What do students say they need from AI policy?
In a trilingual survey of 186 mostly Indonesian and Hungarian respondents, routine AI use coexisted with requests for funded access and clearer rules. The self-selected sample cannot measure national differences, learning gains or policy effectiveness.
The short version
3 key points
- 1.Among 186 respondents, 78.0% reported using AI several times a week or daily and 79.6% perceived an improvement in learning; these are self-reports, not usage logs or measured learning outcomes.
- 2.Institution-funded tool access was selected by 37.1% and clearer AI policies by 24.2%, while policy or training was the most commonly selected barrier at 31.2%.
- 3.The voluntary sample, recruitment routes, languages and response dates were confounded, so the survey cannot identify national effects or show which policy would work.
The answer: students asked first for access and clarity
In this exploratory survey, the most frequently selected institutional support was free institution-funded access to AI tools, chosen by 37.1% of 186 respondents. Clearer institutional AI policies came next at 24.2%. When asked about the biggest barrier to adoption, 31.2% selected educational policies or training, ahead of ethical, economic, cultural and technical barriers. The pattern suggests that institutions should examine whether students can access approved tools and understand acceptable use before assuming that another generic awareness campaign is the answer.
That conclusion must remain local and provisional. The sample was small, voluntary and mainly drawn from Indonesia and Hungary through different recruitment routes. The survey did not audit any institution's written policy or test a training programme. It measured what the respondents perceived and requested. For teachers and university leaders, its value lies in a practical set of questions to ask their own students—not a ranking of countries or proof that funded accounts and clearer rules will improve learning.[1]
What the 186 respondents reported
Seventy-eight per cent reported using AI several times per week or daily, with a 95% confidence interval from 71.5% to 83.3%. ChatGPT appeared in 180 of 186 responses, or 96.8%; Gemini or Bard was selected by 53.8%, Grammarly by 31.7%, DeepL by 31.2% and Claude by 18.3%. Multiple selections were permitted. The most common reported purposes were research or information seeking at 86.0%, problem solving at 71.0%, understanding concepts at 65.6%, writing assignments at 63.4% and translation at 54.8%.
Nearly four in five respondents, 79.6%, said AI improved their learning. That is a perception, not an examination result, assignment-quality measure or causal estimate. More frequent use had a modest positive bivariate association with perceived learning improvement, but both came from the same self-report questionnaire. The study cannot establish that using AI more often caused better learning, or even that respondents used the same tool or feature when answering questions about frequency and outcomes.[1]
How the trilingual survey was conducted
The researchers aligned a common set of questions across Indonesian, English and Hungarian versions. The instrument contained 41 substantive items covering background, access, use, ethics, perceived learning, support and future perspectives. Translation from English used native speakers, independent bilingual back-translation and reconciliation. However, some response formats and options still differed, and the authors did not establish measurement invariance—the statistical condition needed before assuming that a score carries the same meaning across language groups.
Responses were collected from 20 May to 6 November 2025. All 21 Hungarian-language responses arrived on one day through classes; the 30 English-language responses came through WhatsApp groups over two months; and 135 Indonesian-language responses came through WhatsApp over a much longer period. The cleaned dataset contained 186 people from eight countries, but 75.8% lived in Indonesia and 21.0% in Hungary, with only one respondent from each of six other countries. Just over half were undergraduates, 27.4% master's students, 16.7% doctoral students and 4.3% secondary students.[1]
The apparent country contrast is not a country effect
Professor encouragement was higher among Indonesian residents in the full sample, with a mean of 3.41 versus 2.18 for Hungarian residents and an absolute rank-biserial correlation of 0.549. Perceived learning improvement differed little. But residence, questionnaire language, recruitment route and response timing overlapped. All Hungarian-language respondents lived in Hungary, while 130 of 135 Indonesian-language respondents lived in Indonesia. Institutions and referral groups were not recorded, so the analysis cannot separate cultural, linguistic, institutional or recruitment explanations.
