperSubstantial claim2 of 2 sources checked
    • cites Chelli et al., 2024 source

      The Conclusions and Discussion report that ChatGPT and Bard generated convincingly authentic but hallucinated references for systematic reviews, supporting the claim about credible-looking, fabricated and misleading sources. The Strengths and Limitations section says the study was restricted to three models and systematic reviews of rotator cuff pathology; it does not establish that generative AI outputs generally are prone to factual errors across settings. (LLM, conf=0.86)

      Experimental — accuracy not yet measured
    • cites Ferrario et al., 2024 source

      The Introduction (page not provided) says large language models’ accurate responses can create an impression of expertise, and the Final Remarks (page not provided) say ChatGPT is much less reliable for scientific investigations than for general-knowledge answers. This supports the contrast between apparent credibility and possible inaccuracy, but our copy does not substantiate the broader claim that generative AI outputs are often erroneous or that they contain fabricated sources and misleading content. (LLM, conf=0.84)

      Experimental — accuracy not yet measured

  • The quote only partly states this claim.

    Findingown data

    Higher metacognitive bias was associated with greater inclusion of incorrect information (β = 0.31, p = 0.002; ΔR2 = 10%).

    “However, the results indicate that a higher metacognitive bias was moderately associated with integrating inaccuracies (ΔR 2 = 10%).”
    Quote in the paperSubstantial claim
    • Rests on an assumption: The metacognitive bias index used in the study validly measures overconfidence in the context of the writing task.
    • Rests on an assumption: The coding of correct and incorrect information in student articles was reliable and valid.

  • Classified as probably not a substantial claim.

    Study design and conductown data

    Participants were randomly assigned to either the control (independent writing) or experimental (ChatGPT-assisted writing) group prior to the start of the study.

    “Students who volunteered were randomly assigned to either the control (independent writing) or the experimental group (ChatGPT-assisted writing) prior to the start of the study.”
    Quote in the paperSubstantial claim

  • The quote only partly states this claim.

    Study design and conductown data

    The overall interrater agreement for coding information as correct or incorrect was ICC(3,3) = 0.92, indicating excellent consistency among raters.

    “The overall interrater agreement was ICC(3,3) = 0.92, indicating excellent consistency among raters.”
    Quote in the paperSubstantial claim

  • Classified as probably not a substantial claim.

    Premisecitation

    Higher metacognitive bias is associated with a greater likelihood of making errors when dealing with inaccuracies in texts.

    “Additionally, research has shown that higher metacognitive bias is associated with a greater likelihood of making errors when dealing with inaccuracies in texts (Salovich & Rapp, 2021), difficulties in reasoning about controversies (Lang et al., 2021), and greater resistance to correcting misinformed beliefs (Rapp & Withall, 2024).”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites Salovich & Rapp, 2021 source

  • May combine two claims.

    Interpretationcitation

    The negative relationship might also reflect a potential trade-off: students deeply engaged in evaluating knowledge either perceive their self-regulatory strategies as less effective or underestimate them, especially when dealing with conflicting information.

    “Furthermore, the negative relationship might also reflect a potential trade-off: students deeply engaged in evaluating knowledge either perceive their self-regulatory strategies as less effective (Boekaerts & Corno, 2005;Zimmerman & Martinez-Pons, 1992), or underestimate them (Seban et al., 2024), especially when dealing with conflicting information (Maier & Richter, 2014;Stadtler & Bromme, 2007).”
    Quote in the paperNot a substantial claim0 of 3 sources checked
    • cites Boekaerts & Corno, 2005 source
    • cites Zimmerman & Martinez-Pons, 1992

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites Seban et al., 2024

      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). The quote only partly states this claim.

    Interpretationcitation

    The negative relationship between justification for knowing and inclusion of incorrect information may reflect the cognitive demands of critical evaluation, as students who prioritise reasoning must allocate significant cognitive resources to assess reliability and consistency of information.

