The United Nations Educational, Scientific and Cultural Organization (UNESCO)(1) has strived for years to ensure that scientific and technological development is governed by ethical principles, with the aim of establishing international regulations that enhance the benefits of scientific advances and, at the same time, reduce potential risks. Furthermore, it points out that it is urgent for countries that still lack regulations on the ethics of Artificial Intelligence (AI) to formulate and apply them on a priority basis.

Despite the growth in the use of AI in educational contexts, education professionals still lack clear guidelines for its implementation. According to 1 study conducted by UNESCO, only 10% of educational institutions formally monitor its use and only 40 national AI strategies addressing ethical aspects had been identified.(1)

Countries are developing AI training programs to incorporate AI objectives into their curricula.(2) This study arises in response to the international call to contribute to understanding the process of incorporating AI in educational environments and the construction of strategies that promote its ethical, inclusive, and pedagogically relevant use. The objective of this qualitative research is to identify recommendations for the ethical use of AI by university students. The knowledge generated will allow the identification of units of meaning with suggestions about AI for universities, within an international framework interpretation.

Study type. This was a qualitative investigation with a phenomenological design, which allowed exploring and understanding students' subjective experiences regarding the ethical use of AI in university contexts.

Participants. A total of 57 university students participated, selected through convenience sampling based on the resources and access possibilities available to the research team during an international stay, from 17 educational institutions across Latin America and Europe: Mexico, Brazil, Colombia, Argentina, Chile, and Italy. The following inclusion criteria were applied: a) enrolled in a university of any educational sector, and b) having used IA during their academic trajectory. Of the total participants, 64.9% identified as female (n = 37) and 35.1% as male (n = 20), ranging in age from 20 to 24 years.

Instrument. The written composition technique was used for qualitative narrative research, supported by a topic guide from which the participating students individually developed a topic.(16) After validation by three experts in education with a doctorate, the guide was incorporated into Google Forms. It consisted of two categories: a) ethical recommendations on the use of AI and b) sociodemographic data.

Procedure. The research protocol was approved by the private institution of the principal author of this study. The topic guide incorporated into Google Forms was shared with the students via WhatsApp, along with an invitation and agreement for their voluntary and confidential participation through informed consent.

Data analysis. The qualitative software MAXQDA2024 was used to conduct content analysis of the experiences narrated by the students, inductively and without predefined categories, to identify patterns of meaning. This was carried out in two stages according to Manig et al.(17): a) vertical analysis, in which narratives were organized and classified into codes and subcodes; and b) horizontal analysis, in which, using the software's visual tools, statistics of codes and subcodes were obtained, along with a visual representation of MAXMaps with the narratives on ethical recommendations in the use of AI to illustrate the coded segments, i.e., textual quotes from the students.

Ethical regulations and methodological rigor criteria were applied: a) qualitative credibility (18) in MAXMaps; b) data confidentiality (19); c) informed consent (20); and d) internal logical consistency (21). Data analysis was performed independently by the research team.

Five recommendations emerged, ordered by number of coded segments in the qualitative software, from highest to lowest: a) Use AI as support in performing activities (33%), b) Establish policies in the use of AI (24%), c) Use AI for writing (19%), d) Validate the information provided by AI (13%), and e) Train in the use of AI (10%) (Figure 1).

Figure 1. Ethical recommendations for the use of AI in academic activities.

Source: Authors' own elaboration using the qualitative software MAXQDA2024.

Figure 1 shows the recommendations most frequently mentioned by the students. It reflects the concerns arising from their experience interacting with AI. The main characteristics of each recommendation are synthesized below.

a) Use of IA as support in carrying out activities. This was the most outstanding result. Participants recommended using IA only as an auxiliary, not substituting their own intellectual work with results produced by IA. They referred to phrases such as 'only extra help,' 'don't let IA do everything,' and 'let it be an extension to broaden the approach to a topic.'

b) Establish policies on the use of IA. They repeatedly indicated setting limits on the use of IA, referring to time, even mentioning 3 times a week. They emphasized the importance of including citations, showing transparency in authorship, and avoiding copy and paste.

c) Use IA for writing. For writing, they repeatedly referred to paraphrasing based on the results that IA provides, specifying that any idea processed by IA must be generated from the student's own writing.

d) Validate the information provided by IA. This recommendation was among the least frequently mentioned. They recommended corroborating the information that IA provides, verifying the sources that are generated, and contrasting the information with other sources and other IA; they mentioned this to avoid being manipulated by IA and to promote critical thinking.

e) Training in the use of AI. This was the least frequently mentioned recommendation, encompassing the implementation of institutional courses that guide on its proper use, the development of strategies and techniques for ethical treatment of information, and the learning of basic principles on prompt formulation. They also referred to the faculty's mentoring or guiding role in the training process in the use of AI.

Figure 2 presents the five ethical recommendations that emerged, with examples of coded segments and textual narratives from the participating students.

Figure 2. Ethical recommendations for the use of AI and coded segments.

Source: Authors' own elaboration using the qualitative software MAXQDA2024.

The identification of ethical recommendations for the use of AI in academic activities, based on the experiences narrated by university students, yielded 5 recommendations. The following stood out with the greatest number of coded segments: using AI as support in carrying out activities, establishing policies in the use of AI, and using AI for writing. This indicates that students have assumed that AI's role in their academic environment is limited to being only a support for their learning. These findings are consistent with the human-centered approach to AI that UNESCO (7) has reiterated in its international recommendations. The recommendation to have policies on the use of AI perhaps shows that international, national, and any existing institutional frameworks are neither known nor adopted by some of the participants. In this regard, UNESCO (1) has been urging countries that still do not have regulations on AI ethics to articulate and apply them.

Student recommendations also show an acceptance of AI and are feasible to apply due to the technological acceleration of GenAI, which emerges year after year to facilitate learning experiences, including ChatGPT, Deepseek, Gemini, and Perplexity (22).

On the other hand, the less mentioned recommendations were: validating the information provided by IA and training in the use of IA. The UNESCO(6, 7) competency frameworks for teachers and students, in their purposes, refer to the critical judgment that people should exercise regarding IA solutions, especially the hallucinations that IA can produce, as Joshi (23) points out.

Thus, training is key to reducing the possible risks of AI. UNESCO (4) discusses the generation of foundational AI knowledge for lifelong learning and the design of inclusive and sustainable AI. However, since these recommendations are the least referenced, this implies that educational actors should recognize the importance of validating the information provided by AI, an aspect that should be present in training and professional development processes (24). However, it is the lack of training that is observed at this level (25).

The 5 recommendations identified in this study on the use of AI in academic activities, based on the experiences narrated by university students, are consistent with international frameworks and principles, which grants them validity for consideration in the academic community. However, some key recommendations did not emerge, such as those related to the inclusive approach, as well as those about AI being reliable in its algorithm and environmentally sustainable. 3 recommendations stood out prominently: using AI as support in carrying out activities, establishing policies in the use of AI, and using AI for writing. Despite the diversity of the participants' country of origin, it was possible to identify units of meaning with patterns that allowed an understanding of how the recommendations emerged.