Information Quality Requirements for AI-Generated Patient Education Materials in Diabetic Foot Ulcer Care: A Directed Qualitative Content Analysis of Patient and Nurse Perspectives

Abstract

Purpose

To explore the information quality requirements of artificial intelligence generated patient education materials for diabetic foot ulcers from the perspectives of patients and nurses.

Patients and Methods

This single-center qualitative study was conducted at the Fifth Affiliated Hospital of Guangxi Medical University, China, from November 2025 to June 2026. Maximum-variation purposive sampling recruited 15 hospitalized patients with diabetic foot ulcers and 10 nurses. Saturation was assessed separately for each group and defined as no new codes, categories, or relevant insights. It was reached after 13 patient and 8 nurse interviews, with two additional interviews per group for confirmation. Face-to-face semi-structured interviews used standardized artificial intelligence generated diabetic foot ulcer education material as a discussion stimulus. Data were analyzed using directed content analysis guided by Wang and Strong’s information quality framework.

Results

Five themes and 17 subthemes were identified. Four themes mapped to Wang and Strong’s framework: intrinsic information quality (accuracy, believability, objectivity, and reputation); contextual information quality (relevancy, completeness, timeliness, appropriate amount of data, and value-added); representational information quality (ease of understanding, interpretability, consistent representation, and concise representation); and accessibility information quality (accessibility and access security). An additional specific theme comprised clinical safety and actionability. Both groups valued reliable, relevant, understandable, and accessible information, but their priorities differed: nurses emphasized clinical accuracy, professional review, and the risks of unsafe recommendations, whereas patients prioritized relevance to current wound-care needs, ease of understanding and access, and clear guidance for self-care.

Conclusion

Patients and nurses showed different but complementary priorities. Clinical safety and actionability extend conventional information quality considerations. Artificial intelligence may support educational content generation. However, a professional review is necessary for clinically sensitive information, and patient input is essential to ensure usability. Future studies should refine and validate these requirements across settings and user groups.

Introduction

According to the 11^th edition of the International Diabetes Federation Diabetes Atlas, approximately 148.0 million adults aged 20–79 years in China were living with diabetes in 2024, representing the largest number worldwide.¹ Diabetes-related foot ulceration is a major complication of diabetes, with an estimated lifetime incidence of 19–34% and an annual incidence of approximately 2%. Among patients whose diabetes-related foot ulcers have healed, recurrence occurs in approximately 40% within 1 year and 65% within 3 years.² A 2024 systematic review and meta-analysis reported a pooled amputation incidence of 22.4% (95% CI, 18.3–26.5%) among Chinese patients with diabetic foot ulcers (DFUs).³ Mortality is also substantial in this population, with estimated survival rates of 86.9% at 1 year and 50.9% at 5 years.⁴

Health education is a key part of diabetic foot ulcer management. Its quality directly affects patients’ treatment, rehabilitation outcomes, and readmission rates. Health education refers to educational and guiding instructions or materials provided by nurses/medical teams to patients and their families. Common content includes medication management, diet and exercise, blood sugar monitoring, hypoglycemia/hyperglycemia response, foot care, and follow-up visits.^5,6 High-quality health education materials can assist patients and their families in understanding disease risks, mastering foot care skills, promptly identifying abnormalities, and maintaining long-term self-management capabilities.

Artificial intelligence (AI) technology has advanced rapidly in education.⁷ AI increasingly being applied in scenarios such as health consultation, patient education, clinical decision support, and chronic disease management. Generative AI holds immense promise for delivering health information and assistance to patients. Several studies have demonstrated the use of generative AI-generated health information materials in educating patients with various diseases.^8–10

However, some research findings suggest that these materials still have shortcomings in terms of comprehensiveness, readability, and evidence-based application.^11,12 For patients with diabetic foot, the quality of health education materials may influence their ability to identify foot-related risks and engage in self-care behaviors. It is important to further clarify the information quality characteristics and standards of such materials.

Currently, information quality in AI-generated health education materials for diabetic foot ulcers has been evaluated predominantly using quantitative methods, including expert ratings, comparisons with clinical guidelines, assessments of accuracy and completeness, and readability measures.^10,13 Song’s scoping review found substantial variation across studies in the AI models, prompt designs, evaluation instruments, and quality criteria used.¹⁴ These studies have mainly assessed the quality of the AI-generated text itself, with less attention to whether patients can understand and apply the information.

