The AI-Informed Patient: Rethinking Pharma Messaging Before the HCP Visit

A growing share of patients now arrive at their healthcare professional (HCP) appointment having queried a large language model (LLM) before their visit. They may have researched symptoms, compared treatment options, reviewed side-effect profiles, and formed preliminary clinical hypotheses. For pharmaceutical companies, this behavioral shift has meaningful implications for medical communications strategy.

This article summarizes the available evidence on AI-informed patient behavior, examines associated risks and opportunities at the point of care, and discusses how pharma messaging — directed at both HCPs and patients — may need to adapt. Interpretations reflect the author’s synthesis of the cited literature; they do not constitute clinical guidance, promotional claims, or regulatory advice.

1. THE RISE OF THE AI-INFORMED PATIENT: WHAT THE DATA SHOW

Consumer AI adoption in health is growing rapidly. A 2024 survey of U.S. consumers, cited by Sarabu et al. in JMIR1, reported that approximately 48% of respondents had used generative AI for health-related questions, with accessibility, affordability, and ease of use cited as key drivers.

Rock Health’s 2025 Consumer Adoption of Digital Health Survey — which polled 8,000 U.S. Census-matched adults — found that 32% reported using AI chatbots for health information, up from 16% the prior year, with 64% of those users engaging weekly or more often.2

The trend has particularly strong traction among younger cohorts: 45% of Gen Z adults and 48% of Millennials surveyed reported using AI chatbots to find health information.2 A cross-sectional survey published in JMIR Human Factors found that 78.4% of respondents indicated willingness to use ChatGPT for self-diagnosis.3

Most commonly, patients appear to be seeking treatment options based on an existing diagnosis and conducting pre-diagnosis symptom checking, suggesting that clinical hypotheses formed via AI travel with patients into the consultation room.

Writing in NEJM AI, Blumenthal and Goldberg characterise patient use of AI (PAI) as widespread, noting both potential benefits and risks, and highlight the need to help patients engage with these technologies safely and effectively.4 A companion perspective by Goldberg observes that even as the medical establishment cautiously evaluates generative AI for clinical use, patients are already deploying it for diagnostic functions — often without formal guidance.5

2. THE ACCURACY PROBLEM: WHAT PATIENTS MAY BE GETTING FROM AI

Patient adoption of AI health tools would have limited clinical impact if AI consistently delivered accurate, personalised medical guidance. The evidence, however, is considerably more nuanced. A perspective in the New England Journal of Medicine warned that false responses by GPT-4 — so-called “hallucinations” — may be particularly consequential in medical contexts because such errors can be subtle and stated with apparent confidence.6

A development and usability study in JMIR Medical Informatics evaluated reference hallucination rates across multiple AI chatbots and found that general-purpose tools such as ChatGPT and Bing exhibited critical hallucination levels, while more specialised research tools showed negligible rates.7

In cardiology, a simulation study evaluated ChatGPT 3.5 and 4.0 across 25 clinical scenarios pairing common cardiovascular diseases with major comorbidities. Performance deteriorated in complex, multi-morbidity contexts, precisely the patient populations that may most require reliable guidance.8

A scoping review in The Oncologist covering 60 studies found that while LLMs showed some value as health information resources in oncology, notable limitations included variable reliability, hallucinations, and the risk of generating plausible but inaccurate content.9

Drug-drug interaction (DDI) accuracy represents another area of concern. A peer-reviewed study found that ChatGPT-3.5 exhibited limited understanding of contextual patient factors, producing outputs that diverged meaningfully from pharmacist evaluations — a limitation attributed in part to reliance on unverified information sources.10

A mixed-methods survey and interview study published in JMIR found that physicians acknowledged the challenge posed by AI-generated health information, particularly where low AI literacy among patients produced biased expectations that could affect the patient-physician relationship.11

