Engineering student perceptions of generative AI use in learning trust judgment and information seeking

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

The rapid growth of GenAI raises important concerns about academic integrity, overreliance, and the reliability of AI-generated content, particularly when students use such tools without sufficient judgment or verification.

“However, its rapid growth also raises important concerns about academic integrity, overreliance, and the reliability of AI-generated content, particularly when students use such tools without sufficient judgment or verification [3,25,26].”

• cites [3] Chan CKY. Students’ perceptions of ‘AI-giarism’: investigating changes in understandings of academic misconduct. Educ Inf Technol. 2025;30(6):8087–108. sourceSupported
• cites [25] Song N. Higher education crisis: academic misconduct with generative AI. J Conting Crisis Manag. 2024;32(1):e12532. https://doi.org/10.1111/1468-5973.12532. sourceNot checked — no full text of the source
• cites [26] Wu Y. Integrating generative AI in education: how ChatGPT brings challenges for future learning and teaching. J Adv Res Educ. 2023;2(4):7–10. https://doi.org/10.56397/JARE.2023.07.02. sourcePartly supported

  In “The Future Is Now” (page unavailable), the paper raises cheating as a concern amid ChatGPT’s rapid adoption. “Hindered social learning due to reduction of face-to-face interaction” warns that students who overly depend on its answers may lose opportunities to develop critical thinking. The copy does not specifically address the reliability of AI-generated content or students’ verification of its outputs. (LLM, conf=0.85)

  Experimental — accuracy not yet measured

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

Teachers' perceptions of GenAI have been found to become more favorable with greater use, suggesting that experience may reduce hesitation and support broader acceptance in educational practice.

“Similarly, teachers' perceptions of GenAI have been found to become more favorable with greater use, suggesting that experience may reduce hesitation and support broader acceptance in educational practice [15].”

• cites [15] Kaplan-Rakowski R, Grotewold K, Hartwick P, Papin K. Generative AI and teachers’ perspectives on its implementation in education. J Interact Learn Res. 2023;34(2):313–38. https://doi.org/10.70725/815246mfssgp. sourcePartly supported

  In ‘Teachers’ Perceptions of GAI Integration in Education’ (page unavailable), the authors report positive attitudes and suggest that prior experience may help explain them. The ‘Implications’ section (page unavailable) suggests greater exposure might increase confidence and willingness to use AI. The cited passages do not establish that perceptions became more favorable as teachers used GenAI more frequently; that stronger finding is not supported by the body passages reviewed. (LLM, conf=0.82)

  Experimental — accuracy not yet measured

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

Students may distinguish between clearly unacceptable uses, such as submitting AI-generated work, and more nuanced uses, such as using AI to expand or refine their own ideas.

“Related discussions further suggest that students may distinguish between clearly unacceptable uses, such as submitting AI-generated work, and more nuanced uses, such as using AI to expand or refine their own ideas [3].”

• cites [3] Chan CKY. Students’ perceptions of ‘AI-giarism’: investigating changes in understandings of academic misconduct. Educ Inf Technol. 2025;30(6):8087–108. sourceSupported

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

Students emphasize that GenAI tools should supplement rather than replace human teaching and interaction.

“However, students emphasize that these tools should supplement rather than replace human teaching and interaction [6].”

• cites [6] Elliott S, Richter M, Alozie C, Raidis D. Student perceptions of personalised GenAI agents on student learning, success, engagement, and wellbeing: a pilot study of Cogniti AI. Education and Information Technologies. Advance online publication. (2026) https://doi.org/10.1007/s10758-025-09942-7 sourceSupported

May combine two claims.

In the full sample of 56 respondents, 9 (16%) reported not using AI at all, 19 (34%) reported using it a little, 10 (18%) reported using it somewhat, 12 (21%) reported using it often, and 6 (11%) reported using it extensively.

“As shown in Fig. 3, 9 students (16%) reported not using AI at all, 19 (34%) reported using it a little, 10 (18%) reported using it somewhat, 12 (21%) reported using it often, and 6 (11%) reported using it extensively.”

May combine two claims.

Among the 47 respondents who reported at least some AI use, agreement was highest for the items mapped to monitoring and browsing, while lower agreement was observed for the items mapped to starting and verifying.

“Among the 47 respondents who reported at least some AI use, agreement was highest for the items mapped to monitoring and browsing, while lower agreement was observed for the items mapped to starting and verifying.”

May combine two claims.