Existing qualitative studies have shown that the information needs of patients with diabetic foot problems extend beyond the provision of factual knowledge. Hill et al reported areas of agreement regarding the importance of education, but also identified tensions between patients and healthcare professionals concerning perceived risk, responsibility for self-care, and healthcare delivery.¹⁵ Studies focusing on patients’ experiences further found that limited understanding of warning signs, physical limitations, low health literacy, inadequate family support, and ineffective communication could prevent patients from applying health information in practice.^16,17 These studies provide important insight into the content and delivery of diabetic foot education; however, they mainly examined conventional education and self-management rather than the quality of specific educational materials. Moreover, they did not systematically compare how patients and nurses prioritize different information-quality attributes.

Because existing evaluations have mainly focused on isolated attributes such as accuracy, completeness, safety, or readability, a broader framework is needed to integrate the different dimensions of information quality. Wang and Strong’s information quality theory provides such a framework.¹⁸ From the perspective of information users, the theory distinguishes four dimensions of information quality: intrinsic, contextual, representational, and accessibility quality. In relation to AI-generated DFU education, intrinsic quality concerns the accuracy, credibility, and non-misleading nature of the information; contextual quality concerns its relevance, completeness, and timeliness in relation to the patient’s clinical situation; representational quality concerns clarity, conciseness, consistency, and understandability; and accessibility quality concerns the ease, safety, and convenience of obtaining and using the information.

Although the framework was not developed specifically for generative-AI patient education, its multidimensional and user-centered structure provides a useful basis for evaluating information quality beyond textual accuracy alone. Therefore, this study used semi-structured interviews guided by Wang and Strong’s framework to explore the information-quality needs of patients with DFUs and nurses involved in their education.

Materials and Methods

This study is reported in accordance with standardized guidelines for qualitative research (consolidated criteria for reporting qualitative research, COREQ).¹⁹ This study adopted a directed content analysis approach guided by Wang and Strong’s information quality framework.¹⁸

This study used a locally deployed Qwen3-32B model with Q4_K_M quantization (qwen3:32b-q4_K_M) implemented at the Fifth Affiliated Hospital of Guangxi Medical University. This model was served through Ollama (version 0.20.2) and integrated into a retrieval-augmented generation (RAG) application developed using an open-source large language model application-development platform (Dify version 1.4.2). In this system the research team established a local knowledge base comprising 34 evidence-based documents relevant to diabetic foot ulcer care and patient education. These documents were selected through discussion and consensus among the research team. Prompts were then developed around common patient education needs, such as “Please generate daily foot-inspection guidance for patients with diabetic foot ulcers based on the uploaded materials” “Please explain wound-care precautions in language that patients can easily understand” and “Which situations require patients to seek immediate medical attention”. The model primarily generated text-based educational content. Then the research team integrated the relevant content into a single draft covering the key domains of diabetic foot ulcer self-care, including daily foot inspection, wound care, off-loading measures, recognition of abnormal signs, and indications for seeking medical attention. Then the research team reviewed the draft for accuracy, clarity, and suitability. Following this process, a single standardized version of the AI-generated patient education material was finalized. The final version was presented to all patient and nurse participants. The standardized material served as a common stimulus to facilitate discussion and enable participants to reflect on desired characteristics, limitations, and improvement needs of AI-generated patient education materials. The interview outline was developed based on the four dimensions of Wang and Strong’s information-quality theory and included open-ended questions concerning the intrinsic, contextual, representational, and accessibility quality of the material, while allowing participants to raise information needs beyond the predefined framework.

Sampling and Recruitment

This single-center qualitative study was conducted at the Fifth Affiliated Hospital of Guangxi Medical University between November 2025 and June 2026. Maximum-variation purposive sampling was used to recruit two participant groups: 15 patients with diabetic foot ulcers and 10 nurses involved in diabetic foot care and patient education.

Potentially eligible participants in both groups were identified by the researcher according to the predefined inclusion and exclusion criteria. The researcher then approached eligible participants directly, explained the study purpose and procedures, and invited them to participate. Treating nurses were not involved in recruiting patient participants. No patient or nurse who was approached declined participation.

For patients, the researcher reviewed the characteristics of participants already enrolled and sought subsequent eligible patients who could broaden the range of age, sex, educational level, and ulcer history represented in the sample. All patient participants were recruited from the inpatient wards of the study hospital. Prior exposure to AI-generated patient education materials and the frequency of such exposure were also recorded as study-relevant participant characteristics.

