How to Read Citation Metrics: Impact Factor, h-Index, and CiteScore

Quick Answer: How to Read Citation Metrics Without Treating Them as Grades

Use a journal metric to compare journals in the same field, an article citation count to see whether a paper has been taken up yet, and an h-index only with career length and field in view. Never treat a journal's Impact Factor as the quality of one article inside it. Clarivate's 2026 JCR release repeats the same limit in one sentence:

There is no universal "good" Impact Factor or "good" h-index. Leiden's authors wrote that top mathematics journals sat around 3 while top cell-biology journals sat around 30. Compare inside a subject category, or use a percentile / quartile, not a number you memorized from another field.

What Citation Metrics Measure

Citation metrics convert later bibliographic references into a number. They answer "how often was this journal, paper, or author cited in this database, in this window?" They do not answer "is the method sound?" or "should I cite this?" For those questions, read the paper and run a source evaluation.

Students meet these numbers when choosing a journal, scanning a literature review, or watching a supervisor talk about "high-impact" venues. The same number is a weak signal in hiring or funding. DORA was written because committees were using journal averages as a stand-in for a person's file. JCR 2026 still exists to compare journals, not people.

Journal Impact Factor in the 2026 JCR

Eugene Garfield invented the journal impact factor at the Institute for Scientific Information in the 1960s and began publishing it in Journal Citation Reports in 1975. Clarivate still computes it. The Clarivate impact-factor essay (Garfield's 1994 Current Contents text, kept on Clarivate's site) defines the annual JCR impact factor as a ratio of citations to recent citable items:

The 2025 Journal Citation Reports Reference Guide restates the same ratio and names the year. Do not mix "release year" with "JCR year" when you copy a number onto a CV:

Clarivate's 17 June 2026 news release says the 2026 edition "reflects 2025 data" and reports metrics for 22,643 journals across 254 categories, including 521 titles that received a Journal Impact Factor for the first time. Apply the Reference Guide rule to that release: the current JIF is the 2025 Journal Impact Factor — citations in 2025 to items dated 2023 and 2024, divided by articles and reviews from those two years. A "2026 JCR" file is not a 2026 Impact Factor.

Numerator and denominator are not the same set

The most common student mistake is treating JIF as "citations to articles, divided by articles." The 2025 Reference Guide is explicit that the numerator counts every citation to the journal in the JCR year, while the denominator counts only articles and reviews:

A journal that publishes many cited editorials can lift its numerator without growing its denominator. Review journals often sit at the top of a category because reviews are cited as stand-ins for a literature. Garfield already warned that review articles are cited more frequently than typical research articles.

Retracted papers still sit in the denominator

From the 2025 JCR release (2024 data) onward — and still in JCR 2026 — Clarivate removes citation chains that involve retracted or withdrawn items from the JIF numerator. The retracted article itself stays in the article count. Nandita Quaderi's 15 May 2025 Clarivate note is the official wording:

That policy cleans the journal average. It does not clean an article-level Times Cited in Web of Science or Scholar. Retracted work can keep getting cited; check the retraction notice before you lean on a count. A journal that published several papers later retracted may show a lower newest JIF than last year because the numerator was cleaned, not because the remaining articles suddenly got worse.

Failure modes that make an Impact Factor misleading

• It is a journal average, not an article score. Citation distributions inside a journal are skewed. DORA lists that skew as limitation A. Most papers sit below the mean.
• Fields are not interchangeable. The Leiden Manifesto gives the official contrast: mathematics versus cell biology.
• The window is two years. Slow-citing fields (mathematics, parts of the humanities) look weaker than fast-citing biomedicine. Clarivate also publishes a 5-year Journal Impact Factor for that reason.
• Editorial policy can move the number. DORA lists gaming as limitation C. Coercive citation, extra review articles, and journal self-citation all change the ratio.
• Coverage is not the whole literature. Only journals in the Web of Science Core Collection receive a JIF. Clarivate extended JIF to Arts & Humanities and Emerging Sources titles in 2023, but many excellent national-language and niche titles still never appear.
• A missing 2026 metric is not always a missing journal. JCR 2026 now shows some titles with a Forthcoming label when the current-year denominator is incomplete, and links them to last year's figures. Do not invent an Impact Factor for those titles.
• One decimal is enough. Leiden principle 8 says publishing JIF to three decimal places is false precision.

If you need a journal-level number at all, prefer the JCR percentile or quartile in a subject category — or the field-normalized Journal Citation Indicator — over the raw JIF. An Impact Factor of 3.0 can be the top of one category and mid-pack in another. Do not use "IF above 10" as a quality rule; that cutoff is not in Clarivate's definition and it silently excludes entire disciplines.

Journal Citation Indicator (field-normalized JIF companion)

The Journal Citation Indicator (JCI) is Clarivate's answer to the field problem Leiden named. It is the mean Category Normalized Citation Impact for all articles and reviews the journal published in the preceding three years. Martin Szomszor's official introduction is the sentence to keep next to any raw JIF:

A JCI of 2.0 means twice the category average; 0.5 means half. The 2025 Reference Guide adds that the 2024 JCI, for example, used citable items from 2021–2023 and citations those items received from any document indexed between 2021 and 2024. Unlike JIF, JCI is field-normalized, uses a three-year publication window, and counts citations from any later year up to the end of the current year rather than from the JCR year alone. Clarivate still says not to compare physical-science journals with arts-and-humanities journals on JCI as if they were the same market. For arts and humanities categories, JCI rankings are often more informative than a raw JIF ranking.

