Field survey · perceptions research

AI Perceptions in Higher Education — Findings Explorer

Three linked studies on how students and staff perceive generative AI across institutions, subject areas, and research methods.

3 Studies
← Back to case study catalog
Published · Aug 2025

AI Across Borders: Exploring Perceptions and Interactions in Higher Education

Gerard, J., Singh, S., Macleod, M., McKay, M., Rivoire, A., Chakraborty, T., & Singh, M. (2025). Education Sciences, 15(8), 1039. doi.org/10.3390/educsci15081039

Compares quantitative Likert ratings and qualitative comments from 1,211 students at Ulster University (Northern Ireland) and IIT Delhi (India) across subject areas, testing for interactions between institution, subject area, and gender.

1,211

Participants

512

Ulster University

699

IIT Delhi

427

Free-text comments

Sample by gender and institution

Participant counts · Table 2, Gerard et al. (2025)

Ulster UniversityIIT Delhi
View data table
InstitutionMaleFemaleNon-binaryOtherNot statedTotal
Ulster University2012921045512
IIT Delhi563132202699

Awareness ratings, by institution

Mean of 5-point scale (1 = strongly disagree, 5 = strongly agree) · Table 7

Ulster UniversityIIT Delhi

A6 — "I plan to use ChatGPT or similar tools for my coursework in the future" — shows the widest gap between institutions (UU 2.66 vs IITD 4.39).

Perception ratings, by institution

Mean of 5-point scale · Table 8

Ulster UniversityIIT Delhi

Attitude score by subject area — Ulster University

Relative to Arts (reference = 0), from Bonferroni pairwise comparisons · Table 10. Whiskers = ±1 SE. Computer Sciences was the only subject area where UU and IITD ratings converged.

Model: institution F=83.80, p<.001 · gender F=4.88, p=.027 · subject area F=0.84, p=.471 (ns overall) · institution × subject area F=5.47, p<.001. Within UU alone, subject area was significant (F=5.80, p<.001); within IITD alone it was not (F=1.64, p=.163) — ratings there were uniformly high across fields.

View data table
Subject area (UU)Δ vs ArtsSESig.
Arts (reference)0.000——
Biomedical Sciences+0.2060.103ns
Social Sciences+0.3160.093**
Technological Sciences+0.3730.124*
Computer Sciences+0.6160.124***

Qualitative comment sentiment

Share of free-text comments by sentiment · Table 6

PositiveNeutralNegative

IIT Delhi comments skewed more positive than Ulster's; Arts and Biological Sciences comments at UU skewed more negative/neutral, echoing the lower attitude scores in those fields.

In press · Frontiers in Education

Methodological Variation in Studying Staff and Student Perceptions of AI Frontiers in Education

Gerard, J., Macleod, M., Norwood, K., Reid, A., & Singh, M. (in press). Frontiers in Education. doi.org/10.3389/feduc.2026.1716760

Compares three data collection methods (free-text comments, Padlet, focus groups) and four analytical approaches (sentiment analysis, stance analysis, word clouds, thematic analysis) on the same underlying question — do different methods agree?

292

Free-text comments

177

Padlet entries

11

Focus group participants

79%

Auto/manual agreement

Sample by respondent group

Free-text comments, n = 292

Neutral-sentiment share by method

Survey free-text vs. Padlet

Sentiment means clustered near the scale midpoint for both groups (survey F=3.00, p=.03 overall; pairwise ns; Padlet β=−0.148, p=.293) — an overall sentiment score barely distinguished students from staff.

Automated vs. manual sentiment scoring — agreement

RoBERTa model vs. human coder, on 452 scored comments

Of the disagreements, 20% (91 cases) were one point apart on a 3-point scale; only 1% (5 cases) diverged sharply — automated sentiment scoring is a reasonable proxy for coarse agreement, but (per the paper's central point) it flattens the group differences that thematic analysis reveals.

Word-cloud themes: shared vs. divergent

Most frequent terms across student and staff comments

Shared across both groups

toolworkhelpacademiclearningassessment

Divergent by group

students: courseworkstaff: training

Focus-group stance analysis found staff comments skewed more neutral/negative than students, who were more often supportive — a distinction the surface-level sentiment scores above did not capture.

Preprint · reframed, awaiting final draft

From Novelty to Normalisation: Tracking Changing Perceptions of AI in Higher Education, 2024–2026

Gerard, J., Macleod, M., Norwood, K., Reid, A., & Singh, M. arxiv.org/abs/2607.16223

A longitudinal tracking study of how perceptions of generative AI shifted across three survey waves (2024–2026) among four Ulster University stakeholder groups — undergraduates, doctoral researchers, teaching staff, and non-teaching staff.

1,665

Respondents

4

Stakeholder groups

3

Survey waves

'24–'26

Period tracked

This paper has been reframed since the version currently indexed on arXiv — a more recent draft is available. The reported direction of travel: students moved from "tentative experimentation to routine engagement" with GenAI, while staff maintained persistent concerns about academic integrity, assessment design, and critical thinking, with doctoral researchers and non-teaching staff in an intermediate position. Once the updated manuscript is shared, this panel will be rebuilt with the actual wave-by-wave familiarity, usage, and perceived-risk figures rather than this narrative summary.

Explore participant trends by subfield

Loaded by default with the published subject-area sample from AI Across Borders (Table 4). Upload a participant-level CSV to replace it — one row per participant, with columns for institution, subject area, gender, and (optionally) a numeric outcome such as an attitude score.

Drag a CSV here, or . Expected columns: institution, subject_area, gender, and optionally a numeric column (e.g. attitude_score).
Source: Gerard et al. (2025), Table 4

Participants by subject area

Grouped by institution

View data table