For nurses, variation was sought across clinical specialty, years of professional experience, and professional title. Nurse participants were recruited from clinical services involved in DFU care and education, including diabetes care, wound care, chronic disease management, and geriatric nursing. Their duration and frequency of use of AI-generated patient education materials were also recorded as characteristics relevant to the study topic.

Inclusion and Exclusion Criteria for Nurses

Inclusion criteria for nurses: (1) Registered nurses who are officially employed in the hospitals; (2) Responsible for clinical nursing and health education of patients with diabetic foot ulcers in relevant wards or outpatient units such as endocrinology/wound stoma/geriatrics; (3) Nurses have used health education materials generated by generative AI to provide health education to patients in the past year; (4) Voluntarily participate in this study.

Exclusion criteria included (1) Training or rotation nurses; (2) Those who did not work for six months due to a long vacation, further study, or other reasons.

Inclusion Criteria and Exclusion Criteria for Patients

Inclusion criteria for patients: (1) meets the World Health Organization diagnostic criteria for diabetes;²⁰ (2) presents with symptoms, signs, and clinical manifestations of diabetic foot ulcers, meeting the Wagner classification criteria;²¹ (3) Received guidance from nurses on health education materials generated by generative AI in the past year; (4) Has the ability to give informed consent and sign the consent form; (5) Able to communicate in Mandarin or local dialect. Patients were excluded if they: (1) had severe cardiopulmonary, renal, or other medical conditions that could prevent them from participating safely in an interview; (2) had cognitive impairment, altered consciousness, or severe communication difficulties that prevented them from completing the interview; or (3) were clinically unstable at the time of recruitment or interview.

All 15 patient participants were recruited from the inpatient wards of the study hospital. The 10 nurse participants were recruited from relevant clinical nursing services within the same hospital group, including diabetes care (n = 5), wound care (n = 2), chronic disease management (n = 2), and geriatric nursing (n = 1). To safeguard participant privacy, real names were replaced with codes ranging from “N1-N10” and “P1-P15”.

No participants withdrew during the study period.

Interview Outline

Wang¹⁸ reported the information quality can be divided into four categories from the user’s perspective: intrinsic information quality (Intrinsic IQ), contextual information quality (Contextual IQ), accessibility information quality (Accessibility IQ), and representational information quality (Representational IQ). Based on information quality theory, we consulted three diabetic foot experts to design a preliminary interview outline. After conducting pre-interviews with two patients, we finalized the interview outline. The interview outline details are in Figure 1, Semi-structured interview outline for patients and nurses.

Figure 1.

Semi-structured interview outline for patients and nurses.

Data Collection

The interviews were conducted by a researcher who had received qualitative research training, was skilled in interviewing techniques. The researcher obtained a master’s degree in Nursing. She is a current nursing staff member at the hospital. The researcher was amiable, articulate, and possessed strong interpersonal and communication skills. Before the interview, the researcher stated the purpose of the interview, arranged an interview time with the interviewee, and obtained their consent before signing an informed consent form. The face-to-face, semi-structured interviews were conducted in the diabetic foot activity room of hospital. No one other than the researcher and participants was present. The Researcher used a voice recorder to record the whole interview, which lasted between 20 and 40 minutes. Detailed notes were taken during the interviews regarding tone of voice, mannerisms, and facial expressions that could not be conveyed verbally, and these were incorporated into the later data processing. A trusting relationship was established with the researcher, ensuring the smooth and continuous conduct of the interviews. No interviewees dropped out during the interviews. No repeat interviews were conducted; each participant was interviewed once.

Data collection and preliminary analysis were conducted concurrently. Data saturation was assessed separately for the patient and nurse groups and was considered reached when additional interviews yielded no new codes, categories, or relevant insights. Saturation was reached after 13 patient interviews and 8 nurse interviews. Two additional interviews were conducted in each group to confirm saturation, resulting in final samples of 15 patients and 10 nurses.

Data Analysis

This study employed the directed content analysis approach proposed by Hsieh and Shannon to analyze data obtained from interviews with nurses and patients.²² Wang and Strong’s information quality framework was used as the initial coding framework. The four dimensions of information quality (intrinsic, contextual, representational, and accessibility quality) served as a priori coding categories before data analysis. The deductive mapping process between initial codes and the four information quality dimensions is presented in Supplementary Table 1.

All interview transcripts were imported into NVivo 15 for data management and coding. In the first round of coding, three researchers independently read the transcripts, identified meaningful units, and assigned codes. Codes that were conceptually consistent with the four predefined dimensions were mapped to the relevant dimension. This was based on their conceptual meaning and not on specific terminology. Then, within each predefined dimension, similar codes were compared, categorized and grouped into subthemes and overarching themes. Any disagreements during coding and categorization were resolved through discussion to reach consensus.