    “This finding may reflect the cognitive demands of critical evaluation, as students who prioritise reasoning must allocate significant cognitive resources to assess the reliability and consistency of information (Anmarkrud et al., 2019).”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites Anmarkrud et al., 2019 source

  • Classified as probably not a substantial claim. May combine two claims.

    Premisecitation

    Individuals with stronger beliefs about the certainty and reliability of Internet-based knowledge were less likely to critically evaluate the credibility of sources and displayed overconfidence in their ability to work with information.

    “In contrast, Kammerer et al. (2013) observed that individuals with stronger beliefs about the certainty and reliability of Internet-based knowledge were less likely to critically evaluate the credibility of sources and displayed overconfidence in their ability to work with information.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites Kammerer et al. (2013) source

  • Classified as probably not a substantial claim.

    Premisecitation

    Higher metacognitive bias is associated with difficulties in reasoning about controversies.

    “Additionally, research has shown that higher metacognitive bias is associated with a greater likelihood of making errors when dealing with inaccuracies in texts (Salovich & Rapp, 2021), difficulties in reasoning about controversies (Lang et al., 2021), and greater resistance to correcting misinformed beliefs (Rapp & Withall, 2024).”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites Lang et al., 2021 source

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    Working with generative AI tools such as ChatGPT increases productivity in complex problem-solving tasks in educational and professional settings.

    “Working with generative AI tools, such as ChatGPT, increases productivity in complex problem-solving tasks in both educational and professional settings (Noy & Zhang, 2023;Sun & Zhou, 2024;Urban et al., 2024).”
    Quote in the paperNot a substantial claim0 of 3 sources checked
    • cites Noy & Zhang, 2023 source
    • cites Sun & Zhou, 2024 source
    • cites Urban et al., 2024 source

  • May combine two claims.

    Interpretationown data

    ChatGPT assistance enhances the integration of correct human- and AI-generated content, but metacognitive skills remain essential to mitigate the risks of incorporating incorrect AI-generated information.

    “These findings suggest that while ChatGPT assistance enhances the integration of correct human-and AI-generated content, metacognitive skills remain essential to mitigate the risks of incorporating incorrect AI-generated information.”
    Quote in the paperSubstantial claim

  • Classified as probably not a substantial claim. May combine two claims.

    Premisecitation

    The Cognitive Reflection Test (CRT) is a measure of reflective reasoning and the ability to override intuitive, automatic responses.

    “To assess metacognitive bias, the present study employed the Cognitive Reflection Test (CRT; Frederick, 2005), a measure of reflective reasoning and the ability to override intuitive, automatic responses.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites Frederick, 2005 source

  • Classified as probably not a substantial claim.

    Premisecitation

    Metacognitive bias in university students solving non-routine problem-solving tasks was linked to their perceived usefulness of ChatGPT rather than their actual performance.

    “In the context of generative AI use, Urban et al. (2024) found that metacognitive bias in university students solving non-routine problem-solving tasks was linked to their perceived usefulness of ChatGPT rather than their actual performance.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites Urban et al. (2024) source

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Study design and conductcitation

    The sample size was determined a priori using G*Power with α = 0.05, β = 0.80, and an expected effect size of Cohen's d = 0.62, indicating a minimum of 42 participants per group.

    “For the between-group comparison, the sample size was determined a priori using G*Power 3.1.9.6, with α = 0.05, β = 0.80, and an expected effect size of Cohen's d = 0.62.This effect size was estimated based on average between-group differences reported by Urban et al. (2024).The calculation indicated that a minimum of 42 participants per group would be required.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites Urban et al. (2024) source

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    Individuals relying heavily on Internet-based sources are more likely to view knowledge as certain, unchanging and handed down by an authority.

    “It was previously found that individuals relying heavily on Internet-based sources are more likely to view knowledge as certain, unchanging and handed down by an authority (Ståhl et al., 2021).”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites Ståhl et al., 2021

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • May combine two claims.

