How to Teach Nursing Students to Use Generative AI Ethically

Some programs still treat generative AI as a student-conduct problem; others write policies as if every student will eventually use it. A descriptive qualitative study of 25 full-time nursing students at Chongqing Medical University, published in Frontiers in Medicine in September 2026, moves past that split by asking how students actually experience the tools.

The finding is dual: students report efficiency gains while worrying about information accuracy. That tension predicts where GenAI surfaces in the nursing education curriculum: studying, care plans, and evidence summaries. It also makes verification a skill nurse educators cannot treat as optional.

What the Chongqing Study of 25 Nursing Students Found

This study used semi-structured interviews, not a survey, to hear how 25 full-time nursing students at Chongqing Medical University described their own use of generative AI. The design is descriptive and qualitative: researchers used purposive sampling to select students, interviewed them between March and April 2025, and analyzed transcripts with conventional content analysis.

Study design in plain language

Wan and colleagues published the work in Frontiers in Medicine on September 30, 2026, Volume 13, DOI 10.3389/fmed.2026.1916153. The title is "Nursing students' experiences with using generative artificial intelligence: a descriptive qualitative study." The analysis produced three themes and ten subthemes. The publicly available excerpt confirms one theme as "Promoting Innovation in Teaching Methods," and the abstract elsewhere points to a perceptions-and-experiences theme with subthemes related to improving learning efficiency and concerns about information accuracy. The full wording of every theme and subtheme is not available in the excerpt used here, so educators should treat those examples as illustrative rather than a complete accounting.

Why student voice adds something different

Much of the earlier GenAI literature in nursing education measured acceptance, attitudes, and willingness to use tools. That work tends to quantify how open students are, not what they actually do. This study is different because the interview data center lived experience, or nursing student voice: participants explained where GenAI fit into their studying, how it changed their workflow, and what accuracy concerns they raised. For a nurse educator, that shift matters because it reveals the habits and tensions behind the answers early survey items often miss. Conventional content analysis allowed the researchers to group those accounts into themes, which gives educators concrete starting points for classroom discussion and assignment design.

Limits to keep in mind

The findings come from one institution, a purposive sample of 25 students, and self-reported interviews. Student accounts can be shaped by recall, social desirability, or how familiar they are with the tools, so the themes are starting points rather than universal truths. Those design choices are appropriate for exploring experience, but they do not support prevalence claims or generalizable rates. Nurse educators should read the themes as transferable ideas for conversation and curriculum review, not as estimates of how common any behavior is across nursing programs.

Where Nursing Students Turn to Genai: Studying, Care Plans and Evidence Summaries

Nursing students are not waiting for permission to use generative AI, and their use clusters around predictable tasks. The Chongqing study of 25 full-time adult nursing students reports a core tension: students describe efficiency gains in learning while also worrying about information accuracy. The published findings do not separate care-plan drafting from other uses in the available summary, so treat those examples as common classroom patterns rather than confirmed subthemes.

• Study support: students prompt AI to generate practice NCLEX-style questions, explain pathophysiology, or simplify drug mechanisms. Benefit: speed. Risk: plausible but incorrect rationales.
• Drafting care plans: AI drafts nursing diagnoses and interventions. Benefit: a starting structure. Risk: generic plans that skip individualized assessment.
• Summarizing evidence: students ask AI to condense articles or compare guidelines. Benefit: faster orientation to a topic. Risk: fabricated citations and missing nuance.
• Writing feedback: AI tightens case-study narratives or discussion posts. Benefit: clearer drafts. Risk: polishing can mask weak clinical judgment.

The opportunity is pedagogical: these uses mirror real workflows. The challenge is that unverified output enters clinical preparation quietly. Educators who address the efficiency-accuracy tension directly can shape better habits.

Why the Efficiency-Versus-Accuracy Tension Matters for Clinical Safety

The Chongqing interview study gives nurse educators a useful frame: students described GenAI as helping them study faster, yet also worried whether the information was accurate. That tension is a curriculum design problem, not an argument about accepting AI.

Efficiency can hide a verification gap

When students use GenAI to draft care plans or summarize evidence, the speed feels productive. But AI simulation in nursing education research shows outputs can omit guidance or hallucinate details. One review reported hallucination rates as high as 83% in clinical vignettes, and bibliographic reference generation over 25%.¹ A 2026 qualitative meta-synthesis of nursing students found low AI literacy and perceived overreliance reduced information evaluation. These are perceptions and simulated outputs, not proof of performance decline, but they show a predictable pattern: faster generation does not mean safer clinical reasoning.

