The outputs we don’t count: making talk-based research contributions legible
In-person research presentations are first-class and widely valued – but routinely under-documented, writes Ben Kaube
A university can tell you, to the decimal point, how many papers it published and how much funding it won. Ask the same institution where its researchers presented, which audiences they reached, and what ideas they put into circulation through talks, panels, and briefings, and you get dead conference links, cached programme PDFs, and whatever someone happened to save to a personal drive.
The activity is substantial, while the institutional record is not.
Some of the most consequential research communication happens in formats that sit outside the systems we use to track and report on research. Think of the invited keynote that frames a field’s conversation, the policy evidence session where a body of work becomes actionable, or the public lecture that becomes the entry point for industry partners and prospective students. These are first-class research contributions, widely valued but routinely under-documented.
What can’t be seen clearly is hard to preserve and easy to under-credit.
In a recent survey of over 600 researchers, scholarly video ranked alongside replication studies, software, and research databases in demand for greater recognition, all output types that have seen significant infrastructure investment in recent years, and ahead of conference posters, slides, and preprints (Kaier et al., Learned Publishing, 2025). The demand was strongest in the humanities and social sciences, but present across disciplines and markedly stronger among early-career researchers. The underlying problem is not discipline-specific: talks sit outside the systems that make outputs findable, citable, and reusable.
Why these outputs stay invisible
Talks live in the gaps between systems. There is no default place they “belong”, so they end up on conference websites that disappear after a redesign, departmental pages maintained intermittently, personal video accounts, or shared drives owned by someone who has since moved on.
Even when a recording exists, it can be functionally lost. Titles are vague, co-speakers aren’t listed, affiliations are out of date, and the link to the underlying research is lacking. Without persistent identifiers, it becomes difficult to connect “this talk” to “this researcher” and “this institution”, or to assemble a defensible picture of activity at scale. Papers and datasets have DOIs. Grants are increasingly registered with Crossref and linked to outputs through persistent grant IDs. Most talks have nothing.
The problem of ephemeral research talks is not new. A 1958 National Academies report noted that a large share of conference presentations never became full papers, leaving readers with little chance of finding a complete account. More than sixty years later, a Cochrane methodology review covering over 300,000 abstracts found that fewer than half of results initially presented at conferences were subsequently published in full, even after a decade (Scherer et al., 2018). A material share of what scholars present in talks never enters the formal publication record.
We’ve seen this pattern before
Talks are not uniquely difficult. They’re simply late. Datasets were long treated as supplementary material and then quietly lost. Over time, repositories, persistent identifiers, and data citation norms made them citable and indexable, which made it easier to include them in narratives of contribution. Preprints moved from informal circulation to recognised objects with stable records. Grants moved from administrative records to linkable nodes in the scholarly ecosystem.
Across these shifts the move is consistent: give the object a stable reference, attach shared metadata, and create linkages to the rest of the scholarly landscape.
What “legibility” looks like
A minimum talk record needs a title, a date, a venue or host, a format or type (invited talk, panel, evidence session, demonstration, performance), speakers and affiliations at the time, and a stable link to a recording, ideally video.
Once the minimum exists, you can add structured links: associated papers or projects, co-speakers and co-panellists, hosting organisations, grants, and persistent identifiers (ORCID for people, ROR for organisations, DOIs for the talks themselves), so the record connects cleanly to other systems. A few years ago this linking would have been prohibitively manual. AI can now extract speaker names, match affiliations, identify associated publications, and generate structured metadata from a raw recording and its context, making it feasible at scale.
This already works where it has been tried. Seminars are now indexed in Dimensions, a major research discovery platform, with full metadata and an embedded video player, making research talks searchable, citable, and linkable to associated publications in the same way that papers or datasets are. But these remain exceptions. Many institutions have invested heavily in lecture capture for teaching, yet the same infrastructure rarely extends to research seminars and conference presentations. There is no widely adopted, low-friction approach that makes talk-based outputs behave like first-class research contributions across disciplines and institutions.
The archive problem and the workflow problem
If an institution wants talk outputs to be part of its research record, it needs two capabilities.
Backfill (retrospective). A large body of talk-based activity can already be recovered from public traces: recordings on conference and society pages, institutional channels, and video platforms; programme listings in PDFs; announcements in archived web pages. The goal is a structured account: who spoke where, when, and about what.
Backfill changes the evidentiary starting point from ‘here is what we submitted to our last assessment’ to ‘here is a structured, checkable account of what actually happened.
But backfill will always undercount, because it can only find what was recorded and posted somewhere discoverable. If you stop there, you are rebuilding the same record from scratch next time.
Capture (prospective). The harder problem is ensuring that talks happening now and in the future are captured and documented as a matter of course.
This cannot mean asking every academic to upload recordings and input metadata by hand. With the right integrations, capture can be close to automatic: a talk is recorded, a record is created with comprehensive metadata, speakers can claim or correct details, and every item ends up with a stable, citable record. Keeping the record current then becomes a by-product of running events rather than a separate administrative project.
For this to work, governance has to be built into the capture workflow, not bolted on afterwards. Recording rights, consent, embargoes, third-party material in slides, and audience privacy are routine considerations. When speaker preferences and access permissions are managed at the point of capture, these stop being reasons not to record and become part of a consistent, repeatable process.
The power of video
The recording is what turns a record from a claim into evidence. Video is harder to fabricate than textual outputs, and it anchors the contribution in something concrete: what was said, by whom, and in what context. That has direct value for research integrity: misattribution and misrepresentation are harder when the primary source is inspectable.
But the strongest case for capturing talk-based scholarship may be simpler. A talk is where a researcher explains what they’re doing, why it matters, what they’re unsure about, and what they want next. For a prospective collaborator, it’s often more intelligible than a paper. For students, it’s often the best introduction to a lab’s agenda. For funders and industry partners, it supplies the context that formal writing tends to strip out. The Q&A and discussion that follow, when captured, add a layer of challenge and clarification that no other format provides.
Video also preserves things that written outputs routinely strip out: the way uncertainty is expressed, the way methods are explained to a live audience, the clarifications and challenges that emerge in Q&A and discussion. For many research questions, this context is as valuable as the findings themselves.
A record that points to stable video is not only more defensible during evaluation; it is immediately useful to people outside the institution who want to understand the work.
Closing
The research record has been expanding for two decades. We’ve made room for data, preprints, software, and grants. Talk-based scholarship is not a special case. It is one of the main ways ideas move, and the next generation of researchers already sees it that way. The infrastructure to make it legible now exists. What remains is for institutions and funders to decide to use it.
Ben Kaube is Co-founder of Cassyni.