A sensitivity check comparing only 11 Indonesian and 14 Hungarian residents who chose the English questionnaire reduced the professor-encouragement difference and made it statistically uncertain. That does not prove language caused the larger full-sample contrast; choosing English could itself reflect unmeasured characteristics. The policy question also mixed degrees of permissiveness with an 'unclear/no policy' category, so its numerical coding cannot be read as a validated scale of policy quality. The safest claim is that the sampled students described different support climates, not that one country governs AI better.[1]
What teachers and institutions can do with the finding
Teachers can start by making permitted and prohibited uses concrete for the assignments they set. A useful policy should say when brainstorming, translation, feedback or drafting is allowed; what verification and disclosure are expected; and which data must not be entered into a tool. It should also explain how a student without a paid account can complete the work. The survey supports asking those questions because access and clarity were the most selected needs, but it does not identify the best wording or enforcement model.
Institutions should measure local conditions rather than import the reported percentages. A short anonymous audit can separate actual tool use, access barriers, perceived rules and requests for teaching support. Policy documents should then be tested with students and staff for comprehension across the languages used on campus. Assessment redesign, verification exercises and faculty development are plausible responses drawn from the wider literature, not interventions evaluated in this study. Success needs observable outcomes such as fewer unclear cases, more accurate citation and checking, equitable participation and maintained learning—not only higher satisfaction.[1]
Important evidence limits
Recruitment was non-probability and self-selecting, and response rates could not be calculated. The authors performed no prior power analysis. The 186 responses therefore cannot estimate prevalence among Indonesian or Hungarian students, let alone students globally. The instrument was not a validated technology-acceptance scale, several items differed across versions, and language-group equivalence was untested. Small subgroups made comparisons fragile. The broad word AI also covered general-purpose chatbots, translation tools and writing aids without tying reported outcomes to a particular product or task.
The design was cross-sectional and not preregistered. It cannot establish whether student practice developed before institutional policy, whether encouragement increased use or whether use changed learning. Self-reported improvement may reflect enthusiasm, convenience or confidence rather than mastered knowledge. The paper is unusually explicit about these constraints and applies multiplicity adjustments to the main exploratory comparisons, but statistical care cannot repair the sampling structure. The result is a descriptive snapshot of respondents reached in 2025, published on 15 October 2026.[1]
What would change the assessment
Confidence would increase with probability-based or clearly bounded institution-level sampling, recorded recruitment denominators and balanced data collection across languages and time. Researchers should establish measurement invariance before comparing translated scores and link students to audited institutional policies without compromising privacy. Repeated surveys could show whether policy clarity and use change together. Trials could then compare specific supports—funded access, assignment-level guidance, verification training or faculty development—using measured learning, equity, integrity and workload outcomes.
For now, the study's best contribution is not the 78% usage headline. It is the warning that widespread practice can coexist with uncertain support and that apparent cross-country differences may be artefacts of who answered, in which language and through which channel. Educators can act on that by checking their own evidence, publishing plain-language rules and testing whether students can apply them. They should not claim that this survey proves learning gains or supplies a national benchmark.[1]
Funding, disclosure and independence
The authors reported no specific funding and no competing interests. They disclosed that generative-AI tools assisted with manuscript revision, language editing, code review, numerical checking and figure preparation, while stating that the tools did not generate participant responses and that the authors remained responsible for analysis and verification. The relevant Indonesian institutional authority waived formal ethics committee review for the anonymous minimal-risk survey, and participants gave informed consent.
Those disclosures help readers evaluate provenance but do not validate the survey. The practical takeaway remains bounded: among a small, mostly Indonesian and Hungarian convenience sample, frequent reported use sat beside demand for institution-funded access and clearer rules. That is credible as a prompt for local investigation and policy design. It is not evidence that students everywhere use AI at the same rate, that AI improved learning or that a particular governance model works.[1]
What this means for people
- Teachers need assignment-level examples of acceptable use and verification, not only a generic institutional statement.
- Students without paid access can be disadvantaged when coursework quietly assumes premium AI tools, making an explicit access policy important.
- University leaders should test whether guidance is understood across campus languages and measure learning and equity outcomes before calling a policy successful.
