← All research Field study · ICETC 2026, Porto · Accepted

Lowering the barrier: diagnosing readiness before designing the training

Prashanth Shenoy & Mirka Saarela · Faculty of Information Technology, University of Jyväskylä

122 / 119

responses before and after, pooled across two sites

5.39

baseline ease of use, the lowest of five dimensions on a seven-point scale

d = 0.43

the shift in ease of use, the largest of the five

2 of 5

dimensions moved. The other three had nowhere to go

The study in six slides
01 / 06

Training is usually evaluated backwards

The standard way to judge a teacher AI workshop is to measure attitudes afterwards and check whether they improved. It sounds reasonable until you ask what the attitudes were beforehand.

If a cohort already agrees that AI is useful, already intends to use it, and already enjoys the idea, then a post-workshop survey will report high scores no matter what happened in the room.

Instructional design has always put needs analysis first. This study applies that discipline to teacher AI training.

02 / 06

Two rooms, tested for comparability before pooling

Identical 90-minute ChatGPT curriculum-design workshops ran at two Bengaluru-area sites: a coaching institution (44 before, 40 after) and a private school (78 before, 79 after). Pooled, that is 122 responses before and 119 after.

Pooling was not assumed. The two independently recruited cohorts were compared on all five baseline dimensions first, and differed on none of them (all p > .19), so combining them is supported by the data rather than by convenience.

Surveys were anonymous, which means pre and post cannot be individually paired. The comparison is between groups, not within people.

03 / 06

The baseline is the finding

Five dimensions were measured on a seven-point scale: intention to adopt, perceived usefulness, enjoyment, AI self-efficacy, and perceived ease of use.

Intention 5.96
Usefulness 5.89
Enjoyment 5.78
Self-efficacy 5.67
Ease of use 5.39

Three sit near the ceiling before anyone is trained. Ease of use is roughly a quarter to half a point below the rest. The room did not need persuading. It needed fluency.

04 / 06

And the two with room to move are the two that moved

After the workshop, ease of use and self-efficacy show significant differences (d = 0.43 and 0.33). Usefulness, enjoyment and intention do not.

That is not a failed workshop. It is a ceiling. The paper is explicit that no detectable change on a near-ceiling dimension cannot be read as evidence the workshop did nothing for it.

Ease of use d .43 Self-efficacy d .33 Enjoyment d .21 Intention d .18 Usefulness d .09 5.0 5.5 6.0 6.5 before after accent = significant at .05
05 / 06

Item level is where it gets useful

Inside those two dimensions, the movement is not evenly spread. The items that shift hardest are concrete:

  • Effort, that AI does not demand much of it (d = 0.49)
  • Integrability, that it fits the tools already in use (d = 0.47)
  • Resource access and learnability (d = 0.29, 0.27)

Meanwhile confidence in actually using AI in the classroom barely moves at all (d = 0.07). Ninety minutes lowers the perceived cost of the tool. It does not make anyone feel ready to teach with it.

06 / 06

Which is why the instrument exists

A cohort with low ease of use and high usefulness has an accessibility barrier, and structured practice will help. A cohort with high ease of use and low intention has a motivational barrier, and needs something else entirely. Collapse the five dimensions into one readiness score and you destroy the information that tells them apart.

So the sixteen items became AIT3: a five-character readiness code generated in the browser, with nothing transmitted or stored, and a trainer reference card mapping each code to a direction for the session.

Stated honestly: the tool is offered as a design artefact. Whether it produces better session-design decisions is an open question for future field evaluation.

PS

Swipe, or use ← →

The numbers

Dimension Before After d p
Ease of use 5.395.90 0.43.001
AI self-efficacy 5.676.06 0.33.011
Enjoyment 5.786.02 0.21.105
Intention to adopt 5.966.16 0.18.165
Usefulness 5.896.00 0.09.463
Pooled sample, seven-point scale, Welch's t-test with Mann-Whitney U as a non-parametric check. The chart above is drawn on a truncated 5.0 to 6.5 axis so the differences are visible; the full response scale runs from 1 to 7.

How it was measured

A sixteen-item questionnaire, adapted from an existing extension of the Technology Acceptance Model, rated on a seven-point scale and covering five dimensions: AI self-efficacy, perceived usefulness, perceived ease of use, perceived enjoyment and behavioural intention. Internal consistency was acceptable on every dimension at both timepoints, with Cronbach's alpha between 0.75 and 0.91.

The pooled sample was 80% female. Just under half reported nought to five years of teaching experience. Baseline readiness did not differ significantly by gender or by years of experience, so the dimensional pattern is not an artefact of who happened to attend.

What this study cannot tell you

Anonymity was preserved at both sites, which means responses could not be paired. Pre and post are treated as independent cross-sections, so the analysis estimates differences between groups rather than change within individuals. All analyses are exploratory and no correction for multiple comparisons was applied, so the p-values should be read in that light. The pooled sample is heavily concentrated in private-institution settings (106 of 122), which is why no school-type comparison was attempted here. And the confidence intervals around both significant effects are wide relative to their point estimates: a real but modest effect, not a precisely pinned-down one.

Citation

Shenoy, P., & Saarela, M. (2026). Lowering the barrier: A quantitative analysis of generative AI accessibility for teachers in India, diagnosing readiness before designing training. International Conference on Education Technology and Computers (ICETC 2026), Porto, Portugal. Accepted for presentation and publication.

Contact

Ask about any of this

Sessions and seminars on this work usually run to about ninety minutes. The shape of one is set after a training needs analysis with the school or organisation's own leadership, never before it.

Email prshenoy@jyu.fi University of Jyväskylä
WhatsApp Finland · UTC+2