The research team actively sought, and was open to data that were not represented by Wang and Strong’s original framework or contradicted emerging patterns. Additional codes were created for new or diverging ideas that emerged during analysis and discussed among the research team. Emerging codes were incorporated into existing categories when conceptual consistency was found, or left as separate concepts that represented the unique attributes of AI-generated health education materials.

The detailed coding matrix including the mapping process from initial codes to subthemes and themes, is provided in Supplementary Table 2.

Quality Control

The present study utilized investigator triangulation. Three researchers independently coded and analyzed the same data set. As a team, they talked through their differences and eventually reached an agreement. Following initial analysis, member checking was completed by the research team. Researchers sent each of the 15 patients and 10 nurses a copy of the interview summary and a preliminary interpretation of their main views through WeChat. All participants confirmed that the summaries reflected their views accurately and suggested no changes. Field notes and reflective memos were used to make adjustments to interview approaches, which improved the authenticity and accuracy of the data.

Results

The final sample comprised 15 patients with diabetic foot ulcers and 10 nurses. Patients included 8 men (53.3%) and 7 women (46.7%), ranged in age from 40 to ≥70 years, and had educational levels ranging from primary school or below to college or above. Nine (60.0%) were experiencing a first ulcer episode and 6 (40.0%) had recurrent ulcers; 10 (66.7%) had been exposed to AI-generated patient education materials only once in the preceding 12 months. The demographic and clinical characteristics of patients were presented in Table 1. Among nurses, 8 (80.0%) were women. Professional experience ranged from <10 to ≥20 years and titles ranged from staff nurse to chief nurse. Diabetes care (50.0%), wound care (20.0%), chronic disease management (20.0%), and geriatric nursing (10.0%) were represented. Seven (70.0%) had used AI-generated patient education materials for <6 months and 8 (80.0%) used them less than once per month. The demographic and professional characteristics of nurses were presented in Table 2.

Table 1.

Demographic and Clinical Characteristics of Patients (N = 15)

Characteristics	Categories	n (%)
Gender	Male	8 (53.3)
	Female	7 (46.7)
Age	40–59 years	8 (53.3)
	60–69 years	5 (33.3)
	≥70 years	2 (13.3)
Education level	Primary school or below	3 (20.0)
	Secondary school (technical secondary school/high school)	6 (40.0)
	College degree or above	6 (40.0)
Place of residence	Rural area	2 (13.3)
	Urban/town area	13 (86.7)
Foot ulcer status	First foot ulcer episode	9 (60.0)
	Recurrent infection	6 (40.0)
Time since most recent exposure	≤1 month	8 (53.3)
	>1–3 months	4 (26.7)
	>3–6 months	1 (6.7)
	>6–12 months	2 (13.3)
Exposure frequency in the preceding 12 months	Once	10 (66.7)
	2–3 occasions	4 (26.7)
	≥4 occasions	1 (6.6)

Table 2.

Demographic and Professional Characteristics of Nurses (N = 10)

Characteristics	Categories	n (%)
Age	21–29 years	1 (10.0)
	30–39 years	7 (70.0)
	≥40 years	2 (20.0)
Gender	Female	8 (80.0)
	Male	2 (20.0)
Professional title	Nurse	2 (20.0)
	Nurse-in-charge	6 (60.0)
	Associate Chief Nurse	1 (10.0)
	Chief Nurse	1 (10.0)
Years of work experience	<10 years	2 (20.0)
	10–19 years	6 (60.0)
	≥20 years	2 (20.0)
Specialty	Diabetes care	5 (50.0)
	Wound care	2 (20.0)
	Chronic disease management nursing	2 (20.0)
	Geriatric nursing	1 (10.0)
Duration of use of AI-generated health education materials	<6 months	7(70.0)
	6–12 months	3(30.0)
Frequency of use during the preceding 12 months	Less than once per month	8(80.0)
	1–3 times per month	2(20.0)

The analysis identified five overarching themes, 17 subthemes and 38 initial codes concerning the information-quality requirements of patient education materials for DFUs generated by AI. Four themes corresponded to Wang and Strong’s information-quality framework: intrinsic, contextual, representational, and accessibility quality. An additional theme emerged from the interview data about clinical safety and the actionability of self-care guidance. During analysis, the researchers actively examined the transcripts for data that contradicted the emerging categories and themes. No case fundamentally contradicted the final thematic structure; however, variations in participant priorities were retained within the relevant subthemes. Figure 2 illustrates the differing perspectives from which nurses and patients perceive the quality requirements of educational material information.