The safety stakes are concrete

An unverified drug dose, a wrong nursing diagnosis, or a fabricated citation can move from a GenAI answer into a student's care plan. In simulated cases, GPT-4.1 omitted guideline information in 46% of hypercholesterolaemia outputs and 22% of type 2 diabetes outputs; guideline citation rates ranged from 0% to 78.39%. That variability means students cannot assume the tool catches what matters.

Teach verification as a clinical skill

Nurse educators should treat verifying GenAI output as a visible, graded step, just like checking a medication order, one of the evidence-based teaching strategies for nursing educators. Ask students to trace the drug dose, guideline, or citation to a current primary source and document that check. That moves AI use from an integrity rule to a patient-safety habit.

Assignment Templates That Build AI Literacy and Clinical Judgment

Recent qualitative work with 25 full-time nursing students reinforces a familiar pattern: students report efficiency gains while also worrying about GenAI accuracy. These five assignment templates shift grading toward verification and clinical reasoning, not polished AI output. Grade the verification trail itself: require the original AI draft, the correction or comparison, and an explanation of the clinical change.

Related Articles

What NLN, AACN, NCSBN and ANA Guidance Means for Your Program

Generative AI guidance in nursing education has shifted from whether to adopt it to how to govern it, but it still stops short of a single binding set of nursing program accreditation requirements. AACN's AI in Nursing Education Initiative frames governance, equity, curriculum integration, prompt engineering, and a nurse's duty to follow organizational policy and escalate safety or bias concerns.¹ NLN's AI Resource Center and Vision Statement add a teaching layer: foundational and advanced AI competencies, integration into a competency-based nursing curriculum, clinical reasoning, simulation, informatics, and ethics, and faculty development as the priority.²³ NCSBN and ANA have not published equally specific academic-integrity rules in the currently available 2026 materials, so local policy fills the gap.

Three implications for your program

• Policy: Create a school AI governance committee and require human oversight plus source verification, matching NLN integrity safeguards.
• Faculty development: Train faculty on AI literacy, assessment, grading, and curriculum mapping, because no national mandate sets required hours.
• Competency alignment: Map AI skills to AACN's 2026 Essentials⁴ and NLN Core Values, but do not claim accreditation requires AI implementation.

A Tiered AI Policy for Nursing Programs: Permitted, Restricted, Prohibited

Published nursing school policies sort GenAI use mainly by disclosure and attribution rather than by the tool itself. Purdue University School of Nursing's 2026 policy requires disclosure of GenAI-assisted work and addresses privacy, data security, intellectual property, accuracy, bias, and accountability. The table below uses explicit policy language from Rutgers School of Nursing, University of North Alabama, and University of South Carolina Aiken.

Privacy and PHI Rules for Students Using Public AI Tools

Can a nursing student paste a de-identified patient note into ChatGPT to draft a care plan? In most clinical placements, no. Public generative AI tools such as ChatGPT, Gemini, and OpenEvidence are not approved for identifiable patient information, and clinical faculty increasingly treat any real case material as a serious HIPAA exposure.

The non-negotiable rule

No patient identifiers, clinical notes, diagnostic images, patient photos, medical record numbers, dates of service, clinical placement details, or EHR screenshots go into public AI tools.¹² Stanford's Generative AI Policy for MD Students calls patient identifiers "strictly forbidden." De-identification requires more than deleting a name. Case material must meet institutional and HIPAA standards, and only the site privacy officer or faculty can say when a prompt is acceptable.

Quick do/don't list

• Do ask faculty or the site privacy officer before prompting with any clinical material.
• Do use an institutionally approved AI tool when PHI may be involved and permitted.³
• Don't type, photograph, or dictate patient names, MRNs, dates, notes, or images into public tools.¹
• Don't assume a student-created "anonymized" prompt is legally de-identified.

Prompt and data privacy

Prompts and uploaded coursework may be stored or used by the vendor, so students should check tool terms before entering essays, evaluations, or personal course data into any nursing education software. Public consumer chatbots are not HIPAA-compliant just because they process medical text.

