The discovery gap: Why good research still goes unseen

The discovery gap: Why good research still goes unseen

Imagine a materials science lab solving the exact manufacturing problem an aerospace supplier has been chasing for two years. Neither party knows the other exists. Unwittingly, the supplier licenses a lesser solution produced by a lab three time zones away, not because that lab's work is better, but because it is easier to find.

This kind of near miss happens constantly, and it rarely gets logged anywhere. Researchers do not tend to cite "lost partnership because our researcher profile page ranked on page four of a Google search" in their annual report. But multiply that single miss across a research portfolio or a funding cycle and the pattern becomes hard to ignore. Besides lost funding opportunities and industry partnerships, there can be reputational fallout affecting university rankings, recruitment and societal impact — not to mention wasted resources, disrupted careers and the erosion of staff morale. The ripples just keep spreading.

Senior research leaders already know their institutions produce work worth finding. The harder question is whether that work can actually be found by the people and systems that are searching.

How research visibility used to work

For most of the last two decades, "research visibility" meant something fairly simple: a staff directory, a departmental page, maybe a press release when something newsworthy happened. Those assets did their job as long as the audience already knew roughly who or what they were looking for.

A journalist needing a quote on a specific disease pathway could search a university's medical school site and most likely find someone. A program officer who already knew a principal investigator's name could locate their publication list without much trouble.

Why search now starts with the problem, not the institution

The assumption that the searcher already knows where to look is quietly breaking down. Industry partners, funders and policymakers are increasingly starting from a problem or a capability rather than an institution's name or a person's title. A company chasing a materials science breakthrough, for example, does not search "university engineering department." It searches the problem itself, expecting to be matched with whoever can solve it, wherever they happen to work. This type of search is also increasingly unlikely to be conducted by a person clicking through a website, but rather by an algorithm: a funder portal cross-checking disclosures or, increasingly, an AI system synthesizing an answer from whatever it can parse and trust.

This shift changes who wins.

Two institutions can produce research of comparable quality and see very different outcomes with the industry partners, academic collaborators and funders that matter to them, purely because one organized its expertise so that a machine could find, parse and trust it — the other did not.

That is not a hypothetical concern. The UK’s Konfer platformopens in new tab/window is designed to connect industry partners with academics. While it has cataloged over 1.8 million research items and counting, incorrectly structured university pages never make the list simply because Konfer misses them.

Three questions that reveal the gap

A few questions are worth considering here, as they tend to surface the gap faster than any audit could.

1. When an industry partner searches for expertise in a specific field, do they find your researchers, or a competitor's?

2. If a funder portal or an AI search tool queried your institution's current research portfolio, what would it actually surface?

3. Can your institution tell a complete story that connects a given piece of research to the patents, policy citations and real-world outcomes that followed from it, or does that story live in six different people's inboxes?

What decides the outcome is whether the systems doing the matching can see that the research is good, in a form they can trust.

The new variable: AI-powered search

AI-powered search is the new wrinkle changing the shape of the discovery problem, not just its scale.

Search and answer tools that use AI do not browse a website the way a person does. They synthesize an answer from whatever content they can crawl, parse and cross-reference, and they tend to favor sources that are structured, consistent and verifiable over sources that are simply well-written.

A faculty bio with elegant prose and no underlying structure may never surface in an AI-generated answer about who is working on a given problem. Compare that to a competitor's plainer but properly tagged and disambiguated profile, and the winner is invariably the better-structured piece.

This is not currently a fully mapped-out discipline. Answer-engine optimization is still an emerging practice rather than a settled one, and institutions are largely finding their way through it in real time. What is not in question, however, is what research leaders actually care about most.

Funders, industry evaluators and academic collaborators are increasingly relying on these same AI-mediated tools at some point in their process, even when a human still makes the final call.

A website is not a visibility strategy

Most research leaders would say their institution is visible. There is a functioning website and directory, and faculty appear on Google when someone searches their name. On the surface, that looks like enough.

In practice, the traditional model of visibility has always had a ceiling. Expertise buried three or four clicks deep in a departmental site, written in language built for peer reviewers rather than outside audiences, or scattered across pages that do not reference each other, will be unable to reach people who do not already know where to look.

What staying under the radar can cost

The costs of this kind of invisibility are familiar to anyone who has run a research office for an extended period. They tend to fall into three buckets:

1. Financial cost:

   A missed grant cycle, a partnership that goes to a better-known competitor, or philanthropic revenue that never materializes because a prospective donor could not easily see the impact of their gift.

2. Reputational cost:

   A public profile that consistently undersells an institution's actual research strength, compounds over time as funders and journalists form impressions based on what is visible rather than what is true.

3. Time cost:

   Research office staff manually chasing faculty for bios, updated CVs and case studies instead of completing the higher-value work of building relationships with funders and industry contacts.