50% of heavy users agreed that they relied on AI instead of thinking through the problem themselves, compared with none of the light users.

“However, 50% of heavy users agreed that they relied on AI instead of thinking through the problem themselves, compared with none of the light users.”

May combine two claims.

76% of light users reported applying their own judgment rather than copying AI, compared with 33% of heavy users.

“Conversely, 76% of light users reported applying their own judgment rather than copying AI, compared with 33% of heavy users.”

May combine two claims.

Most reported users disagreed that they had relied on AI rather than thinking through the problem themselves, while responses regarding the necessity of AI were more varied.

“Most reported users disagreed that they had relied on AI rather than thinking through the problem themselves, while responses regarding the necessity of AI were more varied.”

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

Students noted that calculation-based items could often be completed more quickly because they mainly involved applying equations or looking up formulas.

“The third theme was faster completion of procedural problems, where students noted that calculation-based items could often be completed more quickly because they mainly involved applying equations or looking up formulas.”

May combine two claims.

Among respondents who reported at least some AI use, higher agreement was observed for items concerning ongoing task support, perceived deeper learning, and the application of their own judgment, whereas lower agreement was observed for using AI to understand the problem requirements initially or to check and validate technical work.

“Among respondents who reported at least some AI use, higher agreement was observed for items concerning ongoing task support, perceived deeper learning, and the application of their own judgment, whereas lower agreement was observed for using AI to understand the problem requirements initially or to check and validate technical work.”

• AssumptionRests on an assumption: The survey items for learning and cognition and for trust and judgment validly measure those constructs.
• AssumptionRests on an assumption: Students' self-reported extent of AI use (Q3) accurately reflects their actual AI use during the assignment.

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

Many students described the assignment as taking roughly 20 to 30 min to complete.

“The first was estimated effort, with many students describing the assignment as taking roughly 20 to 30 min to complete.”

May combine two claims.

Within this convenience sample, respondents generally described AI as a source of conceptual clarification, faster access to information, formula recall, and support when checking their reasoning.

“Within this convenience sample, respondents generally described AI as a source of conceptual clarification, faster access to information, formula recall, and support when checking their reasoning.”

May combine two claims.

For the overreliance item, 30 respondents (63.8%) disagreed or strongly disagreed that they relied on AI instead of thinking through the problem themselves, 8 (17.0%) were neutral, and 9 (19.1%) agreed or strongly agreed.

“For the overreliance item, 30 respondents (63.8%) disagreed or strongly disagreed that they relied on AI instead of thinking through the problem themselves, 8 (17.0%) were neutral, and 9 (19.1%) agreed or strongly agreed.”

May combine two claims.

Respondents also commonly reported perceived benefits for deeper learning and submission quality, although these measures reflect students' perceptions rather than direct evidence of improved learning or performance.

“Respondents also commonly reported perceived benefits for deeper learning and submission quality, although these measures reflect students' perceptions rather than direct evidence of improved learning or performance.”

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

Among the reported-user subgroup, Copilot was selected by one respondent (2.1%).

“Copilot was selected by one respondent (2.1%), while Bing AI was not selected by any respondent in the reporteduser subgroup.”

• AssumptionRests on an assumption: Students' self-reported extent of AI use (Q3) accurately reflects their actual AI use during the assignment.

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

Among the reported-user subgroup, Bing AI was not selected by any respondent.

“Copilot was selected by one respondent (2.1%), while Bing AI was not selected by any respondent in the reporteduser subgroup.”

• AssumptionRests on an assumption: Students' self-reported extent of AI use (Q3) accurately reflects their actual AI use during the assignment.

May combine two claims.

These responses may have been influenced by differences in how students interpreted AI use, including indirect use through Google's AI-generated overview, and may not accurately represent their actual interactions with AI or their behavior while completing the assignment.

“These responses may have been influenced by differences in how students interpreted AI use, including indirect use through Google's AI-generated overview, and may not accurately represent their actual interactions with AI or their behavior while completing the assignment.”

May combine two claims.

Qualitative data suggest a spectrum ranging from minimal or indirect use to more active dependence, with most responses falling in the middle and describing AI as a support tool for speed in clarification and checking understanding.

“Instead, qualitative data suggest a spectrum ranging from minimal or indirect use to more active dependence, with most responses falling in the middle and describing AI as a support tool for speed in clarification and checking understanding.”

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

Because the qualitative coding was conducted by one author, intercoder agreement was not calculated.

“Because the coding was conducted by one author, intercoder agreement was not calculated.”

