Three recent questions submitted to AskTrip made me think about the relationship between the evidence we make available and the knowledge people actually need. They all came in this morning – one at 6.46am, the next at 6.52am and the final one at 7.01am – my assumption is these were all from the same person!

All three concerned sexual and reproductive rights: how they differ across countries, the distinction between sexual rights and reproductive rights, and how sexual rights are defined in the literature.

AskTrip produced answers, but the material it found was limited or only partly relevant. Reading them, I wondered whether a general-purpose AI model, drawing on broader sources, might have been better suited to these questions.

Trip is primarily focused on evidence to inform healthcare. These questions sit at the intersection of health, law, ethics and public policy. That makes them useful examples of what happens when a knowledge need extends beyond the strongest coverage of an evidence resource.

But looking more closely, there was something else going on.

The answers sometimes assessed the available material using criteria that did not fit the question.

The first question asked how sexual and reproductive rights compare across countries. The answer opened:

“Evidence on sexual and reproductive rights shows significant disparities in service quality and access across countries…”

It went on to discuss sexual health services, STI screening and sexual harassment. These are relevant to the realisation of rights, but they provide only part of the comparison requested. A reader could finish the answer knowing that access to services varies, while still knowing little about how countries’ legal protections differ.

The answer acknowledged this:

“many studies discuss service provision or healthcare outcomes rather than explicitly comparing legal frameworks”

That is an important qualification. It also identifies the central problem: the retrieved material largely answers a neighbouring question.

A more useful approach would first establish which countries and which rights the user wanted to compare, then examine legal protections alongside practical access. A country comparison needs both, with dates and sources attached.

The second question asked about the difference between sexual rights and reproductive rights. Its opening distinction was helpful:

“Sexual rights focus on the right to safe, consensual, and pleasurable experiences related to sexuality, while reproductive rights emphasize the ability to make informed decisions about reproduction…”

However, the limitations section criticised the available material for lacking:

“exhaustive empirical validation of the distinctions”

This seems to apply the wrong test. The user is asking for a conceptual explanation. Definitions need attribution, clarity and attention to areas of disagreement. Their usefulness does not depend on being validated in the same way as a treatment effect.

An illustrative alternative would be:

Sexual rights concern autonomy, equality and freedom from coercion in relation to sexuality and sexual relationships. Reproductive rights concern decisions about reproduction, including whether and when to have children. They overlap, but sexuality extends beyond reproduction. A fuller answer should explain how particular organisations define the terms and where those formulations differ.

That is a starting explanation, rather than a complete review. It addresses the distinction directly and makes clear what further sourcing would add.

Significantly, AskTrip had already searched beyond Trip for this answer and found a WHO source. Broader retrieval helped, but the answer still judged the source through an inappropriate empirical lens.

The third question asked how sexual rights are defined in the relevant literature. The answer assembled themes from literature concerning different populations, then observed:

“The literature heavily leans on international conventions and human rights frameworks”

For this question, those frameworks could be central to a good answer. WHO’s sexual health page, for example, explicitly presents a working definition of sexual rights grounded in existing human rights.

A more direct answer could begin with that definition, explain its scope, and compare it with other relevant formulations. It would distinguish an organisation’s definition from rights enforceable in a particular jurisdiction. Population-specific research could then show how those rights are understood or experienced in practice.

Across these examples, I see three questions for an evidence service:

  • Do we hold the sources needed to answer the question?
  • Have we retrieved material that directly addresses it?
  • Are we judging that material using criteria appropriate to the knowledge being sought?

These are separate problems. Adding sources will not necessarily fix an answer that treats a definition as an inadequately validated empirical claim.

Would a general-purpose model do better? Probably. Its broader knowledge could help it explain concepts and recognise relevant legal or ethical frameworks. But fluent writing is not proof of accuracy. Current country comparisons still require checked sources, and a fair comparison would assess accuracy, relevance and attribution rather than how convincing the answers sound. The alternatives above illustrate a different approach; they are not a formal test of competing systems.

For me, the most useful lesson concerns the phrase “limited evidence”. It can describe a genuine lack of research, a gap in a database, or a search that has failed to find suitable material. Those meanings should not be interchangeable.

“The sources we found do not adequately answer this question” is sometimes the more accurate statement.

A question can concern health while requiring knowledge that biomedical research alone cannot supply. The challenge for AskTrip – and for evidence services more broadly – is to recognise the knowledge need, find appropriate sources, and be clear when the fit is poor.