Global context
The survey included three questionnaire languages and respondents in eight countries, but 96.8% of the sample lived in Indonesia or Hungary. It broadens the discussion beyond English-only settings while remaining unsuitable for national or global prevalence estimates.
What the evidence does not yet show
- Only 186 self-selected respondents, mainly in Indonesia and Hungary, with no calculable response rate
- Residence, questionnaire language, recruitment route and response timing were heavily confounded
- No established measurement invariance across the three language versions
- Cross-sectional self-report data cannot show learning gains, causality or policy effectiveness
- Broad AI categories and no link between each reported outcome and a specific tool or use
- Small English-language sensitivity groups and several very small academic subgroups
What to watch next
- Institution-level replication with known recruitment denominators and balanced multilingual samples
- Independent validation of translated measures before cross-context comparisons
- Trials of funded access, clear assignment rules and verification training with measured learning and equity outcomes
- Longitudinal evidence separating changes in student use from changes in policy and teaching practice
Living evidence record
Impact record IAI-11NPYIA
Evidence stage
Studied
Confidence
Supported
Reporting basis
Source analysis
Independent or research support
Present
Record status
Monitoring
Last checked
15 October 2026
Source trail
1 direct source across 1 source type.
People impact
Documented in this record.
Uncertainty
Limits and next checks are explicit.
Stages describe the evidence available—not whether a technology is good or bad. See the public method.
Single-source reporting disclosure
This record analyses one direct source. It can establish what Discover Education published or reported, but it is not independent corroboration of every performance claim or predicted outcome. The confidence label will change only when broader evidence is added.
Evidence trail
Sources used for this report
Links checked 15 October 2026
Research question, topics and regions
How self-selected students reported generative-AI use, perceived learning, institutional encouragement and support needs across three questionnaire languages
Topics: Higher education · Generative AI · Student support · AI governance · Access · Multilingual research
Regions: Asia · Europe · Indonesia · Hungary
This report is labelled source analysis. We summarise and analyse source material in our own words; company statements remain attributed claims until independently supported. Translated summaries preserve the meaning of the original source and link back to it. Read our editorial standards.
Watch the background, then check the evidence
UNESCO
AI-generated illustration · not a video frame
YouTube loads only when you choose. Press play in the player to start. If playback is unavailable, use “Watch on YouTube” below.
Education & human agency · Background explainer
Published 2025-01-23 · 8 min 30 sec
Teacher to Teacher: Is AI reshaping education for the better?
A conversation about how teachers can guide AI use while preserving pupils’ judgement. Watch for the difference between helping a learner think and supplying an answer.
Evidence check: Teachers’ perspectives provide context. They do not establish that a particular AI tool improves learning outcomes.
Source checked 14 October 2026. Check available captions on the publisher’s video page.
The Impact Brief · Free
Keep following the stories that matter to you.
Get our weekday briefing on what changed and why it matters. Choose your topics below.
Continue the story
There’s more to this story
Education
Does generative AI improve self-regulated learning?
A meta-analysis of 17 comparative studies found a positive average effect, but results varied sharply: heterogeneity reached 94.85% and the prediction interval included harm as well as benefit. Almost all evidence came from higher education.
7 min · 1 source
Education
What do 394 higher-education studies show about generative AI?
A peer-reviewed scoping review finds that universities mostly used generative AI for feedback, tutoring and writing. The evidence is substantial but still dominated by surveys and short-term measures, with little attention to equity or learning processes.
10 min · 1 source
Education
Does open AI access still let teachers see student judgement?
In one 33-manuscript university course, structural writing clustered near the top while evidence use and theoretical interpretation still varied. The case supports assessment redesign, not a claim that AI improved learning.
7 min · 1 source
Reader commentary
What does this mean where you are?
Tell us what you have seen at work, in your community or in everyday life. A question is welcome too. No account is needed; we review notes before publishing them.
Explore commentary across the portal →Published reader notes
0No published reader notes yet. You can start the evidence-led discussion above.
Prefer a private correction or response? Contact the newsroom.