When Patients Bring AI-Generated Health Advice into Clinical Consultations in Viet Nam: Verification, Communication and Safe Decision-Making
IMPLEMENTATION INSIGHTS
When Patients Bring AI-Generated Health Advice into Clinical Consultations in Viet Nam: Verification, Communication and Safe Decision-Making
PRACTICAL FRAMEWORKS & TOOLS
Last reviewed: 9 October 2026
A patient arrives at a medical consultation with a screenshot from an AI chatbot. After entering recent symptoms and laboratory results, the patient received a detailed explanation suggesting a serious condition and recommending additional investigations. The patient now wants to know whether the AI is correct and whether the treating physician may have missed something important.
In another situation, a patient with a chronic condition has already changed the timing of a prescribed medication after consulting an AI application. The patient believes the change is safer because the explanation sounded detailed and medically convincing.
These situations raise a practical question for healthcare professionals: how should they respond when patients bring health advice generated by AI systems that the healthcare organization has neither selected nor evaluated?
The appropriate response is not to dismiss the information simply because it came from AI, nor to accept it because the answer appears knowledgeable. The clinician needs to understand what the patient has been told, determine whether any action has already been taken, verify clinically important claims and help the patient make an informed decision based on their actual health circumstances.
For healthcare organizations in Viet Nam, this presents a challenge in clinical communication and patient safety. Unlike AI applications formally introduced into hospital operations, the information originates outside the healthcare organization and may already have influenced the patient's expectations or decisions before the consultation begins.
Patients are bringing a new kind of health information into consultations
Patients have long searched for medical information before visiting healthcare professionals. Health websites, search engines, social media and advice from family members have all influenced how people interpret symptoms and consider treatment options.
Generative AI changes this experience. Instead of searching several websites, patients can describe their concerns in everyday language, upload test results, ask follow-up questions and receive responses that appear personalized to their circumstances.
This can be useful. AI may explain unfamiliar medical terminology, summarize possible causes of symptoms, describe treatment options or help patients prepare questions for their doctors. Some patients may arrive better informed and more confident about discussing their health.
However, the conversational nature of AI can also create an impression of clinical understanding that the system does not necessarily possess.
A chatbot may produce a detailed explanation without access to the patient's complete medical history, physical examination findings, current medications or other information that would influence a professional assessment. It may present several possible diagnoses without adequately distinguishing common explanations from rare or serious conditions.
Patients may not recognize these limitations, particularly when the response uses medical terminology, provides references or presents its conclusions confidently.
By the time a patient reaches the consultation, the AI-generated information may already have shaped their expectations. Some patients may feel reassured. Others may become unnecessarily worried, convinced that they have a particular disease or reluctant to follow previously recommended treatment.
The clinician therefore needs to understand more than the content of the AI response. It is equally important to understand how the patient interpreted that response and whether it has influenced their behavior.
What recent research tells us about patients using AI for medical advice
Recent research highlights an important distinction between the medical knowledge demonstrated by AI models and their usefulness when people rely on them to make health decisions.
A randomized, preregistered study by Bean and colleagues, published in Nature Medicine in February 2026, evaluated whether large language models could help members of the public identify possible medical conditions and choose appropriate courses of action. The study involved 1,298 participants in the United Kingdom who worked through simulated clinical scenarios.
When tested independently with complete scenarios, the models identified relevant conditions in 94.9% of cases and selected the appropriate care disposition in 56.3% on average. However, participants using those models identified relevant conditions in fewer than 34.5% of cases and selected the correct disposition in fewer than 44.2%. Their performance on these outcomes was no better than that of participants using their usual information sources [1].
The researchers identified difficulties in how people interacted with AI, including incomplete information provided by users, misleading model responses and failures to recognize or act on information supplied by the system.
These findings are particularly relevant to clinical consultations. An AI response may appear medically sophisticated, but its usefulness depends partly on the information the patient entered, the questions asked and how the conversation developed.
Another study by Ramaswamy and colleagues, published online in Nature Medicine in February 2026, evaluated ChatGPT Health using 60 clinician-authored clinical scenarios across 16 experimental conditions, producing 960 responses. Among scenarios classified as emergencies, the system recommended care that was less urgent than the reference standard in 52% of responses [2].
These were structured test scenarios, not real patients. The findings should not be generalized to all consumer AI applications or later software versions. Nevertheless, they illustrate why AI-generated advice about symptom urgency requires assessment against the patient's actual clinical condition.
