Literature Gap Analysis in the Browser: Turning Search Results Into a Research Question

A gap in the literature is not an absence of papers on a topic. It is a specific combination of population, method, context, or outcome that existing work does not address. Browser tools can accelerate the mapping stage, but the judgment about what counts as a gap remains human. The useful workflow turns a search export into a coverage matrix and then tests candidate gaps against existing reviews.

Build a corpus before looking for gaps

Run the same structured search in at least two databases. Record the query string, filters, and date. Export titles, abstracts, years, journals, and author keywords.

Deduplicate on DOI, then on normalized title. A corpus of 300–800 records is workable for a focused field; larger sets need sampling or clustering.

Do not include records you cannot screen. A gap claim based on a corpus you never read is a guess with extra steps.

Build a coverage matrix

Choose three dimensions that define the field. Common choices are population, method, and outcome; for technical fields, dataset, model, and evaluation metric. Extract values for each record, either from author keywords or by coding a sample.

A spreadsheet pivot or a small script produces a count per cell. Sparse cells are candidates. A cell with zero records across five years is a stronger candidate than one with two old records.

Automate extraction with care: keyword matching for method terms works reasonably; inferring population from abstracts requires a model and a human check. Report which fields were machine-extracted if the matrix appears in a paper.

Test candidate gaps against reviews

Before claiming novelty, search for reviews and meta-analyses covering the same field. A gap that a 2024 review already names as an open question is not a new finding; it is a documented priority, which is even better for a proposal.

Read the limitations sections of the five most recent relevant papers. Authors routinely state what their design could not test. Those sentences are the most reliable gap statements available.

If a gap survives both checks, write it as a falsifiable question: which population, compared with what, on which outcome, over what period.

Where AI helps, and where it misleads

Models are good at clustering titles into themes, normalizing inconsistent terminology, and drafting a summary of what a cluster covers. They are bad at knowing what has been published recently, at judging study quality, and at detecting that a gap is filled by a paper using different vocabulary.

Never accept "no studies exist on X" from a model. Search for X, then for synonyms, then for the mechanism. Absence of evidence in your corpus is not absence in the literature.

The deliverable, and coding the corpus consistently

Produce three artifacts: the search log, the coverage matrix with counts, and a short memo naming two or three candidate gaps with the evidence for each. That package supports a proposal, a thesis introduction, or a conversation with an advisor, and it is auditable when someone asks how you concluded the gap was real.

A coverage matrix is only as good as the coding behind it. Write a codebook with one definition per dimension value, then have two people code a 30-record sample and compare. Disagreements point to ambiguous definitions, not careless coders.

For machine-assisted extraction, report the model and the prompt version, and validate against the human-coded sample. A model that codes "qualitative" for mixed-methods studies will skew the matrix in a direction you may not notice.

A moving field, and turning a sparse cell into a question

A gap identified in a corpus retrieved in January may be filled by June. Record the search date on every gap claim and re-run the search before submitting a proposal. Reviewers in fast-moving fields check recency.

Set alerts for the exact gap query, not just the broad topic. The specific combination of population and method is what determines whether the gap still exists.

A sparse cell is a description, not a research question. Convert it into a testable formulation: which population, compared with which alternative, on which outcome, over what period.

Write the question and then ask what result would falsify it. If nothing could, it is a topic rather than a question, and the gap claim is not useful.

Justifying feasibility, and presenting the matrix

A gap that no one can study is not an opportunity. Check data availability, sample access, and methods maturity. A gap in a population that is impossible to recruit is a limitation of the literature, not a study you can run.

Include a short feasibility paragraph next to each candidate gap. It is the first thing an advisor or funder will ask about.

Show counts per cell, the search date, and the coding rules. A reader should be able to see why a cell is considered empty without trusting your summary.

Where a cell contains studies you judged out of scope, list them with the reason. That is the difference between a gap analysis and an assertion.

The gap memo, rechecking, and writing a gap that is falsifiable

One page per candidate gap: the cell, the count, the nearest studies, why they do not fill it, and the question that would.

Circulate the memo before writing a proposal. Feedback at this stage is cheap.

Rerun the search and note the date. A gap statement without a recent search is easy to challenge.

"Little work exists on X" is unfalsifiable. "Of the 34 studies on X, none report Y for Z" can be checked by a reviewer with the same search.

State the search date, the databases, and the inclusion rule. Three sentences, and the claim stops being an opinion.

A gap is not a contribution. Pair the empty cell with the method that would fill it, and with the data you can realistically obtain in the time available.

Some gaps are empty because the work is expensive rather than because nobody thought of it. Say which kind you are addressing, and a reviewer will trust the rest of the argument more.