May combine two claims.

No inferential statistical analyses were conducted; therefore, the findings are interpreted as exploratory and specific to this course and assignment.

“No inferential statistical analyses were conducted; therefore, the findings are interpreted as exploratory and specific to this course and assignment.”

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

Structured workshops and guided exposure to tools such as ChatGPT have been associated with stronger teacher confidence and clearer recognition of practical benefits, including resource creation, productivity, and support for lesson planning.

“For example, structured workshops and guided exposure to tools such as ChatGPT have been associated with stronger teacher confidence and clearer recognition of practical benefits, including resource creation, productivity, and support for lesson planning [4].”

• cites [4] Cordero J, Torres-Zambrano J, Cordero-Castillo A. Integration of generative artificial intelligence in higher education: best practices. Educ Sci. 2025;15(1):32. https://doi.org/10.3390/educsci15010032. sourceNot checked yet

  Checked before — its earlier verdict is held back until a new check runs.

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

Group differences by level of AI engagement were examined descriptively using agreement percentages; no inferential tests were conducted.

“Group differences were examined descriptively using agreement percentages; no inferential tests were conducted.”

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

Prior work emphasizes opportunities in curriculum development, simulations, and faculty support alongside challenges related to bias, accuracy, and faculty training.

“while also emphasizing opportunities in curriculum development, simulations, and faculty support alongside challenges related to bias, accuracy, and faculty training [27].”

• cites [27] Yelamarthi K, Dandu R, Rao M, Yanambaka VP, Mahajan S. Exploring the potential of generative AI in shaping engineering education: opportunities and challenges. J Eng Educ Transform. 2024;37:439–45. https://doi.org/10.16920/jeet/2024/v37is2/24072. (Special Issue). sourceNot checked yet

  Checked before — its earlier verdict is held back until a new check runs.

May combine two claims.

These descriptive patterns suggest that greater reported AI use was accompanied by stronger perceptions of usefulness and learning support, but also by indications of greater dependence and lower reported independent judgment.

“These descriptive patterns suggest that greater reported AI use was accompanied by stronger perceptions of usefulness and learning support, but also by indications of greater dependence and lower reported independent judgment.”

Classified as probably not a substantial claim.

Ellis's model conceptualizes information-seeking as a set of related activities that include starting, chaining, browsing, differentiating, monitoring, extracting, verifying, and ending.

“Ellis's model conceptualizes information-seeking as a set of related activities that include starting, chaining, browsing, differentiating, monitoring, extracting, verifying, and ending [7,8,23].”

• cites [7] Ellis D. A behavioural approach to information retrieval system design. J Doc. 1989;45(3):171–212. https://doi.org/10.1108/eb026843. sourceNot checked — no full text of the source
• cites [8] Ellis D, Haugan M. Modelling the information seeking patterns of engineers and research scientists in an industrial environment. J Doc. 1997;53(4):384–403. https://doi.org/10.1108/EUM0000000007204. sourceNot checked — no full text of the source
• cites [23] Rusli FA, Indah RN, Achmad Syam RZ. Analysis of the Ellis model on information seeking behavior of Library Science Department students of Nusantara Islamic University on the determination of internship places. TADWIN Jurnal Ilmu Perpustakaan dan Informasi. 2024;4(2):99–120. https://doi.org/10.19109/tadwin.v4i2.21003. sourceNot checked yet

  Checked before — its earlier verdict is held back until a new check runs.

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

The learning and cognition category and the trust and judgment category each contained only two conceptually distinct items, so they were analyzed at the item level rather than as composite scales.

“Learning and cognition (Q11-Q12) and trust and judgment (Q6 and Q13) each contained only two conceptually distinct items and were therefore analyzed at the item level rather than treated as internally consistent composite scales.”

• AssumptionRests on an assumption: The survey items for learning and cognition and for trust and judgment validly measure those constructs.

May combine two claims.

The exploratory and descriptive design does not support causal conclusions or statistically significant comparisons between user groups.

“Finally, the exploratory and descriptive design does not support causal conclusions or statistically significant comparisons between user groups.”

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

The themes from open-ended responses were checked against the complete response set to ensure they reflected the range of reported experiences.

“The themes were then checked against the complete response set to ensure that they reflected the range of reported experiences.”

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

The field raises concerns about bias, misinformation, and the reliability of AI-generated content, particularly when used without verification in academic settings.

“The field also raises concerns about bias, misinformation, and the reliability of AI-generated content, particularly when used without verification in academic settings [2,20].”