Research also helps explain why patients may trust AI responses. In a cross-sectional study published in the Journal of Medical Internet Research in October 2024, Armbruster and colleagues asked 64 patients to evaluate AI-generated and physician-written responses to questions originally submitted through an online medical platform. Patients rated ChatGPT responses more highly for empathy and perceived usefulness. Specialist physicians separately assessed clinical correctness and potential harm, finding that patients did not consistently distinguish potentially harmful responses from those considered safer [3].
The study compared written responses in an online setting, not AI consultations against in-person medical care. Its relevance lies in showing that the qualities patients appreciate in an explanation are not necessarily sufficient to establish clinical reliability.
Research has also examined physicians' experiences. In a cross-sectional survey published in Frontiers in Public Health in September 2026, Almalki and AlMubarak surveyed 393 physicians in Saudi Arabia. Approximately two-thirds reported encountering patients who had obtained health information from AI tools. Among the 262 physicians reporting such encounters, 79.4% indicated that consultations required more time, while 65.7% reported that patient use of AI made clinical decision-making more challenging [4].
A separate social-media survey conducted in Poland and published in Frontiers in Digital Health in August 2026 examined patients' disclosure of AI use before medical visits. Among 519 respondents reporting pre-consultation AI use, 82.5% selected “No” when asked whether they had informed a physician. However, the researchers did not confirm that the relevant consultations had actually taken place, so this figure cannot be interpreted as the proportion of patients who concealed AI use during completed medical encounters. Fear of a negative physician reaction was among the reasons reported for selecting “No” [5].
These studies involve different populations, research methods and AI systems. They do not establish how frequently Vietnamese patients use AI before consultations or how often this causes clinical harm.
A systematic review by Huo and colleagues, published in JAMA Network Open in February 2025, also identified substantial variation and reporting limitations across 137 studies evaluating chatbot health advice. This reinforces the need for caution when interpreting claims about the reliability or clinical usefulness of these systems [6].
Taken together, the evidence suggests that the quality of an AI-generated answer is only one part of the issue. How patients understand, trust and act on that answer may be equally important.
Begin by understanding what the patient asked AI and what happened afterward
When a patient presents AI-generated health advice, the clinician does not necessarily need to examine every sentence in the conversation.
A more useful starting point is to understand why the patient consulted AI, what information was provided and which part of the response influenced their concerns or decisions.
The original question matters. A patient asking an AI system to explain an abnormal laboratory result may receive a different response from someone asking whether that result indicates cancer. A question that already suggests a particular diagnosis may influence the direction of the generated explanation.
The information supplied to the AI is equally important. A patient may describe a symptom but omit relevant medical history, medications, previous investigations or examination findings. The generated advice may therefore appear reasonable in isolation while being inappropriate for the patient's actual circumstances.
The clinician should clarify what the patient believes the AI has concluded. Did the patient understand the response as a possible explanation, a confirmed diagnosis or a recommendation to change treatment?
This distinction can reveal misunderstandings that are more clinically important than the exact wording of the AI response.
It is also necessary to establish whether the patient has acted on the advice. Some patients may have stopped prescribed medication, changed a dose, started using an over-the-counter product, postponed a medical appointment or decided against a recommended investigation.
Where those actions may affect safety, they deserve prompt assessment.
A short conversation can often identify the relevant concern. The clinician may ask what prompted the patient to consult AI, which statement caused concern and whether anything has changed in the patient's care as a result.
If the original prompt or conversation is available, reviewing it may help clarify the context. However, a screenshot may contain only part of the exchange. Clinicians should not assume that a selected response represents everything the system communicated.
The purpose is to understand the patient's concern and identify claims that could affect clinical decisions, rather than conduct a comprehensive evaluation of the external AI application.
Address immediate safety concerns before debating the AI response
Some AI-generated information can be discussed during a routine consultation. Other situations require more urgent clinical attention.
The first priority is to determine whether the patient currently has symptoms or circumstances requiring prompt assessment, regardless of what the AI advised.
For example, a patient may have been reassured by AI that worsening breathlessness is probably related to anxiety. If the patient's actual presentation raises concern about a serious cardiopulmonary condition, the clinical assessment should not be delayed while the clinician reviews the chatbot's explanation.