Figure 2.

Differing perspectives from nurses and patients.

Theme 1: Intrinsic Information Quality—Trustworthy and Professionally Supported Information

Intrinsic IQ comprised four subthemes: accuracy, believability, objectivity, and reputation. Participants viewed these characteristics as the foundation of safe AI-generated diabetic foot education.

Accuracy was mainly discussed in relation to the correct identification of ulcer characteristics. Nurses were concerned that an incorrect clinical classification could result in inappropriate nursing advice. One nurse explained:

If it misdiagnoses a neurogenic ulcer as an ischemic ulcer, the nursing advice will be different. (N4)

Patients similarly expected AI-generated information to be truthful and free from unsupported treatment suggestions. As one patient said:

If it makes up unfounded remedies, I’d rather not use it. (P1)

Believability was associated with the use of current clinical guidelines, research evidence, and transparent information sources. Nurses emphasized that materials should be aligned with current guidelines:

AI should be aligned with the latest international guidelines in real time. (N9)

Patients also wanted to know the source of the information so that they could judge its credibility:

If the source is clearly indicated, it would be better. (P9)

Objectivity referred to the need for balanced information that accurately reflected the complexity of DFUs. Nurses did not want the materials to provide exaggerated reassurance or make promises that could mislead patients:

AI should not make excessive promises to reassure patients. It should accurately reflect the complexity of the condition and avoid using leading questions. (N2)

Patients preferred explanations supported by concrete evidence rather than general or nonspecific reassurance. One patient commented:

The doctor keeps telling me ‘the results are good’, but I’d like to hear more concrete information. (P6)

Reputation concerned the credibility of the institution, professionals, or developers responsible for the AI system. Nurses considered institutional certification necessary before materials were provided to patients. Patients also reported greater trust in information produced or approved by established hospitals:

I’ll only dare to follow its advice to treat my wound if it was developed by a major hospital or approved by experts. (P9)

Patients and nurses both considered accuracy, credibility, and trustworthy sources essential. However, their concerns differed in emphasis. Nurses focused primarily on the clinical consequences of inaccurate information, including incorrect ulcer classification, inappropriate nursing recommendations, guideline inconsistency, and the need for institutional or professional verification. Patients were more concerned with whether the information appeared truthful, whether its sources were transparent, and whether the developers or approving institution were trustworthy.

Theme 2: Contextual Information Quality—Information Adapted to the Patient’s Clinical Situation

Contextual IQ included relevancy, completeness, timeliness, appropriate amount of data, and value-added. Participants emphasized that AI-generated education should be adapted to the patient’s ulcer characteristics, treatment stage, and everyday care needs rather than providing generic information.

Relevancy was closely linked to clinical classification and individual wound characteristics. Nurses emphasized that patients with different ulcer grades require different care recommendations:

The nursing care for diabetic foot grade 1 and grade 3 is completely different. (N1)

Patients also wanted recommendations that reflected their current wound condition. One patient said:

I just want AI to tell me what kind of dressing I should use at this stage? Can I get it wet? (P4)

Completeness referred to coverage of the whole care process. Nurses wanted AI-generated education to continue beyond the immediate treatment episode and include discharge observation and follow-up:

Patient discharge does not mean the end of care. AI-generated content should cover the entire recovery cycle. (N7)

Patients focused on the completeness of practical daily-care instructions:

Every morning when I wake up, what should I check on my feet? How should I wash them? (P2)

Timeliness reflected the need for information to be consistent with current clinical practice. Nurses were concerned that outdated information could undermine the safety and professionalism of patient education, especially when wound-care technologies and treatment recommendations change:

If the AI uses data from five years ago, the content will be misleading. (N3)

Patients also wanted current, scientific treatment information rather than outdated or potentially harmful informal remedies:

I hope AI can tell me the latest and most scientific methods. (P14)

Participants also discussed the appropriate amount of information. Nurses wanted AI to summarize essential content efficiently within limited education time:

AI-generated content should be highly condensed, ideally compressing the core procedures into 3 minutes. (N6)

Patients, especially older patients, described difficulty processing lengthy responses and preferred several clearly prioritized recommendations:

Couldn’t it just give me the three or five most important pieces of advice? concise is best. (P12)

