Is Web of Science Citation Data Reliable?

Web of Science citation data is widely used in research assessment because it is curated, deeply historical, and structured as a genuine citation network — but a raw citation count from Web of Science is not a neutral, complete measure of quality or influence. It reflects a specific, editorially-selected slice of the literature (the Web of Science Core Collection), counted using rules that treat every citing instance the same regardless of context, discipline, or intent. Whether that data is ‘reliable’ depends entirely on what question it’s being asked to answer: it is a reasonably consistent measure of citation activity within its own indexed universe, and a poor, systematically biased proxy for research quality or importance when used alone, across fields, or without disclosure of what it misses.

What Web of Science citation data actually measures

A Web of Science citation count is the number of times a record has been cited by another item that is itself indexed in the Web of Science Core Collection — not a count of citations from the scholarly literature as a whole. The Core Collection is built from six component indexes (Science Citation Index Expanded, Social Sciences Citation Index, Arts & Humanities Citation Index, Emerging Sources Citation Index, Conference Proceedings Citation Index, and Book Citation Index), each covering a curated list of journals, conference series, or book series that Clarivate’s editorial team has reviewed and admitted against a documented set of quality and impact criteria — publishing standards, peer-review practice, editorial content, and citation activity among them — and continues to monitor after admission. That editorial boundary is the source of both Web of Science’s main strength (a comparatively clean, de-duplicated citation graph within its scope) and its main limitation (anything outside that boundary is invisible to it, no matter how influential).

The genuine strengths

Curated, editorially reviewed indexing. Unlike an aggregator that ingests everything it can find, Web of Science’s selective admission process is intended to filter out low-quality or predatory venues before they ever enter the citation graph, which reduces (without eliminating) the risk of citation counts being inflated by low-quality citing sources.
Long historical depth. The Science Citation Index, the oldest component of the Core Collection, traces to 1964 and Eugene Garfield’s original citation-indexing model — giving Web of Science genuinely long, consistent backward citation coverage for fields it has always indexed well, which matters for longitudinal bibliometric work and for senior researchers’ full career records.
A structured citation network, not just a citation count. Because every indexed record’s reference list is itself parsed and linked, Web of Science supports cited-reference search and citation-network analysis, not only a raw tally — the same structure that CASRAI’s guide to Web of Science Citation Reports covers for generating set-level metrics.
An established position in institutional and funder workflows. Clarivate’s InCites platform, built on Web of Science data, and the annual Journal Citation Reports are already embedded in many universities’ promotion, tenure, and benchmarking processes, which gives Web of Science-sourced figures a familiarity and a degree of cross-institutional comparability that a newer or less widely licensed source may not yet have.

Documented coverage gaps

The limitations below are not fringe objections — they are well established in the peer-reviewed bibliometrics literature comparing citation database coverage, most notably Mongeon & Paul-Hus’s 2016 Scientometrics study ‘The journal coverage of Web of Science and Scopus: a comparative analysis,’ which found substantial, discipline-dependent differences in how much of the published literature each database actually captures.

By discipline

Web of Science’s coverage is strongest in the natural sciences, engineering, and biomedicine — fields where journal articles are the dominant, near-universal output type and where the Science Citation Index Expanded has the deepest, longest-running coverage. Coverage is comparatively weaker in the social sciences, and weaker still in the humanities, where a meaningful share of scholarly output never appears in a Web of Science-indexed journal at all. This is not a defect unique to Web of Science — Scopus shows the same discipline gradient, just with a somewhat wider net in places — but it means a raw citation count means something structurally different depending on the discipline of the person or output being assessed.

By document type

Citation indexes were built around the journal article, and it shows. Books, book chapters, technical reports, working papers, policy documents, and other non-journal outputs are underrepresented relative to how central they are to scholarly communication in fields like law, history, and much of the humanities and social sciences. Web of Science added a dedicated Book Citation Index only in 2011 and the Emerging Sources Citation Index in 2015, specifically to widen coverage of regional, multidisciplinary, and emerging-field journals that its older core indexes had been missing — a direct, if partial, institutional acknowledgment of the gap. Conference proceedings, which carry outsized importance in computer science and parts of engineering, are covered through a separate index (Conference Proceedings Citation Index) with its own, generally narrower, selection criteria.

By language

Web of Science’s core indexes have historically skewed toward English-language journals, a pattern widely discussed in the bibliometrics literature as part of a broader ‘Anglophone bias’ in citation-database coverage. Non-English-language scholarship — common in regional social-science and humanities publishing, and in scholarship from many non-Anglophone countries generally — is therefore underrepresented in both the indexed literature and, by extension, in the pool of items available to cite (and be cited by) other indexed work.

Citation-counting methodology quirks

Self-citations are counted by default. A standard Web of Science citation count, h-index, or Citation Report figure does not exclude an author’s or a journal’s self-citations unless specific records are manually excluded before the count is generated — a single-author paper that cites the author’s own prior work still adds to that author’s cumulative citation total.
Citation window / time-lag effects. Citations take time to accumulate: a paper published eighteen months ago has had far less opportunity to be cited than one published a decade ago, regardless of underlying quality. Comparing raw citation counts across researchers or papers at different career stages, or across recently published versus long-established work, without accounting for this lag produces a systematically misleading comparison.
Every citation counts the same, regardless of context. A citation that credits a paper as foundational to a new finding is counted identically to one that cites it only to critique, correct, or disagree with it, or one that cites it in a single perfunctory background sentence. Citation-context research in the bibliometrics literature has long noted this as a structural limit of any raw citation count, Web of Science’s included — the count reflects that a citation occurred, not why.
Cross-field comparison requires normalization. Citation practices vary enormously by field — some disciplines cite far more heavily and rapidly than others as a matter of publishing convention, independent of quality. A raw, un-normalized citation count compared directly across fields is not a like-for-like comparison. This is precisely why Clarivate’s own InCites platform computes a field-normalized indicator (Category Normalized Citation Impact) rather than presenting raw counts alone for institutional benchmarking — an implicit acknowledgment, from the data provider itself, that raw counts need correction before they support a fair cross-field comparison.