Similarly, a patient who has stopped an essential medicine, taken an excessive dose or started a potentially harmful treatment based on AI advice may require immediate medication review or other intervention.
The level of urgency should be determined from the patient's symptoms, history, examination findings and applicable professional guidance.
Particular care is warranted when AI advice concerns potentially serious symptoms, medication changes, pregnancy-related complications, severe mental health concerns, complex chronic conditions or decisions to delay necessary treatment.
Not every AI recommendation in these areas is incorrect. The concern is that an inaccurate recommendation may have significant consequences if the patient follows it without professional assessment.
Where a patient has delayed seeking care because of an AI response, the clinician should evaluate the current health condition and any possible consequences of the delay. Establishing whether the chatbot's original recommendation was reasonable may be less urgent than addressing the patient's present clinical needs.
The source of the advice should never distract from timely assessment and treatment.
Verify clinically important claims against the patient's actual evidence
AI-generated health advice should be treated as information to evaluate, not as an established clinical finding.
The clinician should identify which claims could materially influence diagnosis, treatment or patient behavior and assess whether they are supported by reliable evidence relevant to the individual patient.
If the AI suggests a diagnosis, the clinician should determine whether the patient's symptoms, history, physical examination and available investigations support that possibility.
A plausible diagnosis mentioned by AI does not establish that the patient has the condition. At the same time, a reasonable concern should not be dismissed solely because it originated from a chatbot. If the information raises a legitimate question that has not been adequately addressed, further assessment may be appropriate.
For laboratory results, the original report should be reviewed, including the units, reference intervals, relevant clinical context and any changes over time. An isolated abnormal result does not necessarily establish disease, while a normal value may not exclude a clinically important condition.
The same principle applies to diagnostic imaging, pathology reports and other clinical documents. An AI-generated interpretation of an uploaded test result should not be treated as an authenticated professional interpretation without appropriate clinical evaluation.
Medication-related advice requires particular attention. If AI recommends changing a dose, stopping a medicine or combining treatments, the clinician should assess the patient's actual prescription, medication history, underlying conditions and relevant risks. Any treatment change should be based on an appropriate professional decision.
Claims about clinical guidelines, treatment effectiveness or contraindications also need scrutiny. AI systems may provide references that are outdated, incomplete, incorrectly interpreted or occasionally nonexistent.
The presence of a citation does not establish that the source supports the statement. Where a claim could materially affect care, clinicians should rely on identifiable evidence, applicable clinical guidance and the patient's actual medical information.
Verification should be proportionate to the consequences of an error. A general explanation of a medical term may require little discussion, while advice to stop anticoagulation, avoid an urgent investigation or use an unprescribed medicine requires more careful assessment.
This allows the clinician to focus on information that matters without allowing the AI-generated document to replace ordinary clinical reasoning.
Communicate without turning the consultation into a disagreement about authority
Disagreements between AI-generated information and a clinician's recommendation can create tension.
A patient may believe that AI has identified a diagnosis the doctor overlooked. The clinician may interpret the patient's questions as a challenge to professional judgment. If the discussion becomes an argument about who is right, the underlying concern may remain unresolved.
A more productive approach is to acknowledge the patient's effort to understand their health and then explain how the information will be evaluated.
A clinician might say, “Thank you for showing me this. I can understand why it raised questions. Let's look at which parts fit your symptoms and test results before we decide what to do next.”
This does not endorse the AI response. It provides an opportunity to evaluate the information without making the patient feel that seeking it was inappropriate.
If the AI advice is broadly accurate, the clinician can acknowledge the useful information while explaining how it applies to the patient's circumstances.
If the advice is misleading, the explanation should focus on the specific problem. The AI may have misunderstood a laboratory value, overlooked an important contraindication, overstated the significance of a symptom or reached a conclusion without relevant clinical information.
Explaining why a recommendation does not fit the patient's circumstances is generally more useful than simply stating that AI cannot be trusted.
Clinicians should also remain willing to reconsider their own assessment when a patient raises a legitimate concern. Good clinical practice does not require rejecting information because it came from an unfamiliar source. It requires evaluating that information appropriately.
The World Health Organization's guidance on large multi-modal models also emphasizes the importance of protecting human well-being, autonomy and safety when AI is used in health-related activities [7]. Although that guidance does not prescribe how to conduct an individual consultation involving patient-generated AI advice, its broader principles are consistent with respectful communication and informed clinical decisions.
