Evidence-based medicine has spent decades improving the supply of evidence: trials, systematic reviews, guidelines, summaries and search tools.

But there is another side we understand much less well: demand.

What are clinicians actually trying to find out when they reach for evidence?

Not the neat questions found in guidelines or research protocols, but the real-world questions that arise in practice: Is this treatment safe for this patient? Does the evidence apply to older adults? What if the patient has renal impairment, pregnancy or multimorbidity? What should I do when local and national guidance differ?

These questions are not always signs that evidence is missing. Often the evidence exists, but the difficulty lies in applying it.

That distinction matters. A repeated clinical question may point to a research gap, but it may also point to something else: poor dissemination, unclear guidance, uncertainty about applicability, or the challenge of translating evidence into action.

This is where natural-language Q&A systems may offer something new. Search logs – the terms people type into a search box – give us fragments: “atrial fibrillation elderly”, for example. Full clinical questions reveal more: the uncertainty behind the search, the context, and what the clinician is really trying to find out.

AskTrip has now received around 19,000 clinical questions. Looking across them, what stands out is not just the range of topics, but the range of reasons clinicians ask. Some questions are about missing evidence; many are about how existing evidence fits messy, real-world care.

Handled carefully, this kind of question log could become a useful new signal for evidence-based medicine. It could help researchers spot recurring uncertainties, help guideline developers see where recommendations are unclear or not reaching practice, and help those building medical AI systems test against real questions rather than artificial benchmarks.

But the cautions matter. Question logs are not neutral. They reflect who uses the system, what the interface encourages, and what people feel comfortable asking. They also need careful governance, de-identification and aggregation.

So this is not an argument for treating question logs as simple truth.

It is an argument for taking them seriously as signals.

Evidence-based medicine will always need to ask: what evidence exists?

But perhaps we should also ask: what are clinicians repeatedly trying to find out?