3. HOW THE AI-INFORMED PATIENT MAY CHANGE THE HCP CONSULTATION

Online health information-seeking before medical consultations is not new. A systematic review in JMIR found that patients who researched their conditions online before visits reported feeling more empowered and asked more focused questions. Physicians found some such encounters productive, while others were complicated by misinformation that required time to address.12

Generative AI may amplify both dynamics. Unlike a search engine that returns multiple sources for the user to evaluate, an LLM delivers a single, synthesised answer. Research summarized by the National Academy of Medicine found that patients may have difficulty distinguishing AI-generated advice from physician-generated advice, underscoring the importance of what the authors term “critical AI health literacy.” 13

A survey experiment study published in JMIR found that information about provider oversight, regulatory approval, and device performance significantly increased the likelihood of patient trust in AI by approximately 14–19%.14

A survey study in JAMA Network Open found mixed levels of trust among U.S. adults regarding whether health systems use AI responsibly — suggesting that the presence of AI in or around the clinical encounter introduces new trust dynamics that neither patients nor HCPs have fully mapped.15

From a systems perspective, O Campos et al. describe generative AI as potentially functioning as a “third agent” in the clinical encounter — acting as either a facilitator of or impediment to therapeutic alliance, depending on how patients integrate AI-sourced information into the consultation.16

4. PHARMA MESSAGING IMPLICATIONS

The pharmaceutical industry has long recognised that patient beliefs and HCP communications are interdependent. AI-informed patients introduce a new and more complex variant of this dynamic, with several specific implications for pharma messaging strategy.

4.1  The Content Baseline Has Shifted

Patients arriving at an HCP visit pre-briefed by AI may have formed working assumptions about disease mechanisms, treatment classes, and specific products. Pharma messaging — whether through medical representatives, digital content, or HCP-directed materials — may need to account for the possibility that the patient has already encountered AI-generated information about the brand, a competitor, or the therapeutic class. A 2024 physician survey commissioned by OptimizeRx found that HCPs are increasingly aware that patients bring external health information into consultations, and that aligned information and timing may influence treatment conversations.17

4.2  HCPs May Need Better Tools to Navigate AI-Generated Patient Beliefs

A systematic review in JMIR examining physician perspectives on internet-informed patients identified consistent themes: HCPs value engaged, curious patients, but find it challenging when patient-acquired information is inaccurate, out of context, or used to challenge clinical recommendations.18 These dynamics may be amplified with AI-informed patients, given that AI-generated content can carry the appearance of authoritative synthesis. Medical affairs teams and medical science liaisons (MSLs) may benefit from equipping HCPs with concise, evidence-grounded talking points designed to address the most common AI-generated misconceptions in a given therapeutic area.

4.3  Scientific Rigor as a Factor in AI-Sourced Information Quality

LLMs synthesize information from multiple sources — including clinical trial data, treatment guidelines, patient forums, and product labelling. The quality and accessibility of a brand’s published evidence base may therefore influence what AI presents to patients. Industry commentary has noted that clear formatting, consistent terminology, and transparent citations may help AI systems extract and present pharmaceutical information accurately.19

4.4  Patient Activation Prior to the HCP Visit

If patients are using AI to prepare for consultations, there may be an upstream opportunity for pharma to support the quality of that preparation — by ensuring that compliant, scientifically sound, patient-comprehensible information is available and findable. Evidence on health information seeking consistently indicates that emotional engagement with health information is associated with more positive shared decision-making outcomes, while social-media-derived information may have negative effects.20 Evidence-based patient education content that AI can accurately draw upon may therefore complement point-of-care strategies.

4.5  Omnichannel Alignment

A 2025 pharma marketing analysis describes an emerging paradigm in which educational messaging reaches HCPs and patients in parallel, with the aim of reducing the burden on physicians while increasing patient engagement.21 When the scientific narrative that a patient encounters via AI before a visit is coherent with the evidence-based messaging their physician receives from pharma channels, the consultation may be more productive. Conversely, fragmentation — contradictory messaging across patient and HCP channels — may create confusion.