The aim is to preserve a constructive relationship while making clear that individual clinical decisions require more than a convincing explanation generated by software.
When AI recommends a test or treatment the clinician has not proposed
A common situation occurs when an AI system recommends an investigation, treatment or specialist referral that was not part of the clinician's original plan.
A patient may request additional imaging, laboratory testing or another intervention because the AI described a possible serious diagnosis.
The clinician should assess whether the request is justified by the patient's actual presentation, relevant evidence, professional guidance and the expected benefits and risks.
An AI-generated recommendation may occasionally identify an appropriate investigation that deserves consideration. In other circumstances, it may encourage unnecessary testing or create concern about a condition that is unlikely in the patient's clinical situation.
The decision should not depend on whether the suggestion came from AI. It should depend on whether the intervention is clinically appropriate.
When further testing is not indicated, the clinician should explain why, including what is currently known, what uncertainty remains and which symptoms or findings would justify reassessment.
Patients may also use AI to question recommended treatment. For example, a chatbot may emphasize possible medication adverse effects without adequately considering the benefits of treatment or the risks of leaving the underlying disease untreated.
In this situation, the clinician should first understand the patient's concern, then discuss the expected benefits, relevant risks and reasonable alternatives.
Shared decision-making remains important. Patients should receive sufficient information to understand the available options, and their preferences should be considered within the applicable professional and legal framework.
Where a patient declines recommended care after relying on AI-generated advice, appropriate counseling and the relevant requirements governing refusal of treatment and documentation must still be observed.
AI-generated information may contribute questions to this discussion, but it cannot substitute for an individualized clinical assessment or the requirements for valid clinical decision-making.
Patients may not always tell their doctors that they used AI
Not every patient who consults AI will mention it during a medical appointment.
Some may regard the interaction as ordinary information seeking. Others may fear that the clinician will dismiss their concerns or view their use of AI negatively.
The Polish survey published in 2026 suggests that anticipated negative reactions from physicians may be a barrier to disclosure, although the study did not establish actual disclosure rates during completed consultations [5].
For healthcare organizations, this is primarily a communication issue rather than a reason to question patients about every digital tool they use.
A clinician does not need to conduct a lengthy interview about technology use during every appointment. However, when a patient's concerns, beliefs or treatment decisions appear to have been influenced by outside information, a neutral question may help.
Asking whether the patient has read or received advice that differs from the proposed plan may reveal information worth discussing.
Patients should not be made to feel that using AI is itself inappropriate. The important question is whether the information has been understood correctly and whether it has influenced decisions that affect care.
Encouraging patients to discuss their concerns openly may help clinicians identify potentially unsafe actions before further harm occurs.
Document clinically relevant information without copying entire AI conversations
When AI-generated advice materially affects a consultation, some information may need to be reflected in the clinical record.
The purpose is to preserve relevant information about the patient's presentation, concerns, actions, professional assessment and agreed plan.
For example, if a patient has stopped a prescribed medication after receiving AI advice, that change and its clinical implications may be important to document. Similarly, when a patient requests an investigation or declines treatment because of information generated by AI, the relevant clinical discussion may need to be recorded.
Documentation should remain proportionate to the circumstances.
There is generally no need to copy an entire chatbot conversation into the medical record simply because the patient has shown it to the clinician. A concise, accurate record of the clinically important claims and actions may be more useful than storing a lengthy transcript containing irrelevant information.
Any AI-generated statement included in the clinical record should be clearly distinguishable from the clinician's own findings and conclusions.
A diagnosis suggested by an external AI system should not appear as a confirmed clinical diagnosis unless it has been appropriately established.
The clinician should also avoid implying that an external AI application has been validated or endorsed by the healthcare organization when that has not occurred.
Where the source or meaning of a clinically important statement remains uncertain, the record should preserve that distinction rather than present the information as verified medical evidence.
Protect patient information when reviewing external AI content
Patients may bring screenshots, chatbot transcripts, AI-generated summaries or uploaded medical documents to their appointments.
These materials may contain sensitive health information, personal identifiers and information about family members or other individuals.
Healthcare organizations should handle such content in accordance with applicable medical-record, confidentiality and personal-data-protection requirements.
An important distinction exists between reviewing information that a patient voluntarily shows during a consultation and entering the patient's clinical data into another external AI application.