• cites [2] Alasadi EA, Baiz CR. Generative AI in education and research: opportunities, concerns, and solutions. J Chem Educ. 2023;100(7):2965–71. https://doi.org/10.1021/acs.jchemed.3c00323. sourceNot checked — no full text of the source
• cites [20] Ng DTK, Chan EKC, Lo CK. Opportunities, challenges and school strategies for integrating generative AI in education. Comput Educ AI. 2025;8:100373. https://doi.org/10.1016/j.caeai.2025.100373. sourceNot checked — no full text of the source

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

The analysis was primarily descriptive and exploratory, consistent with examining how students reported using GenAI and perceived its role in their learning and problem-solving.

“The analysis was primarily descriptive and exploratory, consistent with the purpose of examining how students reported using generative AI during the assignment and how they perceived its role in their learning and problem-solving.”

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

Recent work suggests that AI systems can reshape information-seeking practices by influencing how users search, retrieve, scan, and evaluate information, while also potentially discouraging the triangulation of sources.

“This perspective becomes especially relevant in AI-mediated contexts, as recent work suggests that AI systems can reshape information-seeking practices by influencing how users search, retrieve, scan, and evaluate information, while also potentially discouraging the triangulation of sources [12].”

• cites [12] Hirvonen N, Jylhä V, Lao Y, Larsson S. Artificial intelligence in the information ecosystem: affordances for everyday information seeking. J Am Soc Inf Sci. 2024;75(10):1152–65. https://doi.org/10.1002/asi.24860. sourceNot checked yet

  Checked before — its earlier verdict is held back until a new check runs.

May combine two claims.

The findings were drawn from a convenience sample of 56 students enrolled in one mechanical engineering course and were situated within a single fatigue-failure assignment.

“First, the findings were drawn from a convenience sample of 56 students enrolled in one mechanical engineering course and were situated within a single fatigue-failure assignment.”

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

These findings are specific to one course, one assignment, and a convenience sample and should not be interpreted as evidence of causal effects or actual patterns of AI use.

“These findings are specific to one course, one assignment, and a convenience sample and should not be interpreted as evidence of causal effects or actual patterns of AI use.”

Probably not a substantial claim: may not pass S4 (based).

Such support from GenAI depends on students critically evaluating the explanations rather than accepting them uncritically.

“However, such support depends on students critically evaluating the explanations rather than accepting them uncritically.”

May combine two claims.

Chaining and verifying showed more moderate agreement, suggesting that AI was used by some respondents to explore possible approaches and to check calculations or concepts.

“Chaining and verifying showed more moderate agreement, suggesting that AI was used by some respondents to explore possible approaches and to check calculations or concepts.”

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

Generative AI is often discussed as a resource that may support learning, improve efficiency, and expand access to educational assistance.

“Consequently, GenAI is often discussed as a resource that may support learning, improve efficiency, and expand access to educational assistance [5,6].”

• cites [5] Daher W, Hussein A. Higher education students’ perceptions of GenAI tools for learning. Information. 2024;15(7):416. https://doi.org/10.3390/info15070416. sourceNot checked — no full text of the source
• cites [6] Elliott S, Richter M, Alozie C, Raidis D. Student perceptions of personalised GenAI agents on student learning, success, engagement, and wellbeing: a pilot study of Cogniti AI. Education and Information Technologies. Advance online publication. (2026) https://doi.org/10.1007/s10758-025-09942-7 sourceNot checked yet

  Checked before — its earlier verdict is held back until a new check runs.

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

Concerns persist regarding the reliability of GenAI outputs, as large AI models can produce content that appears highly convincing yet may be false or misleading.

“Simultaneously, concerns persist regarding the reliability of GenAI outputs, as large AI models can produce content that appears highly convincing yet may be false or misleading.”

• cites [24] Shoaib MR, Wang Z, Taleby Ahvanooey M, Zhao J. Deepfakes, misinformation, and disinformation in the era of frontier AI, generative AI, and large AI models. In: 2023 International Conference on Computer and Applications (ICCA). IEEE. 2023. https://doi.org/10.1109/ICCA59364.2023.10401723 sourceNot checked yet

  Checked before — its earlier verdict is held back until a new check runs.

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

A study review illustrates that research has examined GenAI in education through themes such as application, impact, ethics, adoption, and user perspectives.

“A study review illustrates that research has examined GenAI in education through themes such as application, impact, ethics, adoption, and user perspectives [28].”