The value-added of AI-generated materials was related to improved efficiency and continuing educational support. Nurses described the potential value of automatically generated checklists and preparatory materials:

AI could help summarize important points and prepare suitable education materials. (N7)

Participants also viewed the possibility of asking questions after discharge as an advantage over conventional written materials:

The advantage of AI is that patients can ask questions at any time after discharge. Traditional brochures cannot respond to their individual concerns. (N4)

Both groups wanted information that was relevant, complete, current, and practically useful. Nurses emphasized the need to match education with ulcer severity, clinical classification, treatment stage, and the overall care pathway. They also highlighted the value of AI in summarizing information and preparing individualized education efficiently. Patients, in contrast, focused more on whether the recommendations addressed their immediate wound-care questions and daily routines, such as dressing changes, foot inspection, and indications for seeking medical attention. Patients also expressed concern about excessive information and preferred concise advice that could be used after discharge.

Theme 3: Representational Information Quality—Clear, Consistent, and Organized Communication

Representational IQ consisted of ease of understanding, interpretability, consistent representation, and concise representation. Ease of understanding was associated with simple language and visual presentation. Nurses suggested that AI-generated text could be converted into images or interactive illustrations:

The effect will be better if the text can be converted into images or interactive illustrations. (N10)

Patients preferred conversational language and fewer unfamiliar medical terms:

AI should chat with me like with a neighbor, not keep throwing out those obscure medical terms. (P3)

Interpretability concerned the explanation and consistent use of medical terminology. Nurses wanted terminology to remain standardized throughout the material:

If the AI randomly changes the terms ‘necrotic tissue’ and ‘black scab’ in different paragraphs, it will confuse patients. (N1)

Patients wanted technical terms to be explained in relation to their own condition and required actions:

It could also explain what it means and how it might affect my foot. (P5)

Consistent representation was important because patients receive information from multiple sources. Nurses were concerned that conflicting advice from different platforms could undermine patient education:

AI should provide a unified version for the entire hospital or even the entire industry. (N4)

Patients similarly expressed uncertainty when AI advice differed from the advice provided in hospital:

I searched online, asked doctors, and then consulted AI, but everyone gave different advice. Whom should I listen to? (P7)

Concise representation involved important information and presenting it in an organized structure. Patients reported that information became difficult to use when all content appeared equally important:

Sometimes the information is there, but everything seems equally important. (P11)

They preferred information to be organized so that the most important actions could be identified quickly:

The more organized the information the AI provides, the less likely I am to make mistakes during care. (P8)

Patients and nurses agreed that AI-generated materials should be clear, understandable, consistent, and well organized. Nurses placed emphasis on the format and standardization of the information, including the use of visual aids, consistent terminology, and unified recommendations across professional and digital platforms. Patients were more concerned with whether the language was simple, whether unfamiliar medical terms were explained, and whether the most important instructions were easy to identify.

Theme 4: Accessibility Information Quality—Convenient and Safe Access

Accessibility IQ included accessibility and access security. Participants discussed both the practical availability of the materials and the safety of accessing AI systems.

Participants preferred convenient digital access points, including QR codes and familiar mobile applications. Patients with foot ulcers described how mobility limitations could make it difficult to search for or retrieve information:

If there were a QR code in the hospital corridor or on the medicine bag, I could just scan it and get instructions. (P10)

Patients also preferred access through familiar platforms such as WeChat:

If this AI program could be installed directly in WeChat, I could just tap it and ask, “What should I do about red soles?” (P13)

Accessibility also required adaptation to age-related and functional limitations. Patients mentioned the need for large fonts, sufficient contrast, and voice interaction:

The text in these AI-generated educational images has to be very large. My eyesight is poor. (P15)

Another patient preferred voice-based access when pain or limited mobility made screen-based use difficult:

I wish AI could just “speak” to me. (P3)

Access security concerned the protection, storage, and control of patient information. Nurses were concerned about the privacy risks associated with uploading wound photographs:

If patients send photos of their ulcers to the AI, will it leak their privacy? (N5)

Patients also wanted clear information about how their data were stored and deleted:

I want to know where my information goes after I upload it. If I delete it, can it really be deleted? (P6)

Patients and nurses both regarded convenient access as important, but they emphasized different aspects. Patients mainly discussed practical access through familiar digital platforms, QR codes, mobile applications, large fonts, and voice-based functions. Nurses focused more on access security, including the protection of wound photographs, data storage, data control, and privacy.