The latter may involve additional personal-data processing or disclosure and requires assessment under the organization's applicable legal obligations, approved information-management arrangements and internal policies.
Clinicians should not assume that a consumer AI application is authorized to process hospital patient information merely because the patient has used it personally.
Nor should patients routinely be encouraged to upload complete medical records to an external chatbot when a more limited question or an explanation from the treating professional would be sufficient.
Where patients have already shared their information with external AI services, clinicians can explain relevant privacy considerations without implying that the healthcare organization controls those services or is automatically responsible for every independent action taken by the patient.
If a healthcare organization chooses to recommend, integrate or formally provide a particular AI application as part of patient care, that becomes a separate implementation decision requiring appropriate evaluation and governance.
What Viet Nam's current legal framework means
Viet Nam's Law on Medical Examination and Treatment No. 15/2023/QH15, as amended, provides the principal legal foundation for professional responsibilities and patient rights in this situation.
Article 9 establishes patients' rights to receive information and explanations about their health condition and medical care, including guidance on self-monitoring, care and prevention of complications. Article 10 protects patients' dignity and privacy, including the confidentiality of medical-record information, subject to the applicable legal provisions [9].
Article 11 addresses patients' rights to choose examination and treatment methods after receiving appropriate information, explanation and counseling, within the relevant legal framework.
Article 13 separately governs patients' rights to refuse medical examination and treatment or leave a healthcare establishment before completion of treatment, subject to specified conditions. These include written acknowledgment of responsibility following professional counseling and exceptions where treatment is compulsory under the law [9].
These provisions are relevant when patients ask questions, request investigations or express treatment preferences influenced by AI-generated information.
For healthcare professionals, Article 44 establishes duties toward patients, including respect for patient rights, appropriate counseling and provision of information. Article 45 addresses professional duties, including compliance with professional and technical requirements and responsibility for the examination and treatment provided [9].
These obligations support an approach in which clinicians take patient concerns seriously while maintaining appropriate professional assessment and decision-making.
They do not require clinicians to accept AI-generated diagnoses or treatments simply because patients present them during consultations. Nor do they establish a universal requirement to verify every statement contained in an external chatbot conversation.
The Law on Artificial Intelligence No. 134/2025/QH15, effective from 1 March 2026, establishes Viet Nam's broader AI regulatory framework. Decree No. 142/2026/NĐ-CP, effective from 1 May 2026, further details relevant requirements, including risk classification, transparency and obligations applicable to different participants [10,11].
An important distinction must be maintained. A patient independently using a consumer AI application and bringing its output into a consultation does not, by that fact alone, mean that the healthcare organization has deployed the AI system.
The organization's responsibilities depend on what it actually does. Reviewing a patient's questions, independently using external AI for professional clinical activities and formally introducing an AI service into hospital operations are different activities that may carry different obligations.
Patient privacy must also be considered. The Law on Personal Data Protection No. 91/2025/QH15 and Decree No. 356/2025/NĐ-CP, both effective from 1 January 2026, provide relevant requirements for processing and protecting personal data [12,13].
These requirements need to be assessed according to the actual activities involved, particularly when patient information is collected, retained, disclosed or transferred through external digital services.
The approach discussed in this article is not a prescribed Vietnamese Ministry of Health protocol specifically for patient-generated AI advice. It is intended to help healthcare organizations respond consistently and safely within their existing legal, professional and patient-safety responsibilities.
A practical consultation approach for healthcare organizations in Viet Nam
Healthcare organizations do not need to create a new department or separate AI governance process for every patient who arrives with a chatbot response.
A practical approach can be incorporated into ordinary clinical consultations.
The first consideration is whether the patient has an immediate safety concern. Symptoms requiring urgent assessment, potentially harmful medication changes or delays in necessary care should be addressed before examining the broader accuracy of the AI response.
The clinician should then identify the patient's main concern and clarify how the AI information was obtained. This includes understanding what the patient asked, what information was provided, which conclusion influenced their thinking and whether any action has already been taken.
The next consideration is verification. The clinician should focus on claims that could materially affect care and assess them against the patient's clinical presentation, original medical records, relevant investigations and appropriate evidence.
Once the important claims have been evaluated, the clinician should explain which information is supported, which conclusions are uncertain and why an AI recommendation may or may not be appropriate for the individual patient.