Why this cost has always been recoverable, until now

Invisibility of this traditional kind has always been, to some degree, recoverable. A missed opportunity this cycle can be won back next cycle with a sharper pitch, an updated website or a well-placed press release. The damage is real, but it resets, and this is precisely what makes it easy to deprioritize.

If the cost of underinvesting in research visibility is a bad year rather than a structural disadvantage, it is tempting to treat it as a communications problem to solve later rather than an institutional one to solve now.

A familiar scenario: The cross-disciplinary grant

Consider how this plays out in a fairly ordinary scenario.

A funder scans applications for a cross-disciplinary grant that touches public health and materials engineering, two departments that — at most universities — maintain entirely separate web presences with no shared vocabulary between them.

An institution where public health researchers and materials engineers present as a connected, cross-referenced body of work — rather than two unrelated departmental listings — tells a more compelling story about its capacity to deliver on that grant, even if the underlying research quality is identical to a competitor's. An institution that cannot tell a connected story will lose because the story never got told in a form the funder could quickly verify.

Funder verification of university websites becomes even more important in the age of AI. There are two main reasons for this: 

1. Grant proposals developed with AI can include hallucinated or fabricated information; the onus is on the funder to authenticate claims.

2. AI-assisted grant submissions overwhelm funders’ systems. Funders are increasingly looking at verification methods beyond proposals themselves, often involving researcher context and how it appears on university websites.

Comfort with an increasingly ineffective status quo is getting harder to justify, not because the traditional cost of invisibility has worsened, but because a second, less forgiving cost is appearing alongside it.

Passing the human test, failing the machine test

Here is the uncomfortable part for institutions feeling confident about their web presence. Looking fine to a human visitor and being discoverable to a machine are no longer the same test, and an institution can pass one while failing the other.

Academic collaborators, funders and industry partners are moving — unevenly but steadily — from browsing institutional websites to querying structured databases and AI-powered tools.

These tools search by capability and problem rather than department name or job title, the same way an industry buyer now does. Research that reads perfectly well to a human on a faculty page can be functionally invisible to the system doing the actual matching if that page is not structured, indexed and disambiguated in a way a machine can parse.

Translation? A beautifully written bio means very little to a large language model trying to determine whether a given researcher is, in fact, the person who actually published the paper it is citing.

Why this cost compounds instead of resetting

What makes this cost different from the traditional visibility gap is that it does not reset, but compounds.

An institution that falls behind in how AI systems and structured databases index and represent its expertise is also further behind next year, because rankings, funding decisions and partnership shortlists increasingly draw on the same underlying structured signals.

Research on metadata and discoverability directly backs this. A study of more than 100 scholarly booksopens in new tab/window found that titles with persistent identifiers and well-structured metadata were consistently easier to find in Google Scholar than those without, even though the search engine claims not to use that metadata directly in its index. The structure behind the content, not just the content itself, is doing real work in determining what gets found.

The disambiguation problem

Disambiguation is a good illustration of how granular this problem gets. A researcher who has published under a maiden name, moved between three institutions over a career or simply shares a common surname with several other academics in their field creates exactly the kind of ambiguity that a human reader can usually resolve with a little context, but that an automated system either resolves correctly through structured identifiers or gets wrong at scale.

Persistent researcher identifiers exist precisely to solve this, but an identifier only helps if an institution's underlying records are clean and consistently linked to it. A research information system full of duplicate, unclaimed or unlinked profiles can actively degrade an institution's discoverability whenever a rankings funder database or AI tool has to guess which "J. Smith" actually did the work.

Compliance is raising the stakes

Compliance pressure sharpens this further, on both sides of the Atlantic.

In the United States, the National Security Presidential Memorandum-33opens in new tab/window (NSPM-33) directs federal agencies to establish consistent disclosure requirements for researchers receiving federal research funding, including information about foreign financial support, affiliations and other relationships that could create conflicts of interest or research security concerns.

The Family Educational Rights and Privacy Act (FERPA) adds a parallel layer of pressure, governing how institutions handle student and researcher education records and reinforcing that the same information cannot be treated one way for compliance purposes and another way for public-facing visibility. Together, these requirements quietly change what a researcher profile is — an institutional record that needs governance, version control and accuracy. An outdated or inconsistent profile is now a compliance exposure, not just a missed marketing opportunity.

Europe carries its own version of this pressure, with data-sovereignty and governance expectations that similarly push researcher information out of the "website content" category and into something closer to institutional infrastructure. Ungoverned profiles, in other words, are underselling an institution while also acting as a liability with an audit trail.

At the same time, the audiences institutions most need to reach are shifting in ways that make discoverability even more consequential.

