Exploring Multimodal LLM Engagement on Prompt Engineering Competency Among Non-Tech Students in India

Authors

  • Meltem Alkoyak-Yildiz
  • Sandrapuri Aswin Tejasvi
  • Venkatesh Viswanathan
  • Luis Felipe Hernandez Mora

DOI:

https://doi.org/10.34190/icer.3.1.5685

Keywords:

Prompt Engineering, Multimodal Interaction, LLM Competency, AI Literacy, Higher Education

Abstract

Large Language Models (LLMs) have become increasingly prevalent in the daily use of university students’ educational contexts, yet the specific perceptual LLM engagement patterns that contribute to effective LLM usage remain weakly understood. This exploratory cross-sectional study (N = 75) examines which factors of LLM engagement most strongly predict self-perceived prompt engineering competency among non-tech university students in India. Academic participants completed a survey measuring: Self-Perceived Prompt Engineering Competency (dependent variable), Basic Modality Use of LLM Input (text, images, audio, video, code), Strategic Integration & Application (purposeful combination of modalities), Depth of Use (usage of basic queries vs complex multi-modal tasks), and Self-Readiness & Access to AI tools, Primary Adoption Barrier to use AI. Non-parametric Wilcoxon signed-rank tests, multiple regression, and barrier analyses through Chi-square tests were conducted. Wilcoxon signed-rank tests determined whether scores significantly exceeded the theoretical scale midpoint of 3.0. Three of five constructs were significantly above the midpoint: Self-Readiness & Access (W = 1323, p < .001, r = .782, large), Prompt Engineering Competency (W = 2197, p < .001, r = .768, large), and Integration & Application (W = 1008, p = .002, r = .463, medium). Basic Modality Use (W = 516, p = .997) and Depth of Use (W = 496, p = .963) did not exceed the midpoint. These results indicate that despite high perceived competency, multimodal and complex LLM engagement remains limited. Results of the multiple regression analysis show that the four-predictor regression model explains 41.6% of the variance in competency (F(4, 70) = 12.5, p < .001). Integration & Application was the strongest unique predictor (B = 0.51, p < .001), followed by Self-Readiness & Access (B = 0.34, p = .006). Neither Basic Modality Use nor Depth of Use contributed unique variance. A Chi-square goodness-of-fit test revealed a significantly unequal barrier distribution (χ²(5, N = 75) = 37.6, p < .001, w = 0.708). Lack of knowledge was the most prevalent Adoption Barrier to use AI (37.3%) and scored significantly lower on competency than students reporting privacy concern (U = 158, p = .034, r = .373). This study demonstrates that Prompt Engineering Competency is driven specifically by the purposeful integration of LLMs into academic workflows, rather than frequent exposure or access alone. Implications for AI literacy curriculum design are further discussed.

Downloads

Published

2026-11-13