The consultation should conclude with a clear plan. The patient should understand what has been established, what remains uncertain, whether additional assessment is needed and which symptoms or changes should prompt further medical attention.
Where the concern requires more time or specialist input, an appropriate follow-up discussion, additional review or referral may be arranged according to the clinical circumstances.
The clinician does not need to determine whether the external AI application is generally reliable. The immediate responsibility is to make appropriate decisions about the patient presenting for care.
Making the approach workable in Vietnamese clinical settings
The practical arrangements will depend on the healthcare setting.
In a busy outpatient clinic, clinicians may have limited time to review lengthy chatbot conversations. It is therefore reasonable to prioritize the patient's most important concern and any information that could affect immediate clinical safety.
A patient bringing several pages of AI-generated analysis does not necessarily require the clinician to review every statement during the appointment. The relevant question is which parts could influence the current clinical decision.
Healthcare organizations may encourage patients to bring original laboratory reports, prescriptions and other medical documents rather than relying solely on summaries prepared by AI. This allows the clinician to verify important information directly.
Where an organization already uses digital appointment or patient communication systems, it may consider allowing patients to submit their principal questions before a consultation through an approved channel. Such arrangements should not automatically involve transmitting medical information to external AI services.
Professional education can remain focused on practical cases. Clinicians may benefit from discussing scenarios involving unsupported AI-generated diagnoses, misleading interpretations of test results, inappropriate medication advice and delayed care following incorrect reassurance.
Communication skills are equally important. A clinician who understands AI limitations but responds dismissively may still fail to identify the patient's real concern or an unsafe action that has already occurred.
In Viet Nam, language and health literacy can introduce additional complexity. Patients may bring information written in Vietnamese, English or a combination of medical terminology and everyday language. Some may rely on relatives to interpret the information or discuss treatment options on their behalf.
Clinicians should establish what the patient personally understands, who is involved in decisions and whether communication support is needed, while respecting patient rights and confidentiality.
The US Agency for Healthcare Research and Quality's Health Literacy Universal Precautions Toolkit provides useful methods for encouraging questions, explaining information clearly and checking understanding through teach-back [8]. These are practical communication approaches, not mandatory Vietnamese requirements for consultations involving AI.
These considerations are particularly important for patients with multiple chronic conditions, complicated medication regimens or limited access to specialist services.
The objective is not to discourage patients from seeking information. It is to help them understand the difference between general medical explanations and advice that has been assessed for their specific clinical circumstances.
Learn from significant cases without creating unnecessary reporting
Healthcare organizations may benefit from reviewing selected encounters in which patient-generated AI advice contributed to a meaningful safety concern.
Examples include patients who stopped prescribed treatment, delayed urgent medical assessment, used inappropriate medication combinations or misunderstood serious test results after relying on external AI information.
The review should focus on the clinical consequences and opportunities for improvement.
Not every disagreement between AI and a clinician represents a patient-safety incident. Similarly, a patient asking about a diagnosis suggested by a chatbot is not, by itself, evidence of unsafe behavior.
Organizations should distinguish ordinary information seeking from situations involving clinically significant misinformation, potentially harmful actions or delayed care.
Where repeated problems occur, the findings may help improve patient education, medication counseling, clinician communication or procedures for directing patients toward appropriate professional assessment.
Local evaluation should also recognize that an absence of reported incidents does not establish that patient-generated AI advice is harmless. Some patients may never disclose their AI use, and the consequences of inaccurate information may not become apparent immediately.
For most healthcare organizations, integrating this learning into existing quality-improvement and patient-safety activities will be more practical than creating a separate reporting process for every AI-related conversation.
From competing answers to better clinical decisions
Patients bringing AI-generated health advice into clinical consultations should not automatically be viewed as a problem. Their questions may reflect a genuine effort to understand symptoms, evaluate treatment options or prepare for a difficult decision.
AI can provide useful explanations and help patients formulate questions they might otherwise hesitate to ask. However, a detailed and convincing answer does not establish that its conclusions are correct or appropriate for an individual patient.
The clinician's role is to understand what the patient has been told, determine whether any action has already been taken, verify claims that could affect care and explain the reasoning behind the agreed clinical plan.
This requires attention to both information accuracy and the quality of communication.
Dismissing AI-generated information without discussion may leave important concerns unresolved. Accepting it without appropriate verification may introduce avoidable risk. A safer approach is to evaluate the relevant claims respectfully and independently, using the patient's actual clinical information.