    Premisecitation

    Those with higher confidence in their ability to resist inaccuracies often made more errors, revealing a lack of metacognitive calibration.

    “Those with higher confidence in their ability to resist inaccuracies often made more errors, revealing a lack of metacognitive calibration (Salovich & Rapp, 2021).”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites Salovich & Rapp, 2021 source

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    Epistemic beliefs positively correlate with academic achievement across educational stages, though the overall effect size is small (r = 0.16).

    “The meta-analysis by Greene et al. (2018) found that epistemic beliefs positively correlate with academic achievement across educational stages, from elementary school to graduate studies, though the overall effect size is small (r = 0.16).”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites Greene et al. (2018) source

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductcitation

    In the present study, epistemic beliefs were treated as static.

    “Second, in the present study, epistemic beliefs were treated as static, yet they may be influenced by situational factors, such as task demands and repeated interactions with information sources (Hofer & Sinatra, 2010).”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites Hofer & Sinatra, 2010 source

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Study design and conductown data

    All participants passed two attention checks and completed the required tasks; thus, no exclusions were made.

    “All participants passed two attention checks and completed the required tasks; thus, no exclusions were made.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductnone given

    The rubric categorising information as correct or incorrect from both texts was developed prior to the coding process.

    “The rubric categorising information as correct or incorrect from both texts was developed prior to the coding process.”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim. May combine two claims.

    Premisecitation

    Human users often fail to recognise AI-generated text and are generally overconfident in their ability to do so.

    “Previous studies suggested that human users often fail to recognise AI-generated text and are generally overconfident in their ability to do so (Casal & Kessler, 2023;Fleckenstein et al., 2024).”
    Quote in the paperSubstantial claim0 of 2 sources checked
    • cites Casal & Kessler, 2023 source
    • cites Fleckenstein et al., 2024 source

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductreasoning

    The study does not account for developmental differences in epistemic beliefs and metacognitive processes.

    “First, the study does not account for developmental differences in epistemic beliefs and metacognitive processes.”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim.

    Premisecitation

    Higher metacognitive bias is associated with greater resistance to correcting misinformed beliefs.

    “Additionally, research has shown that higher metacognitive bias is associated with a greater likelihood of making errors when dealing with inaccuracies in texts (Salovich & Rapp, 2021), difficulties in reasoning about controversies (Lang et al., 2021), and greater resistance to correcting misinformed beliefs (Rapp & Withall, 2024).”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites Rapp & Withall, 2024 source

  • May combine two claims.

    Interpretationown data

    While ChatGPT can enhance performance, students' ability to accurately assess their own thinking remains crucial to effectively integrate information from human-generated sources not included in the training dataset of generative AI tools.

    “These findings suggest that while ChatGPT can enhance performance, students' ability to accurately assess their own thinking remains crucial to effectively integrate information from human-generated sources not included in the training dataset of generative AI tools.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Study design and conductown data

    There were no significant differences between the groups in terms of gender, age, field of study, academic level, prior knowledge about vitamin C or prior experience with ChatGPT.

    “As can be seen, there were no significant differences between the groups in terms of gender, age, field of study, academic level, prior knowledge about vitamin C or prior experience with ChatGPT.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    Exposure to authoritative sources can influence how individuals integrate conflicting views, often reducing critical and evaluative thinking.

    “Finally, exposure to authoritative sources can influence how individuals integrate conflicting views, often reducing critical and evaluative thinking (Ku et al., 2014).”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites Ku et al., 2014

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • Classified as probably not a substantial claim. May combine two claims.

    Study design and conductreasoning

    The ChatGPT environment was neither fine-tuned nor artificially modified; participants interacted with the standard, unaltered ChatGPT interface to mirror real-world usage scenarios.

