When Everyone Can Create Content, Trust Becomes the Product

For most of modern medicine, access to knowledge has been the problem. Scientific information was expensive to produce, difficult to distribute, and concentrated in universities, medical journals, conferences, and professional societies. 

The internet transformed distribution. Artificial intelligence is changing something more fundamental: who can produce medical content, how quickly it can be created, and how difficult it is becoming to distinguish between information that looks authoritative and knowledge that deserves to be trusted. A clinical summary, literature review, patient explanation, or analysis of a scientific paper can now be produced in seconds, translated, simplified, and distributed almost instantly. 

This is an extraordinary achievement, but it creates a paradox. As credible-looking medical information becomes cheaper to produce, trust becomes harder to manufacture. The answer to infinite medical information may not be fewer voices, but better signals of who — and what — deserves to be trusted. 

This is no longer hypothetical. In 2026, 81% of physicians surveyed by the American Medical Association reported using AI professionally, more than double the 38% reported in 2023. The most common reported use was summarizing medical research and standards of care, cited by 39% of respondents.(1) 

AI is already positioning itself between scientific evidence and the professional who needs to interpret it. 

There are compelling reasons for this adoption. The volume and complexity of medical knowledge exceed what any individual can reasonably process. AI can search enormous bodies of information, synthesize evidence, and reduce the time professionals spend navigating it. Used well, this should make medicine better. But what happens when the tools physicians use to navigate knowledge can reproduce the language and appearance of expertise without necessarily reproducing the mechanisms that made that expertise trustworthy? 

Much of the debate has focused on hallucinations. The concern is legitimate, but the larger issue is not simply that AI can be wrong. Physicians make mistakes, journals publish corrections, and guidelines change. AI introduces a different problem: incorrect information can be delivered with much the same fluency and apparent sophistication as correct information. 

The World Health Organization has warned that large language models can generate authoritative-sounding responses that are false or seriously inaccurate. Its guidance also highlights automation bias: the risk that clinicians or patients may overlook errors or delegate decisions that should remain subject to human judgment.(2) 

Research increasingly supports that concern. A 2025 study of AI-generated clinical summaries manually evaluated 12,999 sentences and found hallucinations in 1.47% of them. That number may sound reassuring, yet 44% of the hallucinated sentences were classified as major because they could affect diagnosis or management if left uncorrected(3). 

Separately, research published in Nature Communications has shown that strong performance on conventional medical questions does not necessarily mean a model reliably recognizes the limits of its own knowledge. 

These findings point to something more consequential: medicine can no longer assume that the appearance of authority is evidence of authority itself. For generations, medical information traveled with contextual signals. The name of a journal meant something. So did the credentials of an author, the reputation of a university, the endorsement of a scientific society, or the institution where a physician practiced. None guaranteed correctness, but together they helped professionals decide where to allocate two scarce resources: attention and trust. 

Generative AI can reproduce many of those external signals remarkably well. It can write in academic language, organize an argument, cite literature, and explain complex science with apparent confidence. An expert analysis and an unreliable AI-generated explanation can increasingly look similar on the surface. Plausibility and credibility can no longer be treated as synonyms. If presentation becomes a weaker proxy for quality, the question shifts from “What does this content say?” to “Why should I trust it?” Who produced it? What evidence supports it? Can a claim be traced to its original source? Is that evidence current? Was AI involved in transforming it? Who reviewed it? And is there an identifiable professional or institution prepared to stand behind it? 

These are becoming questions of infrastructure. Healthcare increasingly needs systems that make credibility visible alongside content: verified professional identities, traceable scientific sources, transparent review mechanisms, and identifiable institutional accountability. Trust should not simply be declared through a badge or label. It should be inspectable. That means preserving provenance around medical knowledge. A physician should be able to understand who produced a claim, what evidence supports it, how that evidence was summarized, who reviewed the result, and when it was last validated. 

AI-generated content does not necessarily need to be treated as inferior; it needs to exist within a system where its origin and evidentiary basis can be examined. The answer to information abundance is not a return to old gatekeepers. It is better signals. This also changes the role of the physician. If machines can access and synthesize more medical knowledge than any individual could retain, it is tempting to assume that the value of physician knowledge must decline. 

The opposite may prove true. When information was scarce, expertise depended heavily on possessing knowledge. When information becomes abundant, expertise increasingly involves exercising judgment over knowledge. A physician must determine whether evidence applies to a particular patient, reconcile contradictory recommendations, recognize uncertainty, question an algorithmic suggestion, and accept responsibility for the eventual decision. AI can shorten the distance between a question and a possible answer. It cannot remove the responsibility of deciding whether that answer is appropriate. 

Physicians appear to recognize this tension. The AMA survey also found that 88% considered validation of safety and efficacy important for broader AI adoption, while 85% wanted physicians consulted or directly involved in implementation decisions.(4) This is not technological resistance. It is a profession adopting a powerful technology while demanding stronger reasons to trust it. 

The same shift could reshape the organizations that sit between medical knowledge and medical professionals. For years, digital medical information revolved around content, audience, and distribution. Generative AI changes that model. If credible-looking content can be produced at enormous scale and minimal marginal cost, volume alone becomes a less meaningful advantage. Value moves toward what surrounds the content: professional identity, scientific provenance, institutional endorsement, peer relationships, independent review, and reputation. 

For journals, universities, and medical societies, this reinforces functions they have performed for generations. For professional networks such as IntraMed in Latin America or Doximity in the United States, it creates a related opportunity. Their long-term value may depend less on how much information they distribute and more on whether they connect that information to verified professionals, credible institutions, and accountable communities. 

None of this argues for slowing AI adoption. The same technology can help check references against original publications, flag outdated evidence, expose disagreements between guidelines, and preserve provenance as information is summarized. The WHO’s governance framework similarly emphasizes transparency, expert participation, rigorous evaluation, oversight, and accountability.(5) 

But technology alone cannot close the loop. Machines can help generate and cross-check information; responsibility for trusting and acting upon it still belongs to identifiable people and institutions. 

Medicine spent decades building infrastructure to make knowledge more accessible. AI now offers the possibility of making that knowledge easier to navigate, understand, and apply. Access will remain a challenge, particularly across unequal health systems. But alongside it, a different scarcity is emerging. As information becomes easier to create, confidence in that information becomes more valuable. The question is no longer only how to give professionals access to more knowledge, but how to help them determine which knowledge deserves their trust because medical information may become effectively infinite. Human attention will not. Neither will trust.

Sources:

- American Medical Association. 2026 Physician Survey on Augmented Intelligence. AMA Center for Digital Health and AI, March 2026.
- World Health Organization. Ethics and Governance of Artificial Intelligence for Health: Guidance on Large Multi-Modal Models. WHO, 2025.
- A Framework to Assess Clinical Safety and Hallucination Rates of LLMs for Medical Text Summarisation. npj Digital Medicine (2025).
- Nature Communications (2025), study evaluating reliability and limitations of large language models in medical reasoning.