5. REGULATORY AND ETHICAL CONSIDERATIONS

At its 2023 Annual Meeting, the American Medical Association called on physicians to educate patients about the benefits and risks of AI-generated medical advice, and directed the association to work with the federal government to protect patients from false or misleading AI-generated information.22 Blumenthal and Goldberg identify constitutional limits on government’s regulatory authority over patient AI use, but also highlight opportunities for private actors — including healthcare companies — to help patients engage with generative AI more safely.4

For pharma specifically, these dynamics may create both responsibility and opportunity. Brands that develop resources to help patients critically evaluate AI-generated health information — and that share those resources with HCP partners — may build trust with both audiences. This is consistent with a broader shift in medical communications toward transparent, evidence-grounded content, as described in the CREATE trust communication framework published in JAMA Internal Medicine.23

6. PRACTICAL RECOMMENDATIONS FOR PHARMA MEDICAL AND COMMERCIAL TEAMS

Based on the evidence reviewed, the following considerations may be relevant for pharma companies:

  • Conduct AI landscape audits per therapeutic area. Systematically query major LLMs with questions your target patients are likely to ask, and benchmark responses against approved data and labelling. Identify gaps or misconceptions to address through publication strategy and patient education.
  • Equip HCPs with AI-ready discussion guides. Develop evidence-based talking points designed to address common AI-generated narratives patients may present. Frame these as complementary clinical context rather than rebuttals.
  • Invest in publication quality and accessibility. Robust peer-reviewed evidence, structured abstracts, plain-language summaries, and patient-accessible content may increase the likelihood that AI systems surface accurate information about your therapy.
  • Align patient and HCP messaging architectures. The scientific narrative that an HCP hears from a representative, the information a patient finds via AI, and the content provided in digital patient support programmes should be coherent. Misalignment may erode trust.
  • Develop AI health literacy resources. Consider co-developing, with patient advocacy groups and HCP societies, brief guides to critical AI health literacy — what to ask AI, how to evaluate its responses, and when to seek physician input.

CONCLUSION

The AI-informed patient is an emerging and growing phenomenon. Generative AI has increased access to health information while also introducing risks of misinformation, misplaced confidence, and clinical confusion. For pharmaceutical companies, these dynamics suggest that the competition for therapeutic trust begins before the patient enters the HCP’s office.

Pharma medical and commercial teams that understand this shift and build messaging strategies robust enough to work in an environment where patients arrive already briefed may be better positioned to support productive HCP conversations and improve treatment outcomes. The evidence base reviewed here is still maturing; recommendations should be adapted as new data emerge.