Follow the funding shift

Government funding of research is shrinking within the United States and much of Europe, even as overall research spending grows. In 2023, businesses funded 75% of total U.S. research and developmentopens in new tab/window performed across all sectors, compared with 18% funded by the federal government. Between 2012 and 2023, the business share of basic research funding rose from 21% to an estimated 35%, while the federal government's share fell from 52% to 41%. The pattern holds across the EU, where an analysis of R&D expenditure by source of fundsopens in new tab/window shows that the government share fell from 32.9% in 2013 to 29.3% in 2019, before rising only modestly since, while business funding grew to 56.7% of the total by 2023. Japan and South Korea show a different story: both have run business-dominated R&D funding models for decades — Japanese enterprises have long accounted for roughly 70% of national R&D spendingopens in new tab/window, and Korean government allocations have held steady around 20%opens in new tab/window for years — so what's unfolding in the US and Europe isn't a shift toward an already-established Asian model, it's a newer and still-open-ended departure from their own postwar funding settlement.

Industry, in short, is doing more of the work that government funding used to do, which means the institutions best positioned to be found and trusted by industry partners are the ones best positioned to backfill what public funding no longer covers on its own. And yet the track record of universities actually capturing that industry relationship, at scale, is not strong. Public research institutions have historically accounted for a strikingly small share of the innovation pipeline that industry cares most about.

Joint patenting activity between public research institutions and industry represented only about 1.6% of total patent applicationsopens in new tab/window across Organisation for Economic Co-operation and Development countries as of the last major assessment. This is not due to a lack of knowledge.

Universities are not short on patentable, licensable, collaboration-worthy research. The problem is matching, which is exactly what structured, machine-readable discoverability is built to solve.

It is also worth noting that industry funding replacing state funding is not a like-for-like swap. The former is overwhelmingly focused on applied research with a short time to market, ensuring ROI but neglecting fundamental research. If universities want to get funding for fundamental projects, they must work harder — and that includes efforts directed toward showcasing research.

The stakes get higher at the consortium level

This gap tends to widen further at the level of cross-institutional and consortium-based work.

National research programs, regional innovation clusters and multi-university alliances increasingly need to present a single image of capability to industry and government funders — not just a collection of individual profiles. A funder or industry partner evaluating a research alliance wants to see the group’s collective strength, disambiguated and cross-referenced, not five separate websites with five separate formats and no shared structure between them.

Institutions that have solved discoverability at the individual researcher level still often fail this test at the consortium level, simply because the underlying records were never built to be combined in the first place.

Together, these forces point in the same direction. Academic collaborators, funders and industry partners — the audiences with the most influence over an institution's funding, reputation and partnerships — increasingly find and evaluate research through systems that reward structure over prose. Institutions that adapt to that shift pull further ahead each year, while institutions that do not fall further behind.

A governance problem, not a technology one

None of this is really a technology problem in the narrow sense, despite having a technical component. The problem is governance.

Research information — once treated as a scattered collection of bios, CVs and departmental listings maintained by whoever had time — has to function as a single, trustworthy and machine-legible institutional asset, because that is what the audiences on the other end increasingly expect.

This has real implications for how research leaders should evaluate the systems they use to manage and present this information, whether that is a research information management system (RIMS) already in place, one under evaluation or an internally built platform. It is worth resisting the urge to treat this as a straightforward buy-versus-build decision or a matter of picking the tool with the longest feature list. The institutions that get the most out of these systems tend to be the ones that first agree on what "discoverable" actually needs to mean for their researchers, funders and industry partners. Only then should they evaluate whether a given RIMS can deliver.

A few qualities matter more than a feature checklist would suggest:.

• A single source of truth for researcher and output data, so that a profile, a publication record and a compliance disclosure do not quietly drift out of sync as three separate documents maintained by three different people.

• Genuine researcher and affiliation disambiguation, not simple name matching, so that someone with common initials or a career spanning multiple institutions gets attributed correctly by collaborators, funders and AI systems alike.

• Underneath both of these sits a requirement that is easy to overlook: structured, standards-based metadata and markup that machines can actually parse, rather than well-written prose that only a human reader can appreciate. Because compliance has become inseparable from visibility, governance workflows need to be built around disclosure requirements directly, not bolted onto a communications publishing process after the fact, so that NSPM-33, FERPA and equivalent requirements are met as a byproduct of good data hygiene rather than a separate scramble every reporting cycle.

• The ability to trace a given piece of research to its downstream impact — whether that shows up as a patent, a citation in policy or a shift in practice — in a form that is structured enough for an algorithm to credit and specific enough for a person to believe.

The AI-mediated discovery layer that is emerging around all of this is still taking shape. No institution has this fully solved yet, and anyone claiming otherwise is getting ahead of the evidence. What is already clear is the direction the RIMS category itself is moving — toward research information that is more structured and automated. The institutions that treat their research information as governed infrastructure, rather than scattered content, are the ones that will keep showing up when it matters. The research your institution produces may already be strong enough to win the next grant or the next industry partnership. The open question is whether it is structured well enough to be found.