Theme 5: AI-Specific Theme—Clinical Safety and Actionable Guidance

Two additional subthemes emerged from the interview data that were not fully represented by the information-quality framework.

Clinical safety reflected concerns about the potential harm caused by incorrect information. As a nurse said:

Even a small error in the suggestion could be risky. It may delay appropriate treatment. (N4)

Nurses regarded professional review as necessary before materials were used in patient education:

AI can help us generate information, but it cannot replace our review. (N7)

This finding extended the concept of IQ beyond the properties of the generated text to include professional oversight.

Actionability referred to whether patients could translate information into specific self-care behaviors. Participants preferred checklist formats and practical decision guidance. One patient said:

Each item is clearly listed as 1, 2, 3, and I check them off as I complete them. (P8)

Patients also wanted clear guidance about when professional care was required:

It should tell me when I must go to the hospital. (P11)

Nurses also valued structured formats that could support consistent education:

It should generate flowcharts or step-by-step structures. What to do in the first step, what to do in the second step. (N8)

Actionability represented the connection between understanding information and performing appropriate self-care. Participants did not regard information as high quality merely because it was accurate. It also needed to tell patients what to do, how to do it, and when to seek professional assistance.

The two groups also differed in their emphasis on the AI-specific characteristics that emerged from the data. Nurses primarily emphasized clinical safety and professional review. They considered AI-generated information to be an aid to education rather than a substitute for nurses’ clinical judgment. They believed that generated content should be reviewed before being provided to patients. Patients focused more on actionability, including step-by-step instructions, checklist formats, and clear guidance about when to seek medical care.

Discussion

This study found that patients and nurses have both commonalities and significant differences in their information quality needs for AI-generated patient education materials on diabetic foot ulcers. Both groups of participants hope that the information is accurate, credible, relevant to the condition, easy to understand, and accessible. However, the emphasis on evaluating the quality of information differs between the two groups. Nurses are more concerned with whether the information is safe, standardized, and professionally reviewed. Patients, on the other hand, are more concerned with whether the information is relevant to their current issues, whether it is easy to understand, and whether it can be directly applied to their daily care. The most obvious difference is that nurses emphasize clinical safety in AI-specific theme. Patients, on the other hand, place more emphasis on actionability, hoping to receive clear operational steps.

The reasons for these differences may be related to the different roles and tasks of the two types of participants in patient education. Nurses need to make risk assessments and be responsible for patient safety, so they pay more attention to the accuracy, professionalism, and potential clinical consequences of misinformation. Especially in DFU management, the nursing measures corresponding to different wound types, severity, and treatment stages are not the same. Therefore, if AI-generated content has any judgment errors, it may affect subsequent nursing recommendations. After evaluating the performance of AI chatbots in the management of diabetic foot ulcers, Shiraishi et al found that there are still differences in accuracy and the authenticity of references among different AI tools.¹⁰ This also indicates that in scenarios like DFU, which require strong clinical judgment, nurses emphasizing professional review is practically necessary.

In contrast, patients are more concerned with whether the information can directly help them with their daily self-management. After leaving the hospital, patients need to perform foot inspections, wound care, and abnormality identification on their own, so they not only need “correct information” but also “information that is easy to understand and implement”. Gotlieb et al found that the public may misunderstand some commonly used clinical medical expressions.²³ This study further found that patients’ needs go beyond just “understanding” medical information; they also include understanding how this information relates to their current wounds and what actions should be taken next. It can be seen that nurses are more concerned with safety and professional oversight before the use of information, while patients are more focused on understanding and actual implementation during the use of information.

The emergence of the themes of clinical safety and actionability suggests that traditional dimensions of information quality may be insufficient to fully evaluate AI-generated patient education materials. Wang and Strong’s information quality framework mainly focuses on the information itself.¹⁸ But for AI-generated health education materials, once the information is used by patients for wound care and medical decision-making, its impact no longer remains confined to the text itself. Nurses emphasize clinical safety, indicating that even if a piece of content appears accurate and complete on the surface, it can still pose risks if it is not suitable for the current patient. Patients emphasize actionability, which means that even if a piece of information is easy to understand, its practical utility is limited if it does not clearly tell the patient what to do next. Rebitschek et al found that when evaluating health information provided by generative AI, the AI-generated content still has shortcomings in terms of evidence-based practice and the presentation of evidence.²⁴ This implies that even well-articulated AI-generated information cannot be solely judged based on these criteria. Therefore, the evaluation of AI-generated patient education materials cannot only address the correctness of the content but also further assess the clinical safety of the information and whether it can guide patient actions.