References

  1. Yau JY, Saadat S, Hsu E, Murphy LS, Roh JS, Suchard J, Tapia A, Wiechmann W, Langdorf MI. Accuracy of Prospective Assessments of 4 Large Language Model Chatbot Responses to Patient Questions About Emergency Care: Experimental Comparative Study. J Med Internet Res. 2024 Nov 4;26:e60291. doi: 10.2196/60291. https://pubmed.ncbi.nlm.nih.gov/39496149/
  2. Rock Health. 2025 Consumer Adoption of Digital Health Survey. San Francisco: Rock Health; 2026. https://rockhealth.com/insights/the-tortoise-and-the-hare-of-care-health-ai-insights-from-rock-healths-2025-consumer-adoption-survey/ Last accessed 07/06/2026
  3. Shahsavar Y, Choudhury A. User Intentions to Use ChatGPT for Self-Diagnosis and Health-Related Purposes: Cross-sectional Survey Study. JMIR Hum Factors. 2023 May 17;10:e47564. doi: 10.2196/47564. https://pubmed.ncbi.nlm.nih.gov/37195756/
  4. Blumenthal D, Goldberg C. Managing Patient Use of Generative Health AI. NEJM AI. 2025;2(1). doi:10.1056/AIpc2400927 https://ai.nejm.org/doi/abs/10.1056/AIpc2400927
  5. Goldberg C. When Patients Take AI into Their Own Hands [Patient Portal]. NEJM AI. 2024;1(5). doi:10.1056/AIp2400283 https://ai.nejm.org/doi/full/10.1056/AIp2400283
  6. Lee P, Bubeck S, Petro J. Benefits, Limits, and Risks of GPT-4 as an AI Chatbot for Medicine. N Engl J Med. 2023 Mar 30;388(13):1233-1239. doi: 10.1056/NEJMsr2214184. PMID: 36988602. https://pubmed.ncbi.nlm.nih.gov/36988602/
  7. Aljamaan F, Temsah MH, Altamimi I, Al-Eyadhy A, Jamal A, Alhasan K, Mesallam TA, Farahat M, Malki KH. Reference Hallucination Score for Medical Artificial Intelligence Chatbots: Development and Usability Study. JMIR Med Inform. 2024 Jul 31;12:e54345. doi: 10.2196/54345. PMID: 39083799; PMCID: PMC11325115. https://pubmed.ncbi.nlm.nih.gov/39083799/
  8. Hao WR, Chen CC, Chen K, Li LC, Chiu CC, Yang TY, Jong HC, Yang HC, Huang CW, Liu JC, Li YJ. ChatGPT Performance Deteriorated in Patients with Comorbidities When Providing Cardiological Therapeutic Consultations. Healthcare (Basel). 2025 Jul 3;13(13):1598. doi: 10.3390/healthcare13131598. PMID: 40648622; PMCID: PMC12249446. https://pubmed.ncbi.nlm.nih.gov/40648622/
  9. Chen D, Avison K, Alnassar S, Huang RS, Raman S. Medical accuracy of artificial intelligence chatbots in oncology: a scoping review. Oncologist. 2025 Apr 4;30(4):oyaf038. doi: 10.1093/oncolo/oyaf038. PMID: 40285677; PMCID: PMC12032582. https://pubmed.ncbi.nlm.nih.gov/40285677/
  10. Radha Krishnan RP, Hung EH, Ashford M, Edillo CE, Gardner C, Hatrick HB, Kim B, Lai AWY, Li X, Zhao YX, Raubenheimer JE. Evaluating the capability of ChatGPT in predicting drug-drug interactions: Real-world evidence using hospitalized patient data. Br J Clin Pharmacol. 2024 Dec;90(12):3361-3366. doi: 10.1111/bcp.16275. Epub 2024 Oct 2. PMID: 39359001; PMCID: PMC11602951. https://pubmed.ncbi.nlm.nih.gov/39359001/
  11. Heinrichs H, Kies A, Nagel SK, Kiessling F. Physicians’ Attitudes Toward Artificial Intelligence in Medicine: Mixed Methods Survey and Interview Study. J Med Internet Res. 2025 Aug 26;27:e74187. doi: 10.2196/74187. PMID: 40857713; PMCID: PMC12421205. https://pubmed.ncbi.nlm.nih.gov/40857713/