For healthcare organizations in Viet Nam, this is an opportunity to strengthen patient-centered communication while preserving professional accountability and safe decision-making.
When patients bring AI-generated health advice into a consultation, the goal should not be to decide whether the doctor or AI is right. It should be to ensure that the patient leaves with a clearer understanding, an appropriate clinical plan and greater confidence in making safe health decisions.
Professional education notice
This article is intended for professional education and knowledge exchange. It does not constitute individualized medical, diagnostic, prescribing, legal, regulatory, personal-data-protection, accreditation or quality-management advice.
The approaches discussed are proposed clinical communication and patient-safety practices. They are not a prescribed Vietnamese Ministry of Health procedure, mandatory checklist for patient-generated AI advice, WHO clinical protocol or international certification requirement.
Applicable obligations depend on the patient's clinical circumstances, the healthcare professional's scope of practice, the information presented, the actual use of AI by the healthcare organization and the relevant legal requirements.
The research cited involves different populations, simulated scenarios, AI models and evaluation methods. Findings from international studies should not be interpreted as estimates of patient AI use, disclosure behavior or clinical harm in Viet Nam.
Recommendations concerning verification, communication, clinical documentation, privacy and the management of potentially unsafe AI-generated advice are intended to support existing professional and patient-safety responsibilities.
References
Bean AM, Payne RE, Parsons G, et al. Reliability of LLMs as Medical Assistants for the General Public: A Randomized Preregistered Study. Nature Medicine. 2026;32:609–615. DOI: 10.1038/s41591-025-04074-y.
Ramaswamy A, Tyagi A, Hugo H, et al. ChatGPT Health Performance in a Structured Test of Triage Recommendations. Nature Medicine. 2026;32:1671–1675. DOI: 10.1038/s41591-026-04297-7.
Armbruster J, Bussmann F, Rothhaas C, Titze N, Grützner PA, Freischmidt H. “Doctor ChatGPT, Can You Help Me?” The Patient's Perspective: Cross-Sectional Study. Journal of Medical Internet Research. 2024;26. DOI: 10.2196/58831.
Almalki M, AlMubarak S. Navigating the Artificial Intelligence-Informed Patient: A Cross-Sectional Study of Physicians' Perceptions and Clinical Practices. Frontiers in Public Health. 2026;14:1932876. DOI: 10.3389/fpubh.2026.1932876.
Wójcik S, Rulkiewicz A, Domienik-Karłowicz J. Generative AI Use Before Medical Visits: Disclosure-Item Responses, Trust, and Care-Seeking Behaviors in a Cross-Sectional Social-Media Survey in Poland. Frontiers in Digital Health. 2026;8:1933451. DOI: 10.3389/fdgth.2026.1933451.
Huo B, Boyle A, Marfo N, et al. Large Language Models for Chatbot Health Advice Studies: A Systematic Review. JAMA Network Open. 2025;8(2). DOI: 10.1001/jamanetworkopen.2024.57879.
World Health Organization. Ethics and Governance of Artificial Intelligence for Health: Guidance on Large Multi-Modal Models. Geneva: World Health Organization; 2024. ISBN: 978-92-4-008475-9.
Agency for Healthcare Research and Quality. Health Literacy Universal Precautions Toolkit. Third edition. Edited by Cindy Brach. Rockville, MD: AHRQ; 2024. AHRQ Publication No. 23-0075.
National Assembly of Viet Nam. Law No. 15/2023/QH15 on Medical Examination and Treatment, as amended. Adopted 9 January 2023; effective 1 January 2024.
National Assembly of Viet Nam. Law No. 134/2025/QH15 on Artificial Intelligence. Adopted 10 December 2025; effective 1 March 2026.
Government of Viet Nam. Decree No. 142/2026/NĐ-CP Detailing a Number of Articles and Measures for Implementation of the Law on Artificial Intelligence. Issued 30 April 2026; effective 1 May 2026.
National Assembly of Viet Nam. Law No. 91/2025/QH15 on Personal Data Protection. Adopted 26 June 2025; effective 1 January 2026.
Government of Viet Nam. Decree No. 356/2025/NĐ-CP Detailing a Number of Articles and Measures for Implementation of the Law on Personal Data Protection. Issued 31 December 2025; effective 1 January 2026.