What DORA, CoARA, and the Leiden Manifesto say about this

None of the above is a niche critique — it is the explicit basis for the major responsible-research-assessment frameworks the field has converged on. The San Francisco Declaration on Research Assessment (DORA) discourages using journal-level and citation-based metrics as a proxy for the quality of an individual researcher’s contributions in hiring, promotion, or funding decisions. The Leiden Manifesto‘s ten principles include explicit guidance directly on point: quantitative indicators should support, not replace, qualitative expert judgment, and citation-based assessment must account for variation in publication and citation practice across fields. The Coalition for Advancing Research Assessment (CoARA), through its 2022 Agreement on Reforming Research Assessment, commits signatory institutions to moving away from inappropriate reliance on journal- and publication-based metrics, including citation counts pulled from any single database, Web of Science included. None of these frameworks say citation data from Web of Science is worthless — they say it should never be the sole or decisive input into a judgment about a person’s or a body of work’s quality.

Practical guidance for research administrators

Never present a single Web of Science citation figure as a stand-alone quality judgment. Pair it with narrative, qualitative context — the practice both DORA and the Leiden Manifesto recommend — especially in tenure, promotion, and hiring decisions.
Cross-check against at least one other source — typically Scopus or OpenAlex — particularly for researchers in social sciences, humanities, or fields with heavy conference-proceedings or book output, where Web of Science’s coverage gap is largest. See CASRAI’s Scopus vs. Web of Science vs. OpenAlex comparison for how the three differ on coverage and access.
Name the source, the exact search or record set, and the pull date whenever a Web of Science-sourced citation figure appears in a dossier or report — an unlabeled citation count or h-index is not independently checkable, and figures from different databases are not interchangeable.
Use a field-normalized indicator, not a raw count, for any cross-field or cross-institution comparison — Category Normalized Citation Impact via InCites, or an equivalent normalized metric from whichever source is in use.
Disclose known coverage gaps relevant to the case at hand — a humanities scholar’s Web of Science citation count understates their scholarly reach in a structurally different way than it would for a molecular biologist, and a review process that treats the two figures as comparable is applying the metric outside where it’s reliable.
Don’t exclude self-citations selectively or inconsistently. If self-citations are excluded for one candidate’s figures, apply the same exclusion rule to every candidate being compared, and document that the adjustment was made.

Frequently asked questions

Is Web of Science citation data accurate?

It accurately counts citing instances from within its own indexed universe — the Web of Science Core Collection. It is not a complete or unbiased count of all scholarly citation activity, because it only reflects citations from sources Clarivate has admitted to that index.

Why do citation counts differ between Web of Science, Scopus, and Google Scholar?

Each source indexes a different, only partially overlapping, set of publications and citing literature, so the same underlying body of work produces a different raw citation count on each platform. This is a coverage difference, not a computational inconsistency in how a citation is defined.

Does Web of Science undercount citations in the social sciences and humanities?

Its coverage is comparatively weaker there than in the natural and life sciences, largely because those fields rely more heavily on books, regional journals, and non-English-language publishing that historically fell outside Web of Science’s core indexes, notwithstanding later additions like the Book Citation Index and Emerging Sources Citation Index.

Should a promotion or tenure committee rely on a Web of Science h-index alone?

No responsible-assessment framework recommends this. DORA, the Leiden Manifesto, and CoARA all explicitly caution against using any single citation-based figure, from any one database, as a stand-alone measure of an individual’s quality or impact.

Does Web of Science count self-citations in its citation totals?

Yes, by default. Self-citations are included in a standard citation count, h-index, or Citation Report figure unless specific records are manually excluded before the figure is generated.

Related CASRAI resources

Web of Science — dictionary definition.
Web of Science Citation Report: What It Measures and How to Read It.
How to Search Web of Science Effectively.
Scopus vs Web of Science vs OpenAlex comparison.
H-Index Inflation.
DORA, CoARA, and the Leiden Manifesto — the major responsible-research-assessment frameworks.
Persistent Identifiers & Research Information Systems — the CASRAI cluster hub for this topic.

Sources

Mongeon, P. & Paul-Hus, A. (2016). ‘The journal coverage of Web of Science and Scopus: a comparative analysis.’ Scientometrics, 106(1). Peer-reviewed comparative-coverage study.
Clarivate, Web of Science Core Collection product and editorial-selection documentation (clarivate.com/academia-government/scientific-and-academic-research/) — Core Collection component indexes, journal evaluation process, and InCites/Category Normalized Citation Impact.
DORA, San Francisco Declaration on Research Assessment (sfdora.org).
Hicks, D., Wouters, P. et al. (2015). ‘The Leiden Manifesto for research metrics.’ Nature, 520.
Coalition for Advancing Research Assessment (CoARA), Agreement on Reforming Research Assessment, 2022 (coara.eu).