  12. Tan SS, Goonawardene N. Internet Health Information Seeking and the Patient-Physician Relationship: A Systematic Review. J Med Internet Res. 2017 Jan 19;19(1):e9. doi: 10.2196/jmir.5729. PMID: 28104579; PMCID: PMC5290294. https://pubmed.ncbi.nlm.nih.gov/28104579/
  13. Campos H, Salmi L. Critical AI Health Literacy as Liberation Technology: A New Skill for Patient Empowerment. NAM Perspect. 2025 Dec 8;2025:10.31478/202512a. doi: 10.31478/202512a. PMID: 41970622; PMCID: PMC13064910. https://pmc.ncbi.nlm.nih.gov/articles/PMC13064910/
  14. Zhu X, Stroud AM, Minteer SA, Yoo DW, Ridgeway JL, Mooghali M, Miller JE, Barry BA. Key Information Influencing Patient Decision-Making About AI in Health Care: Survey Experiment Study. J Med Internet Res. 2026 Jan 12;28:e75615. doi: 10.2196/75615. PMID: 41525463; PMCID: PMC12795307. https://pubmed.ncbi.nlm.nih.gov/41525463/
  15. Nong P, Platt J. Patients’ Trust in Health Systems to Use Artificial Intelligence. JAMA Netw Open. 2025 Feb 3;8(2):e2460628. doi: 10.1001/jamanetworkopen.2024.60628. PMID: 39951270; PMCID: PMC11829222. https://pubmed.ncbi.nlm.nih.gov/39951270/
  16. Campos HO, Wolfe D, Luan H, Sim I. Generative AI as Third Agent: Large Language Models and the Transformation of the Clinician-Patient Relationship. J Particip Med. 2025 Aug 11;17:e68146. doi: 10.2196/68146. PMID: 40826831; PMCID: PMC12361538. https://pubmed.ncbi.nlm.nih.gov/40826831/
  17. OptimizeRx. Conversations that Convert: 2024 Physician Survey on HCP+DTC Pharma Marketing Strategies. OptimizeRx Corporation; 2024. [Company-commissioned; not peer-reviewed. Disclose as such in any external use.] Available from: https://www.optimizerx.com/white-papers/conversations-that-convert-2024-survey-pharma-marketing-strategies
  18. Lu Q, Schulz PJ. Physician Perspectives on Internet-Informed Patients: Systematic Review. J Med Internet Res. 2024 Jun 6;26:e47620. doi: 10.2196/47620. PMID: 38842920; PMCID: PMC11190621. https://pubmed.ncbi.nlm.nih.gov/38842920/
  19. Envision Pharma Group. 10 AI Game-Changers Set to Redefine Pharma and Medical Communications in 2026. Envision Pharma Group; 2026 Jan 21. https://www.envisionpharmagroup.com/news-events/10-ai-game-changers-set-to-redefine-pharma-and-medical-communications-in-2026/ Last accessed 07/06/2026
  20. Song M, Elson J, Haas C, Obasi SN, Sun X, Bastola D. The Effects of Patients’ Health Information Behaviors on Shared Decision-Making: Evaluating the Role of Patients’ Trust in Physicians. Healthcare (Basel). 2025 May 24;13(11):1238. doi: 10.3390/healthcare13111238. PMID: 40508852; PMCID: PMC12155490. https://pmc.ncbi.nlm.nih.gov/articles/PMC12155490/
  21. OptimizeRx. 2025 Pharma Marketing Predictions: AI, Media, and Audience Alignment. OptimizeRx Corporation; 2025. https://www.optimizerx.com/blog/2025-pharma-marketing-predictions
  22. American Medical Association. Educate Patients about Misleading AI-Generated Medical Advice [AMA Policy]. Chicago: AMA; 2023. https://www.ama-assn.org/practice-management/digital-health/educate-patients-about-misleading-ai-generated-medical-advice Last accessed 07/06/2026
  23. Allen MR, Schillinger D, Ayers JW. The CREATE TRUST Communication Framework for Patient Messaging Services. JAMA Intern Med. 2024 Sep 1;184(9):999-1000. doi: 10.1001/jamainternmed.2024.2880. PMID: 39073805. https://pubmed.ncbi.nlm.nih.gov/39073805/