Assessment Redesign and Student Disclosure Statements

One path treats AI-assisted work as an integrity violation to be detected; another treats it as a clinical reasoning exercise to be assessed. Nursing programs need the second path because detection alone does not teach safe use, and the Next Generation NCLEX rewards clinical judgment.

Assess the process, not just the final document

Process-based assessment asks students to submit drafting notes, prompt summaries, or an annotated source trail alongside the finished assignment, a practice reflected in nursing faculty online teaching strategies. Add a brief oral defense or an in-class clinical reasoning check for major papers and care plans. If a student cannot explain the decisions behind the final product, the work has not met the standard, regardless of how it was produced.

A disclosure statement that works

Ask students to complete three short statements: "I used [tool] to [purpose]. I revised [what I changed]. I verified [specific source or guideline]." For example: "I used ChatGPT to generate an initial care plan outline. I revised the assessment and interventions, and I verified each drug dose and citation against the course textbook and current guidelines. I did not enter patient identifiers."

Rubric criteria for AI-assisted work

• Verification evidence: The student shows how each AI generated claim was confirmed against a primary source.
• Accuracy of citations: Every citation resolves to a real, retrievable source.
• Clinical reasoning in the student's own words: The final summary and judgment are clearly independent of AI phrasing.

Where acceptable assistance becomes misconduct

Acceptable use is disclosed, verified, free of patient data, and defensible by the student. Misconduct shows up as undisclosed AI use, unverified content that passes as fact, or protected health information entered into a public prompt.

Testing AI for Bias and Handling Unsafe Outputs in Class

Bias testing has moved from a quality-assurance afterthought to a patient-safety expectation in nursing education, where addressing nursing education implicit bias is now a core concern. The Obermeyer et al. 2019 study of a commercial population-health algorithm showed how using predicted costs as a proxy for need caused the tool to underestimate illness among Black patients at the same risk score, a bias pattern with clear implications for nurse educators healthcare disparities. Correcting the proxy would have increased the proportion of Black patients selected for extra care by more than 50%.¹ Recent LLM research points the same direction: Jia et al. (PLOS One, 2026) found implicit bias in all 10 safety-aligned clinical LLMs tested, with mean IAT bias above zero. A debiasing prompt reduced GPT-4o bias by a mean of 0.262 (P = 0.026).²

A comparative classroom exercise

Have students run one patient scenario through a generative AI tool, varying only demographics such as race, sex, age, or insurance status. Ask them to compare each output against current clinical guidelines and highlight where recommendations shift without clinical justification. Chest radiograph underdiagnosis bias in underserved populations³ and race-based corrections in kidney function or pulmonary function interpretation⁴ offer real-world anchors for discussion.

Escalation path for unsafe AI output

• Stop using the output for any clinical or academic decision.
• Verify against a trusted source, such as a guideline, drug reference, or faculty-approved text.
• Document the prompt, the output, and the discrepancy.
• Tell the instructor or preceptor.

This routine makes equity and patient safety explicit parts of nursing ethics, not abstract AI policy. When students learn to interrogate outputs for differential risk, they practice the same clinical judgment required at the bedside.

A Step-by-Step Rollout Plan for Nurse Educators and MSN Students

A GenAI rollout fails when it is treated as a one-time policy update. The schools that get this right sequence change across curriculum, assessment, and faculty development, not just a syllabus paragraph.

Start with six ordered moves

• Audit current use. Ask students and clinical instructors what GenAI tasks they already perform, including study support, drafting, and summarizing.
• Adopt a tiered policy. Move from blanket bans to permitted, restricted, and prohibited tool uses tied to specific course activities.
• Redesign two assignments. Choose one clinical reasoning task and one evidence synthesis task, then make verification of AI output a graded step.
• Add disclosure and rubric language. Require students to state how they used GenAI and add a rubric line for source checking and clinical judgment.
• Train faculty. Match each change to nurse educator competencies in curriculum design, assessment, and the use of information technologies. This also gives instructors a shared language for evaluating AI use.
• Review each term. Revisit the policy after each cohort, using real student work and incident reports to adjust permitted uses.

Keep the first action small

Do not try to launch all six moves at once. Start this week by convening one faculty meeting to inventory where students already report using GenAI, then bring that list to Nursing Program Director Leadership for the tiered policy decision.

When you are ready to map these moves to competency-level planning, the AI-CSNE readiness article on nurseeducator.com is the next read.