A patient has never had more health information available to them, or been able to reach it more easily. They can research before the appointment, sit through the explanation in the consultation room, and ask a chatbot to fill the gaps on the way home. By every measure of access, this is the best-informed moment in the history of medicine.
It has not made patients better informed. In 2024-25, NHS written complaints reached a record 256,777, up 6.1 percent on the year. Communication was the single largest category of complaint in hospital care. As information became abundant, complaints about communication did not fall. They rose.
The problem was never access. It is what happens in the three places a patient's information now lives, and what none of them does.
Before the consultation room
Most patients research before they arrive with eight in ten people beginning a health query on a general search engine, and they choose what to read by convenience rather than accuracy.
The information that ranks is rarely written or checked by a clinician. The AI answers they increasingly rely on come with fewer warnings than they used to. In 2022, around a quarter of chatbot responses to health questions carried some note that the model is not a doctor. By 2025, fewer than one in a hundred did.
There is a version of this phase that does work well, though. Preoperative educational video, authored by the clinician, improves patients' knowledge, understanding and preparedness and lowers their anxiety. The patient arrives prepared by their clinician, not by whatever surfaced first in their search.
In the consultation room
Then comes the consultation, and most of it does not survive the walk out of the building, let alone getting home.
The evidence here is old and unflattering and has not been overturned. Patients forget between 40 and 80 percent of what a clinician tells them almost immediately. The more they are told, the smaller the share they keep. So, the conversations that really matter, from the diagnosis to supporting informed consent, are the ones least likely to be remembered, because they are the ones delivered under the most stress.
This is just how memory works, and it makes a single spoken explanation, however clear, the wrong format for information a frightened person needs to hold onto.
After the consultation room
Patients then fill the gaps later and they increasingly fill them with AI. More than 40 million people now use ChatGPT every day for health questions. Seven in ten of those conversations happen outside clinic hours, in the evenings and at night, which is precisely when the clinician is not reachable and the questions arrive.
The instinct is to ask whether the answers are accurate. That is the wrong question. The more useful finding is about context.
In the largest study of its kind, published in Nature Medicine this year, Oxford researchers had people use leading chatbots to work through real medical scenarios. The people using AI did no better than people relying on ordinary methods. More telling, the model on its own, handed the full scenario, did better than the same model paired with an actual patient. The patient was the bottleneck.
Anyone who has spent time with these tools knows why. The quality of the answer depends almost entirely on the context in the question. A clinician knows which details matter, however a patient, most often, does not. They cannot supply the history, the comparison, the relevant detail that would change the answer, because knowing what is relevant is the expertise they are trying to borrow. The tool is only as good as a prompt the patient is not equipped to write.
The pattern
Before the consultation, the information is not clinician-curated. During it, the information is forgotten. After it, the information is missing the context only the clinician could have added.
Abundance does not touch any of these. More search results, more leaflets, more chatbot answers. None of it closes a gap that was never about volume. We have spent two decades making medical information easier to find and have left the harder problem almost exactly where it was, which is whether the patient understands and trusts what they are told.
The answer is not to take the clinician out of any of these moments. It is to let the clinician's explanation reach into all three. Content the patient can prepare with before they arrive, return to after they leave, and trust because it carries the name and judgement of the expert that is treating them or people just like them. Authored by the clinician, in the clinician's words, available in the patient's language and at the patient's own pace.
As medical misinformation rises and trust in content drops, the answer comes from where it has always been when it comes to medical information: the clinical professionals.
References
- NHS England Digital (2025). Data on Written Complaints in the NHS, 2024-25.
- Kessels, R.P.C. (2003). Patients' Memory for Medical Information. Journal of the Royal Society of Medicine, 96(5), 219-222.
- Bean, A., Payne, R., et al. (2026). Reliability of large language models as medical assistants for the general public: a randomized, preregistered study. Nature Medicine. Oxford Internet Institute and Nuffield Department of Primary Care Health Sciences.
- OpenAI (2026). AI as a Healthcare Ally: How Americans Are Navigating the System With ChatGPT. Reported in Fierce Healthcare and Healthcare Dive, January 2026.
- Effectiveness of the video medium to supplement preoperative patient education: A systematic review (2022). Patient Education and Counseling.