Olivier Gryson, PharmD, MSc
25 years of experience in digital marketing in the pharmaceutical industry
Special focus on AI Search in Pharma Marketing

Frequently Asked Questions

Rock Health’s 2025 Consumer Digital Health Survey (n=8,000 U.S. adults) found 32% used AI chatbots for health information — double the prior year’s 16% — with 64% engaging weekly or more. A U.S. consumer survey cited by Sarabu et al. reported approximately 48% had used generative AI for health questions. These are self-reported U.S. figures and may not generalise to other markets.1,2

Younger adults show the highest reported rates: Rock Health’s 2025 survey found 45% of Gen Z and 48% of Millennials used AI chatbots for health information. However, willingness is not limited to younger cohorts: a JMIR cross-sectional survey found 78.4% of all respondents would be willing to use ChatGPT for self-diagnosis, suggesting broader potential uptake.  [2,3]

Evidence indicates important accuracy limitations. A NEJM perspective warned that GPT-4 hallucinations in medical contexts may be subtle yet stated with high confidence. A JMIR Medical Informatics study found critical hallucination rates in ChatGPT and Bing. In cardiology, AI performance deteriorated significantly in multi-morbidity scenarios. A scoping review in The Oncologist (60 studies) cited poor reliability and hallucinations as key disadvantages. DDI accuracy also showed meaningful gaps versus pharmacist evaluation.6,7,8,9,10

A JMIR systematic review found that online pre-consultation research was associated with increased patient engagement but also with misinformation requiring physician time to correct. Generative AI amplifies these dynamics by delivering a single synthesised answer rather than multiple sources. A 2025 JMIR study found physicians acknowledged AI-sourced information could disrupt the patient-physician relationship when it introduces biased expectations.  [11,12]

Evidence suggests many cannot. Research reviewed by the National Academy of Medicine (2026) found patients may be unable to distinguish AI-generated from physician-generated advice, and may trust inaccurate AI outputs. A JMIR survey experiment found that disclosing provider oversight and regulatory approval status increased patient trust in AI by 14–19%, highlighting that transparency about AI provenance matters.13,14

A 2024 JMIR systematic review found HCPs value engaged patients but struggle with inaccurate, out-of-context, or adversarial patient information. A 2024 OptimizeRx physician survey (company-commissioned) found HCPs are increasingly aware of patients’ externally acquired health information and indicated it may influence treatment conversations when well-timed. HCPs appear to value concise, evidence-based tools that contextualise, rather than simply contradict, patient-sourced information.17,18

LLMs synthesise information from available published sources. The quality, structure, and accessibility of a brand’s evidence base — including peer-reviewed publications, structured abstracts, and plain-language summaries — may influence whether AI accurately represents the therapy. Industry commentary (Envision Pharma Group, 2026; not peer-reviewed) suggests clear formatting, consistent terminology, and transparent citations may help AI systems extract pharmaceutical information correctly.19

The AMA’s 2023 Annual Meeting called on physicians to educate patients about the risks of AI-generated medical advice and directed work with the federal government to protect patients from false or misleading AI content. Blumenthal and Goldberg in NEJM AI note constitutional constraints on government regulation of patient AI use, but highlight significant opportunities for private healthcare actors to support safer patient AI engagement. The regulatory landscape remains evolving.4,22

Alignment between patient-facing and HCP-facing content may be increasingly important. A 2025 OptimizeRx marketing analysis (company-commissioned) describes a parallel-messaging paradigm aimed at reducing burden on physicians while increasing patient engagement. When scientific messaging is coherent across channels, AI-shaped patient beliefs may reinforce rather than complicate HCP communication. The CREATE trust communication framework (JAMA Internal Medicine, 2024) provides relevant principles: transparent, evidence-grounded, patient-centred content.21,23

Critical AI health literacy is defined by the National Academy of Medicine (2026) as the ability to assess, evaluate, and appropriately act on AI-generated health information. Patients with low AI literacy may be more susceptible to acting on inaccurate AI outputs. Pharma companies that co-develop AI literacy resources with patient advocacy groups and HCP societies — helping patients know what to ask AI and when to defer to a clinician — may build trust with both audiences while contributing to a safer information ecosystem.13

Follow the conversation on LinkedIn

I regularly share reflections on pharma marketing, search behavior, and the impact of AI on healthcare communication.

Follow me on LinkedIn

This article was written with the assistance of generative AI technology and reviewed for accuracy.