Which papers generate the most research gaps?

When we mine research gaps from the literature, one question surfaces quickly: are gaps evenly distributed across papers, or does a small fraction of papers generate a disproportionate share? After processing 1.1M+ open research gaps from 780,000+ papers in our 7.9M-paper local library, the answer is unambiguous. The distribution is steep, and one paper type explains most of the concentration.

Why this matters

Understanding which papers yield the most documented gaps has practical value for researchers. If clinical practice guidelines are structurally richer in gap signals than most primary research articles, you can target your own literature review more precisely. It also matters for how we build and interpret the corpus: a library with strong coverage in medicine will produce a gap index that reflects that weighting. This analysis maps the concentration and its shape, across fields and across time, drawing on the same corpus that powers the Research Gap Finder.

Clinical practice guidelines account for the most gaps per paper

The paper with the highest gap density in our corpus is a 2026 Surviving Sepsis Campaign guideline, tagged by our extraction pipeline with 77 documented open questions. The next two highest are a 2024 Chinese hypertension management guideline (53 gaps; DOI: 10.1097/cm9.0000000000003431) and a 2026 European Society of Gastrointestinal Endoscopy update on peptic ulcer bleeding (40 gaps). All three are clinical practice guidelines.

The pattern holds across the top-ranked papers. The majority are guidelines, consensus statements, or structured systematic reviews. This is structurally predictable: guidelines are the one paper type explicitly required to grade the strength of their recommendations. Every "weak" or "conditional" recommendation signals that the evidence base is insufficient. Every section labelled "future research priorities" is a direct extraction target for our pipeline.

This has a practical implication for gap searches. Searches in clinical medicine and related fields will surface denser cross-sectional coverage than searches in, say, pure mathematics, where paper conventions rarely include formal research-priority sections.

Medicine dominates by field, but the distribution is broader than expected

Across the non-boilerplate gap corpus, the field breakdown is as follows:

Medicine's lead is substantial, but thirteen fields contribute meaningfully. For a gap search in biology, education, engineering, or social science, the underlying corpus has real density. Mathematics and humanities are the outliers: their paper conventions work against gap-signal extraction from text, so coverage there is thin relative to research output.

The field labels in our index come from each paper's source metadata, cross-checked during harvest. They are broad: "medicine" covers everything from cardiology to nursing science. If you search a specific clinical subfield, the narrower topic index returns more precise results.

Abstracts are the richest single source within a paper

About 600,000 gaps, roughly 42% of the non-boilerplate corpus, were extracted from abstracts. A further 330,000 came from dedicated research-gap sections, about 169,000 from future-research sections, and the remainder from conclusions, discussion sections, and limitations statements.

This has an implication that many authors do not expect: stating open questions in the abstract is the highest-leverage act for gap indexing. A sentence such as "It remains unclear whether this effect generalises to adolescent populations in low-income settings" is directly extractable; "more research is needed" is not. Our extraction pipeline looks for specificity, not sentiment.

Authors who bury their future-research statements in a final paragraph are not hiding them from extraction, but they are reducing the probability that those statements are tagged as gap signals. Inside the methodology agent explains the same specificity principle from the reviewer's side: vague gap claims in a submitted paper weaken the originality score just as vague gap statements in a source paper reduce extractability.

Older foundational papers carry more gaps per paper

We divided the prolific gap-source papers (those with at least five documented gaps) by publication era:

Two observations stand out. First, pre-2000 papers average 7.4 gaps each when they appear in the prolific group. These tend to be foundational methodological works whose open questions have accumulated over decades precisely because no subsequent paper resolved them. A gap mined from a 1985 paper remains in the open corpus until we index a paper that closes it, which is how our gap-open determination works.

Second, the 2000-2009 era is nearly absent: only three papers from that decade meet the five-gap threshold. This is a genuine coverage gap in our corpus rather than a property of the literature. Our harvest pipelines prioritise open-access papers from recent years and from pre-digital archives; the 2000s sit awkwardly between the two. We are actively extending coverage into that decade.

The gap synthesis pages partially compensate for library-coverage unevenness by grouping individual gap statements from multiple papers into topic-level syntheses. A topic with thin 2000-2009 coverage may still have a well-grounded synthesis drawn from adjacent decades.

What this means for authors

If your paper is a clinical practice guideline, a systematic review, or a large cohort study with a structured limitations section, your open questions have a direct pathway into the indexed gap corpus. The extraction pipeline processes these paper types well.

For most primary research papers, the abstract is the highest-leverage location for stating open questions. Specificity matters: name the population, the intervention, the context, or the methodological question you could not resolve. A sentence that frames the gap precisely is far more likely to surface in a Research Gap Finder search than one that gestures at future work broadly.

You can search any topic free at the Research Gap Finder to see whether your field's questions are already indexed, and one unlock writes your gaps section with source citations. The AI Review scorer also surfaces gap-related feedback: the originality agent checks whether your claimed contribution addresses a genuine gap in the literature, grounded in the same corpus described here.

FAQ

What is a "research gap" in your corpus?

A research gap is a statement extracted from a published paper that identifies a question the paper's own evidence does not resolve. We extract these from abstracts, conclusions, future-research sections, and dedicated gap sections. Each gap is linked back to its source paper by DOI. For the full operational definition, see what counts as a research gap.

Why do clinical guidelines generate so many documented gaps?

Guidelines are consensus documents that grade the strength of their recommendations. Every recommendation graded "weak" or "conditional" signals that the evidence base is insufficient, which is the definition of an open research question. This makes guidelines unusually dense in direct gap statements, compared with a primary research article where future directions are often mentioned briefly in the discussion.

Are gaps from older papers still treated as open?

Yes, unless a later paper in our library explicitly reports resolving the question. A gap mined from a 1985 paper remains in the open corpus until we index a paper that closes it. This is why pre-2000 papers can still be prolific gap sources: the questions they raise may simply not have been answered in the portion of the literature we have indexed.

Does the field distribution affect which gaps I can find?

Yes. Medicine, biology, and education have the densest coverage, so searches in those fields will return more indexed gaps. Mathematics and humanities have the thinnest coverage. If your field is underrepresented, we recommend combining the gap search with a broader topic keyword to draw on adjacent fields.

How many citable gap pages are there?

More than 1,900 grounded gap syntheses have permanent pages under /research-gaps, each with source citations and field context. The broader index of 1.1M+ open gaps is searchable by topic, though not every gap has its own dedicated page.

AI Review reports are written by AI agents. They are pre-submission feedback, not an editorial decision.