The issue isn't whether to use AI in research. It's knowing how to use it.
)
Generative artificial intelligence is already part of study and research practices. Therefore, pretending that students do not use it does not promote academic integrity; it only leads to clandestine, unregulated, and unguided use. And, above all, it results in missed opportunities for genuine dialogue and learning on the subject.
On the other hand, normalizing any use of AI on the grounds that “AI is just a tool” does not solve the problem either. When a tool assists in an activity, is it stifling a skill that that activity is precisely meant to develop?
Two recent documents provide important references for this discussion: the American Historical Association’s (AHA) principles for teaching history and the guidelines from the Society for Science—which organizes the Regeneron ISEF—on the use of generative AI in science projects.
AI produces answers, not evidence
The AHA makes a fundamental distinction: artificial intelligence produces text, images, audio, and video, but it does not produce truths. Its results are generated by identifying patterns and may contain errors, biases, nonexistent references, and a false sense of certainty.
This is especially dangerous in research. A well-written text can conceal a claim that lacks evidence. A reference cited in academic format may simply not exist. A historically convincing image may depict a scene that never took place.
Therefore, AI should not be treated as a source. It can:
- Help identify opportunities;
- Organize questions;
- Present some initial approaches.
Therefore, every piece of information must be traced back to its original sources, the data, and the scientific literature.
The Difference Between Supporting and Replacing
The Regeneron ISEF table (based on the AHA text) makes this threshold more concrete.
| Activity | Acceptable use? | Terms and Conditions |
| Ask generative AI to identify or summarize the main points of an article before reading it, using the results as a starting point for a literature review. | Yes | Acceptable without an explicit statement. |
| Asking AI to summarize a book or article and including that summary in a literature review without reading the original work. |
No |
It is never acceptable, since there was no actual contact with the book or article. |
| Use an AI chatbot as a writing tool to help generate and develop ideas. |
Yes | An explicit statement may be required, depending on the circumstances. Prompts must be recorded in the research log. |
| Use generative AI to draft the research plan, abstract, article, or poster. |
No |
The initial draft must be the student's own independent work. Guidance or revisions after the original document has been produced may be permitted, provided that such assistance is acknowledged and documented. |
| Ask generative AI to write the abstract or a section of the article and submit the text as your own work. |
No | It is never acceptable. |
| Write the summary and ask the AI to refine the language without changing, adding, or replacing the main points. |
Yes | Acceptable without an explicit statement only when the changes are limited to grammar and syntax. |
| Use AI to write the initial code for the project. |
Yes |
It is necessary to explicitly state which parts of the code were generated by AI and to keep a record of the prompts. |
| Ask generative AI to create a flowchart, graph, or image for an article or presentation. |
Yes |
The material must be clearly identified as having been generated by AI, with an explicit statement explaining how it was created. |
| Write the paper and ask the AI to add new arguments or points to the text. |
No | It is never acceptable. |
| Use AI to generate conclusions, next steps, or project developments. |
No | It is never acceptable. |
| Use AI to help identify appropriate statistical tests or programs. |
Yes | The prompts must be recorded in the research journal. The data must be interpreted by the student researcher. |
| Use AI to collect data, write the research plan, and provide references that support the paper’s claims. |
No | It is never acceptable. |
| Ask the generative AI to generate an initial bibliography. |
No | It is never acceptable. |
| Ask the AI to correct the structure or formatting of the bibliography. |
Yes | Acceptable without an explicit statement, but all references must be reviewed and verified. |
Source: SOCIETY FOR SCIENCE. Use of Generative AI to Support a Research Project.
AI can be used to:
- Help develop ideas;
- Improve grammar and syntax;
- Suggest statistical tests;
- Assist with the initial code development.
However, the student must keep a record of the prompts, document the generated text, and personally take responsibility for interpreting the results.
On the other hand, it is not acceptable to use AI to write:
- Research plan;
- Abstract;
- Sections of the article;
- Conclusions;
- Future work;
- More points to add to the text;
- Results from in-house production.
The criterion goes beyond the tool used; rather, it defines the function to which the AI was applied—for example, whether the AI:
- It improves the expression of an idea developed by the researcher; this is a form of assistance;
- By formulating the hypothesis, interpreting the data, and writing the conclusion, she has taken the student's intellectual role.
See also:
- The next educational inequality will be inequality in the questions asked
- Law No. 15,436/2026: A New Approach for Highly Gifted and Talented Students
- When the Future Became Too Small
Recording usage is part of the method
An important aspect of the ISEF guidelines is the requirement that prompts be recorded in the logbook. This shifts transparency from a generic statement—"I used artificial intelligence"—to a verifiable record:
- Which tool was used;
- At what stage;
- What questions were asked;
- Which responses were used;
- What was reviewed, corrected, or rejected;
- Who made the final decision?
This record allows us to evaluate the process, not just the final product. In scientific research, it is not enough to present a correct answer; it is necessary to demonstrate how it was arrived at.
The rules depend on the context
The two documents are not identical. The AHA permits, subject to verification, the use of AI to generate an initial bibliography. The ISEF prohibits this practice in student projects. The difference is telling: there is no universal authorization to use AI. The rules must take into account the field of study, the age of the researchers, the educational objectives, and what is being evaluated.
At a science fair, the learning process and the student’s intellectual independence are part of the very purpose of the evaluation. For this reason, the criteria must be more rigorous.
Students must continue to be researchers
The key question is not “Was AI used?”, but rather: Can the student still explain and defend every decision made in the assignment?
He must understand the problem, read the sources, justify the method, interpret the data, recognize limitations, and support his conclusions. Artificial intelligence can enhance his ability to work, but it cannot replace the process through which he becomes capable of thinking scientifically.
Banning everything does not teach people how to use it. Allowing everything does not teach people how to research. Between these extremes lies a truly educational task: setting clear boundaries, demanding transparency, and ensuring that artificial intelligence remains a support tool—while authorship, judgment, and responsibility remain, and will always remain, human.
About the author:
*This text does not necessarily reflect the opinion of Bett Brasil.
Share on social media:
References:
SOCIETY FOR SCIENCE. Use of Generative AI to Support a Research Project. Regeneron ISEF, 2026.
Categories
- Learning Strategies
- Innovation