    “To ensure ecological validity, the ChatGPT environment was neither fine-tuned nor artificially modified; participants interacted with the standard, unaltered ChatGPT interface to mirror real-world usage scenarios.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductnone given

    The study relied on self-reports and separate-task measures of metacognitive bias, which may not fully reflect the dynamic nature of these constructs.

    “Finally, the study relied on self-reports and separate-task measures of metacognitive bias, which may not fully reflect the dynamic nature of these constructs.”
    Quote in the paperNot a substantial claim
30 of 30

  • The paper relies on this without support.

    There is currently no clear conceptualisation of epistemic beliefs about AI-generated knowledge.

    “Despite the importance of epistemic beliefs in working with multiple information sources, there is currently no clear conceptualisation of epistemic beliefs about AI-generated knowledge.”

  • The paper relies on this without support.

    No research has yet examined how epistemic beliefs and metacognitive accuracy affect students' actual use of ChatGPT-generated content.

    “Recent studies have conceptualized ChatGPT as an epistemic authority; however, no research has yet examined how epistemic beliefs and metacognitive accuracy affect students' actual use of ChatGPT-generated content, which often contains factual inaccuracies.”

  • The paper relies on this without support.

    The expert-written article used in the study contained correct information.

    “The results showed that the students who wrote with the assistance of ChatGPT included significantly more correct expert knowledge in their articles compared to those who wrote independently (R 2 = 9%).”

  • The paper relies on this without support.

    The ChatGPT-generated article used in the study contained correct information.

    “Our findings further showed that students working with ChatGPT integrated its correct content more often than those who worked independently (R 2 = 19%).”

  • The paper relies on this without support.

    The ChatGPT-generated article used in the study contained incorrect information.

    “The present study showed that ChatGPT assistance neither significantly helped nor harmed in reducing the inclusion of incorrect information in students' writings.”

  • The paper relies on this without support.

    The metacognitive bias index used in the study validly measures overconfidence in the context of the writing task.

    “Higher metacognitive bias (ie, overconfidence) was associated with a greater inclusion of incorrect information, β = 0.31, p = 0.002.”

  • The paper relies on this without support.

    The coding of correct and incorrect information in student articles was reliable and valid.

    “The results showed that the students who wrote with the assistance of ChatGPT included significantly more correct expert knowledge in their articles compared to those who wrote independently (R 2 = 9%).”
7 of 7

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

  • Urban et al. (2025) exposed students to expert and ChatGPT content containing known correct and incorrect claims, allowing incorporated information to be scored, but did not directly observe checking.

    “Urban et al. (2025) exposed students to expert and ChatGPT content containing known correct and incorrect claims, allowing incorporated information to be scored, but did not directly observe checking.”

    From Beyond checking: verification quality, reliance calibration, and learning in generative AI-assisted higher education (2026) · cites this paper as Urban et al., 2025

  • Self-regulation in GenAI environments includes epistemic monitoring, verification, and the ability to resist the automatic acceptance of plausible AI output.

    “These findings are especially important because they show that self-regulation in GenAI environments is not limited to effort management or planning; it also includes epistemic monitoring, verification, and the ability to resist the automatic acceptance of plausible AI output.”

    From Generative AI, Cognitive Offloading, and Learner Agency in Higher Education: A Scoping Review (2026) · cites this paper as Urban et al., 2025

  • Urban et al. (2025) could score the accuracy of content incorporated into academic writing but not whether students later detected similar errors independently.

    “Urban et al. (2025) could score the accuracy of content incorporated into academic writing but not whether students later detected similar errors independently.”

    From Beyond checking: verification quality, reliance calibration, and learning in generative AI-assisted higher education (2026) · cites this paper as Urban et al., 2025

Statistics that don’t add up

Each reported p-value recomputed from its test statistic and degrees of freedom.

  • Inconsistent p-value: χ 2 (24) = 24.48,p = .44

    Minor

    The paper reports p = .44 here; recomputing from the test statistic and its degrees of freedom gives p = .434. The check cannot tell whether the statistic or the p-value is the misprint.