• cites [28] Yusuf A, Pervin N, Román-González M, Noor NM. Generative AI in education and research: a systematic mapping review. Rev Educ. 2024;12(2):e3489. https://doi.org/10.1002/rev3.3489. sourceNot checked — no full text of the source

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

The survey instrument was developed based on the study's conceptual framework [19] and research focus on learning, trust, and information-seeking.

“To examine students' perceptions of generative AI in engineering learning, a survey instrument was developed based on the study's conceptual framework [19] and research focus on learning, trust, and information-seeking.”

• cites [19] Mohammed Y, Qandil H. Generative AI and student learning processes: a conceptual framework for engineering education. In: Proceedings of the ASEE Conference for Industry and Education Collaboration (CIEC). American Society for Engineering Education; (in press).Not matched

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

May combine two claims.

Browsing also showed comparatively high agreement, reflecting the perceived value of AI for obtaining explanations and clarifying concepts.

“Browsing also showed comparatively high agreement, reflecting the perceived value of AI for obtaining explanations and clarifying concepts.”

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

AI use becomes problematic when it constitutes unauthorized assistance, particularly in contexts where institutional or course-level expectations are unclear.

“Prior literature notes that AI use becomes problematic when it constitutes unauthorized assistance, particularly in contexts where institutional or course-level expectations are unclear [1,25].”

• cites [1] Abd-Elaal ES, Gamage SH, Mills JE. Artificial intelligence is a tool for cheating academic integrity. In: 30th Annual Conference for the Australasian Association for Engineering Education (AAEE 2019): Educators becoming agents of change: innovate, integrate, motivate. Brisbane, Queensland: Engineers Australia. 2019. p. 397–403.Not checked yet

  Not checked in the free run.

• cites [25] Song N. Higher education crisis: academic misconduct with generative AI. J Conting Crisis Manag. 2024;32(1):e12532. https://doi.org/10.1111/1468-5973.12532. sourceNot checked — no full text of the source

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

Recent systematic review evidence indicates that the responsible adoption of GenAI in higher education depends on a combination of academic support, student self-regulation, and clear institutional governance.

“Recent systematic review evidence indicates that the responsible adoption of GenAI in higher education depends on a combination of academic support, student self-regulation, and clear institutional governance.”

• cites [14] Kangwa D, Msafiri MM, Fute A. Exploring the factors that promote a balance between academic integrity and the effective use of GenAI tools in higher education: A systematic review. J Comput Assist Learn. 2025;41(5):e70109. sourceNot checked — no full text of the source

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

In engineering education, problem-solving extends beyond retrieving information and instead requires students to identify, analyze, interpret, and evaluate complex problems through iterative reasoning and application.

“In engineering education, this lens is particularly important because problem-solving extends beyond retrieving information and instead requires students to identify, analyze, interpret, and evaluate complex problems through iterative reasoning and application [13].”

• cites [13] Husin M, Usmeldi U, Masdi H, Simatupang W, Fadhilah F, Hendriyani Y. Project-based problem learning: improving problem-solving skills in higher education engineering students. Int J Sociol Educ. 2025;14(1):62–84. https://doi.org/10.17583/rise.15125. sourceNot checked yet

  Checked before — its earlier verdict is held back until a new check runs.

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

Existing work shows that although GenAI may strengthen autonomy, personalization, and learner confidence, excessive reliance may also create risks such as cognitive overload, reduced critical thinking, and weaker social learning.

“Existing work shows that although these tools may strengthen autonomy, personalization, and learner confidence, excessive reliance may also create risks such as cognitive overload, reduced critical thinking, and weaker social learning [26].”

• cites [26] Wu Y. Integrating generative AI in education: how ChatGPT brings challenges for future learning and teaching. J Adv Res Educ. 2023;2(4):7–10. https://doi.org/10.56397/JARE.2023.07.02. sourceNot checked yet

  Not checked in the free run.

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

Q15 responses were analyzed among the 47 reported AI users; 46 of these respondents provided a non-missing response.

“Q15 responses were analyzed among the 47 reported AI users; 46 of these respondents provided a non-missing response.”

• AssumptionRests on an assumption: Students' self-reported extent of AI use (Q3) accurately reflects their actual AI use during the assignment.

May combine two claims.

Starting showed the lowest agreement, indicating that respondents were comparatively less likely to report using AI to understand the initial problem requirements.

“Starting showed the lowest agreement, indicating that respondents were comparatively less likely to report using AI to understand the initial problem requirements.”