The results of this study suggest that AI is more suitable as an auxiliary tool for nurses to conduct patient education, rather than replacing nurses. AI can be used to quickly generate, organize, and adjust educational content, and provide more targeted explanations based on the questions raised by patients. However, for specialized content such as wound assessment and risk identification, nurse review is still required.²⁵ Zhu and others found that supervision by medical professionals can enhance patients’ trust and acceptance of medical AI.²⁶ At the same time, the information that has undergone professional review also needs to be adjusted according to the actual usage needs of the patients. For example, use simple language, highlight key points, provide step-by-step instructions, and clearly indicate when professional help is needed. Existing research rarely incorporates patient feedback into the quality evaluation of AI-generated health education materials.¹⁴ Patient involvement should not be limited to satisfaction evaluation after the development of AI-generated health education materials, but should be integrated into the process of forming information quality standards. Therefore, in DFU patient education, AI improves the efficiency of information generation and adjustment, nurses are responsible for ensuring clinical safety and professional standards, and the actual needs of patients determine how the final information is presented.

Limitations

The results of this study are significant for the development and enhancement of patient education materials aided by AI by focusing on the diabetic foot ulcer patients. Nevertheless, this study has its limitations. All participants were recruited from a single hospital, and the patient sample was restricted to hospitalized individuals who were clinically stable, able to communicate, and willing to participate in an interview. Eligibility also required prior exposure to AI-generated health education materials. These potential selection effects limit the transferability of the findings to outpatients or individuals with severe functional impairment.

The face-to-face interview may have introduced social desirability bias. Patients may have been reluctant to criticize health education provided by nurses, whereas nurses may have tended to present their professional practices more favorably. The interviewer was a nurse employed at the study hospital and was therefore familiar with the clinical setting. Although this insider role facilitated contextual understanding, it may also have influenced participants’ willingness to express negative views and interpretation of the data. The findings should be interpreted as context-dependent accounts, and their credibility and transferability should be considered in relation to the characteristics of the participants and the study setting.

Conclusion

Patients and nurses had different perspectives on AI-generated education for DFU care. Nurses were more concerned with the clinical accuracy, safety and need for professional review. Patients wanted to know if the information was relevant to their current wound, easy to follow and useful for everyday self-care. The additional subthemes of clinical safety and actionability imply that existing information quality criteria do not fully capture what matters when AI-generated information is used in patient care.

AI-generated patient education materials should be judged not only by the quality of the text but also by how safely and actionably the information can be used. In DFU care, AI can help prepare and tailor educational content, while nurses remain responsible for reviewing clinically sensitive information. Patients should also be involved in judging whether the final material is clinically safe and actionable in real-life self-management. Future research should examine these findings through multicenter qualitative studies, broader stakeholder-consensus work, and cognitive testing with users of different literacy levels to further refine and validate the identified information-quality requirements.

Funding Statement

This work was supported by the Central Guidance for Local Science and Technology Development Funds Project (ZY24212012, F. W.), Guangxi Zhuang Autonomous Region Health Commission Geriatric Nursing Key Specialty Construction Project, and Nanning Scientific Research and Technology Development Program Project (20235077). The funders provided financial support for the research but had no role in the study design, data collection, data analysis, interpretation of the results, preparation of the manuscript, or the decision to submit the manuscript for publication.

Abbreviations

DFUs, diabetic foot ulcers; AI, Artificial intelligence; COREQ, consolidated criteria for reporting qualitative research; IQ, information quality.

Data Sharing Statement

The interview data are not publicly available due to participant confidentiality and ethical restrictions. Selected data supporting the findings are presented in the article as anonymized quotations. Additional de-identified data may be available from the Li Wang corresponding author upon reasonable request, subject to ethical approval and confidentiality requirements.

Ethical Considerations

This study was approved by the Ethics Committee of the Fifth Affiliated Hospital of Guangxi Medical University (KY2025-decision-138), and strictly followed the principles of the Declaration of Helsinki. Informed consent has been obtained from all the respondents.

Author Contributions

All authors made a significant contribution to the work reported, whether that is in the conception, study design, execution, acquisition of data, analysis and interpretation, or in all these areas; took part in drafting, revising or critically reviewing the article; gave final approval of the version to be published; have agreed on the journal to which the article has been submitted; and agree to be accountable for all aspects of the work.

Disclosure

All authors have no relevant financial or non-financial interests to disclose.