Illustration: supervised small-group learning with technology in a medical school setting — not a photograph from the cited study.
Short version
A medical college in India tested a different way of studying: students regularly answered difficult questions from memory and used ChatGPT during supervised sessions with teachers.
A few months later, students performed better on complex exam-style questions. The biggest improvements were seen among students who had lower scores at the beginning.
How the study worked
The study lasted six months and involved third-year medical students.
Students were divided into four groups:
- two groups started the new learning sessions immediately;
- two groups started later.
This allowed researchers to compare results before and after the teaching method was introduced.
The learning module itself lasted four weeks. Once a week, students attended a two-hour session where they:
- answered challenging multiple-choice questions;
- focused on applying knowledge instead of memorising facts;
- used ChatGPT to discuss topics and get explanations;
- reviewed mistakes together with teachers.
Importantly, ChatGPT did not replace teachers. It was used as an additional learning tool.
What the researchers found
According to the study:
- almost all students completed the full program;
- average test scores improved over time;
- the improvement was still visible months later.
The most interesting result was that students with weaker initial scores improved the most.
This suggests the method may help reduce the gap between stronger and weaker students instead of only raising average scores.
Why this matters
The study does not show that “ChatGPT makes people smarter.”
Instead, it highlights the value of combining three things:
Retrieval practice Trying to remember information from memory helps learning more than simply rereading notes.
Discussion and explanation ChatGPT can quickly provide explanations and alternative ways to understand a topic.
Teacher supervision Teachers guided the process, corrected mistakes, and prevented students from relying entirely on AI-generated answers.
The combination of these elements is likely what produced the positive results.
Could this work outside medical education?
Probably yes.
The study suggests AI tools can support learning when students:
- actively think for themselves;
- regularly test their own knowledge;
- receive feedback;
- avoid using ChatGPT as a shortcut for ready-made answers.
This approach could also be useful for:
- language learning;
- exam preparation;
- programming education;
- technical and scientific subjects.
Important limitations
There are several limitations to keep in mind:
- the study was conducted in only one college;
- students were not randomly assigned individually;
- most publicly available details come from the abstract only;
- it is unclear whether the benefits would last a year or longer.
AI tools also change very quickly. A teaching workflow designed around one version of ChatGPT may need adjustments in the future.
Final thoughts
The study does not prove that ChatGPT can replace teachers or traditional education.
However, it does suggest something important: AI can improve learning when it is used as a guided support tool rather than a machine for instant answers.
The key seems to be active thinking, repeated recall, feedback, and discussion — not passive copying from AI.