Verify everything it produces. Do not feed it sensitive information.
As artificial intelligence grows more capable of mimicking scientific reasoning, two Spanish research groups have stepped forward with a framework for keeping human judgment at the center of nursing science. Their ten principles, published in Enfermería Clínica, do not reject the technology — they insist that its speed and fluency must never be mistaken for wisdom. In a field where the texture of human experience is itself the subject of inquiry, the stakes of getting this wrong extend well beyond methodology.
- Generative AI can fabricate references, invent data, and produce authoritative-sounding text that contains no truth — a vulnerability researchers call 'hallucination' that demands constant, active scrutiny.
- Nursing research faces a particular danger: AI systems trained on published text may flatten the complexity of lived experience, erasing therapeutic relationships, social determinants, and culturally specific perspectives in favor of smooth but hollow summaries.
- Researchers from Fisabio and Universitat Jaume I have responded with ten practical guidelines — verify everything, protect sensitive data, document AI use openly, and deploy the tools only where they genuinely strengthen human work.
- The framework aligns with the new European AI Act and data protection regulations, treating governance not as red tape but as the structural guarantee that research remains honest and the people behind the data remain protected.
- The trajectory is clear: AI is entering scientific practice whether frameworks exist or not, and this work argues that the only responsible path is integration with vigilance — keeping researchers, not algorithms, in the role of thinker and decision-maker.
Two Spanish research groups — the Fisabio Foundation and Universitat Jaume I of Castelló — have published ten principles for using generative AI responsibly in nursing research, addressing a tension that will only intensify as these tools become more capable and more tempting.
The appeal is genuine. AI can accelerate the most laborious stages of research: formulating questions, reviewing literature, organizing findings, preparing publications. But the researchers found themselves confronting a specific and serious vulnerability. These systems can generate text that reads like rigorous science — coherent, complete, authoritative — while containing fabricated references, invented data, or interpretations that do not exist. This phenomenon, known as hallucination, is not a minor flaw. It is a structural risk that demands continuous verification of everything the technology produces.
The danger runs deeper in nursing research than in many other fields. Nurses study how patients experience illness, how professionals navigate ethical dilemmas, how social conditions shape health outcomes. These are qualitative, context-dependent phenomena. An AI system may produce summaries that sound plausible while quietly erasing the therapeutic relationship, ignoring social determinants of health, or defaulting to biomedical frameworks and English-language sources at the expense of nursing-specific and culturally particular perspectives.
The team's recommendations are practical without being naive: use AI only where it genuinely improves human work, verify all outputs, protect sensitive data, preserve the full complexity of what is being studied, and document openly how the tools were used. The framework also situates AI use within the new European AI Act and existing data protection regulations — not as bureaucratic compliance, but as the architecture that keeps research honest.
What the researchers ultimately propose is not a prohibition but a discipline: AI may serve nursing science, but only if nurses and nursing scientists remain the ones doing the thinking, the questioning, and the deciding. Rigor governs the technology — not the other way around.
Two research groups in Spain have spent months wrestling with a question that will only grow more urgent: How do you let artificial intelligence help with scientific work without letting it break the work itself?
Scientists from the Fisabio Foundation and Universitat Jaume I of Castelló have now published their answer—a set of ten principles for using generative AI responsibly in nursing research. The work, published in Enfermería Clínica, emerged from researchers at the university's nursing research group and Fisabio's eNURSYS team, operating within a joint research unit focused on information systems, care technologies, and the measurement of nursing quality.
The appeal of these tools is real. Generative AI can accelerate the grinding parts of research: formulating questions, sifting through literature, organizing findings, preparing results for publication. But the researchers found themselves staring at a particular kind of danger. These systems can produce text that reads like rigorous science—coherent, authoritative, complete—while containing fabricated references, invented data, or interpretations that simply do not exist. The field calls this "hallucination," and it is not a minor glitch. It is a fundamental vulnerability that demands constant vigilance.
The risk cuts deeper in nursing research than it might in other fields. Much of what nurses study cannot be reduced to numbers or biomedical categories. They investigate how patients experience illness, how professionals navigate ethical dilemmas, how social conditions shape health outcomes. These are qualitative phenomena, rooted in context and complexity. An AI system, trained on vast quantities of published text, may generate summaries that sound plausible and read smoothly while actually flattening the texture of human experience. It might miss the therapeutic relationship entirely. It might overlook the social determinants that shape a patient's reality. It might default to biomedical frameworks and English-language sources while ignoring perspectives specific to nursing or to particular cultural contexts.
Beyond these methodological hazards lie governance questions. The researchers emphasize that AI use in science must operate within frameworks of transparency and accountability. Data protection matters. The new European Artificial Intelligence Act matters. These are not bureaucratic afterthoughts—they are the scaffolding that keeps research honest and protects the people whose data makes research possible.
The ten recommendations the team proposes are practical but not naive. Use AI only when it genuinely improves human work, not as a shortcut. Verify everything it produces. Do not feed it sensitive information. Preserve the full complexity of what you are studying. Document openly that you used these tools and how. Keep measuring whether they are actually making your science better or just faster.
What emerges from this work is not a prohibition on AI in nursing research. It is something more useful: a framework for integration that does not pretend the technology is neutral or that speed equals quality. The researchers are saying that nurses and nursing scientists can use these tools—but only if they remain the ones doing the thinking, the questioning, the deciding. The technology serves the rigor. The rigor does not serve the technology.
Notable Quotes
Generative AI systems can produce seemingly rigorous texts that contain errors, nonexistent references or inaccurate interpretations— The research team, published in Enfermería Clínica
AI may generate coherent summaries or plausible interpretations that nevertheless oversimplify the complexity of patients' and professionals' experiences— The researchers, on risks specific to nursing research
The Hearth Conversation Another angle on the story
Why does nursing research need its own guidelines? Couldn't researchers just apply general AI best practices?
Because nursing is built on understanding human experience in all its messiness. A biomedical AI trained on published literature might generate something that sounds right but misses the therapeutic relationship, the social context, the things that actually matter to patients and nurses.
You mean the AI could oversimplify?
Exactly. It could produce a coherent summary that strips away the complexity—the very thing nursing research is trying to preserve and understand.
What about the hallucination problem? How bad is that really?
Bad enough that you cannot trust anything an AI generates without checking it. It will invent references that do not exist, misquote sources, create plausible-sounding interpretations that are simply false. In research, that is not acceptable.
So the guidelines are basically saying: use AI, but do not trust it?
More like: use AI when it genuinely helps, but verify everything, document what you did, and never let speed compromise the thinking that makes research worth doing.
What about bias? Can AI reproduce the biases in existing research?
Yes. If the training data is skewed toward biomedical approaches or English-language sources, the AI will amplify those biases. It will not naturally seek out nursing perspectives or culturally specific knowledge.
So this is really about keeping humans in control?
It is about keeping humans responsible. The technology is a tool. The researcher has to remain the one asking the questions, making the judgments, and answering for the work.