Before teaching people to use AI, teach them to check

Mint (a fictional character) teaches a first-year foundation course. This term the reports look much better: tidy language, quotations from thinkers, each with a book title and year. But when she opened the books cited, she could not find the quotations anywhere.

This happens in real classrooms. The five studies brought together here help explain why, and suggest that what needs teaching may not be how to use the tools.

An answer that looks credible may not exist

A study by Assoc. Prof. Dr Pavel Slutskiy (published in 2025) gave 240 Chulalongkorn undergraduates the same task: find out what a particular philosopher thought on a topic and support it with a quotation from his work, using any online source. About one student in five chose a chatbot as the main source.

Thirty-seven submissions included a quotation attributed to the philosopher, with the title of the work and the year, just like a real citation. Not one of the quotations appears in the work cited, and all 37 students were confident they had quoted the original. The problem is not only that the tool was wrong. It is that the answer arrived looking as if it had already been checked.

What causes stress is quantity, not difficulty

A study by Asst. Prof. Dr Teerada Chongkolrattanaporn (published in 2023) surveyed 803 first-year students during fully online teaching. The form of technology stress that stood clearly above the rest was the sense of too many tools and tasks arriving at once. The sense that the technology was hard to use was fairly low.

Science students showed the same picture, with overload at 3.52 out of 5, above every other dimension. The remedy is therefore not more training in the tools, but fewer systems per programme and deadlines across courses that do not pile up together.

Quizzes help recall; applying knowledge comes from conversation

A study by Assoc. Prof. Dr Smith Boonchutima, Asst. Prof. Dr Teerada Chongkolrattanaporn and Dr Wassayut Kongjan (published in 2023) tried interactive media in an online master's course with 23 students. Mean scores rose from 6.77 to 8.20 after using the media, but repeating the same quiz added nothing and participation fell by about half. Students said the quizzes helped short-term memory, while applying the material to real work came from talking with the lecturer and classmates and getting direct feedback. They also asked for the slides as a plain PDF to annotate, even though a web presentation with videos was already provided.

Fluent use is not the same as discernment

A study by Dr Muhammad Ibtesam Mazahir and Yaseen (published in 2025) surveyed 500 adults in Pakistan who use media at least an hour a day. Only 41% could correctly identify biased or misleading news. Younger people handled digital tools more easily, but the interviews showed that many still could not tell which content was credible.

A collection by Giray and colleagues (published in 2024), co-written by Dr Robbie Jan Vincent T. Buelo, brought together the reflections of eight lecturers and administrators in the Philippines in the first months after ChatGPT launched. All of them used AI in real work, but they also met fabricated references and students submitting AI-written work as their own. At the time, none of the four institutions had a full AI policy. The authors agreed that prohibition would not work, and proposed separating AI that helps polish language or manage references from AI that produces the answer.

Four points for teachers and trainers

1. 1

   Teach people to go back to the source

   Have learners confirm every quotation and figure from AI against the original, and count the checking towards the grade. A book title and a year do not prove a quotation exists.

2. 2

   Say clearly which kinds of AI use are allowed

   Separate help with language or citation formatting from having AI write the answer, and ask learners to say what they used, rather than banning everything.

3. 3

   Cut the number of tools before adding training

   If learners are stressed by quantity, teaching more tools does not help. Use fewer systems and schedule deadlines across courses so they do not collide.

4. 4

   Use quizzes in moderation and add time to talk

   Quizzes help recall, but repeats lose value fast. Open live discussion at regular intervals, because applying knowledge comes from debate and the teacher's feedback.

Reading this evidence with care

The chatbot study used one task at one university, collected when students were using ChatGPT 3.5; tools that search the web and attach links may behave differently, so the source split is a snapshot of that time. The technostress study was conducted during fully online teaching, which is not a normal condition. The interactive media study had a single group of 23 students, no comparison group, and the lead researcher taught the course. The Philippine study is eight authors' own reflections, not a survey. The Pakistan study covers only people using media at least an hour a day, and its urban–rural difference carries no test statistic.

Tools change fast; the habit of checking lasts

AI and online tools will change many more times. What repays a teacher's investment is the habit of going back to the source, being open about which tools were used, and organising learning so it does not bury learners under so much that they have no time to think.

If your institution or organisation is setting guidelines for AI in learning, the Center of Excellence in Communication Innovation at Chulalongkorn University's Faculty of Communication Arts would be glad to talk and share what we have learned.