    Reported p = .44 · recomputed p = .434

  • p-value fits an unstated one-tailed test: F(1,93) = 11.43,p < .001

    Minor

    The paper reports p < .001 here; recomputing from the test statistic and its degrees of freedom gives p = .001. The reported value is what a one-tailed test gives, but the paper does not say near this result that a one-tailed test was used.

    Reported p < .001 · recomputed p = .001

2 of 2

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.

  • Doesn’t add upχ²(24) = 24.48reported p = .44 · recomputed p = .434

    “χ 2 (24) = 24.48,p = .44”

  • Doesn’t add upF(1, 93) = 11.43reported p < .001 · recomputed p = .001

    “F(1,93) = 11.43,p < .001”

  • Consistentχ²(2) = 1.61reported p = .45 · recomputed p = .447

    “χ 2 (2) = 1.61, p = .45”

  • Consistentt(96) = 0.68reported p = .5 · recomputed p = .498

    “t(96) = 0.68, p = .50”

  • Consistentχ²(2) = 0.22reported p = .9 · recomputed p = .896

    “χ 2 (2) = 0.22, p = .90”

  • Consistentχ²(1) = 1.01reported p = .32 · recomputed p = .315

    “χ 2 (1) = 1.01, p = .32”

  • Consistentt(96) = 0.08reported p = .94 · recomputed p = .936

    “t(96) = 0.08, p = .94”

  • Consistentt(96) = 0.58reported p = .56 · recomputed p = .563

    “t(96) = 0.58, p = .56”

  • ConsistentF(1, 93) = 20.41reported p < .001 · recomputed p < .001

    “F(1,93) = 20.41,p < .001”

  • ConsistentF(1, 93) = 0.76reported p = .38 · recomputed p = .386

    “F(1,93) = 0.76, p = .38”

  • ConsistentF(1, 96) = 16.77reported p < .001 · recomputed p < .001

    “F(1,96) = 16.77,p < .001”

11 of 11

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

Flagged first, then by citations

Research works printed but not linked: 2

  1. Bråten I., 2016, Handbook of individual differences in reading: Text and context, 67
  2. 10.3102/0034654306700108 · DOI

Cited by

Sorted by citations

Related

Authors
Urban, M., Brom, C., Lukavský, J., Děchtěrenko, F., Hein, V., Svacha, F., Kmoníčková, P., & Urban, K.
Affiliations
  • Czech Academy of Sciences (CZ)
  • Charles University (CZ)
  • Slovak Academy of Sciences (SK)
Corresponding author
Urban, Marek
ORCID
Venue
British Journal of Educational Technology
Publisher
Wiley
Published
2025-04-09
Volume
56
Issue
5
Pages
1897-1918
ISSN
  • 0007-1013 (ISSN-L)
  • 1467-8535
DOI
10.1111/bjet.13591
OpenAlex
W4409296452
Cited by
47
References
63
Citations by year
2026: 39 · 2025: 8
FWCI
92.91
Type
journal article
Language
en
Open access
Licence
onlinelibrary.wiley.com/termsAndConditions#vor (vor)
Locations
British Journal of Educational Technology (publishedVersion)
Funders
Grantová Agentura České Republiky (24‐11515S)
Topics
  • Social Sciences › Psychology › Developmental and Educational Psychology › Educational Strategies and Epistemologies
  • Physical Sciences › Computer Science › Information Systems › Artificial Intelligence in Education
  • Social Sciences › Psychology › Developmental and Educational Psychology › Innovative Teaching and Learning Methods
Keywords
ChatGPT, epistemological beliefs, metacognitive accuracy, metacognitive bias, AI-generated content, factual inaccuracies, hallucinations, academic writing, information integration, generative AI in education, university students, critical evaluation of information, GPT-4o, AI-assisted writing, epistemic authority, misinformation risk, information literacy, metacognition, epistemology

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