The h-Index (Hirsch, 2005)

Jorge E. Hirsch proposed the h-index in Proceedings of the National Academy of Sciences on 15 November 2005 (doi:10.1073/pnas.0507655102; PMC PMC1283832). He designed it as one number that mixes productivity and citation impact for an individual researcher, using Web of Science counts:

Google Scholar Metrics restates the same rule for a publication and gives a worked set you can check by hand. Those five counts are Scholar's example, not a real journal:

Sort an author's papers by times cited. Walk down the list until rank no longer stays at or below the citation count. That rank is h. Papers cited more than h times do not raise h until enough other papers also clear the new threshold. That is why one famous paper cannot manufacture a high h-index, and why a long list of uncited papers cannot either.

Hirsch's own 2005 physicist numbers — not a universal ladder

The old "graduate student 2–5, mid-career 15–25" bands you still see on blogs are not in Hirsch's paper. What Hirsch did publish, for physicists and with "large error bars," is this:

Hirsch immediately added that life-science h values run much higher than physics, and that large collaborations inflate h. The Leiden Manifesto later put field and database dependence in one paragraph you should keep next to any author metric:

Limitations of the h-index

• It never falls. Outdated work stays in the count. Hirsch noted this explicitly.
• It tracks age. Two extra decades of publishing raise h even if recent work is thin. Hirsch's m (h divided by scientific age) was his attempt to compare different seniorities.
• Coauthors are invisible. Hirsch suggested normalizing h when collaboration sizes differ. High-energy experiment and some biomedical consortia produce large h values that do not mean one person wrote the papers.
• The database changes the number. Scholar includes theses, preprints, and more books; Web of Science and Scopus are curated. Always name the source: "h = 18 (Scopus, 2026-03)" is a fact; "h = 18" is not.
• It ignores the tail above h. A paper with 4,000 citations and a paper with h + 1 citations add the same amount once they are inside the core.
• It is a poor hiring rule. Leiden principle 7: read the portfolio. DORA recommendation 1 forbids using journal metrics as a surrogate for a person; the same logic applies to treating h as the file.

Article Citation Counts

"Times cited" is the simplest metric and the one students can actually use while building a literature review. It is still not a quality score. Review articles, methods papers, and papers in large fields accumulate citations faster. A 2025 article with 3 citations is not "worse" than a 2005 methods paper with 800. Retracted work can keep getting cited even after Clarivate removed those chains from the journal JIF; check the notice before you lean on a count.

Do not memorize bands such as "100 citations = influential." Those cutoffs are not in Clarivate, Scopus, or Scholar documentation. Instead ask: How old is the item? How large is the field? Is it a review? Who is citing it — a methods section, a critique, or a passing mention? Cited by in Scholar, or Times Cited in Web of Science, lets you read the citing papers.

Those four items are identifiable publications with DOIs. Open them in Scholar or Web of Science and you will see different totals. That difference is the lesson: the count is a property of the index, not a physical constant of the paper.

CiteScore, Eigenfactor, and Other Journal Indicators

CiteScore (Elsevier / Scopus)

CiteScore is Elsevier's Scopus journal metric. Unlike JIF, the citation window and the publication window are the same four calendar years, and the document types in the numerator match the denominator. Elsevier's Scopus Support page (last updated 3 June 2025) gives the official method and still uses CiteScore 2024 as the worked year:

Elsevier's CiteScore product page now offers a CiteScore 2025 fact sheet. Apply the same official rule to the next frozen year — do not invent a different window:

That four-year window is why CiteScore and JIF for the same title rarely match. CiteScore also includes conference papers, book chapters, and data papers that JIF's "article and review" denominator omits. Elsevier freezes the annual value around May–June and shows a monthly CiteScore Tracker until then. Use CiteScore percentiles in a Scopus subject area the same way you use JCR quartiles: to stay inside a field. Elsevier's own "golden rules" on that page match DORA and Leiden: use qualitative judgement, and never one metric alone.

Eigenfactor and Article Influence

Carl Bergstrom and Jevin West's Eigenfactor metrics treat citations as a network, in the same family as eigenvector centrality / PageRank. A citation from a highly cited journal counts more than a citation from a little-cited one, and journal self-citations are excluded. Bergstrom, West, and Wiseman described the pair in The Journal of Neuroscience 28, no. 45 (2008): doi:10.1523/JNEUROSCI.0003-08.2008. The Eigenfactor Score scales with journal size (more articles, more total influence). The Article Influence Score divides that influence by the number of articles so you can compare it with an impact factor. Scores appear in Journal Citation Reports; the project site is eigenfactor.org.

How the main journal metrics differ

Google Scholar Metrics and Author Profiles

Scholar publishes journal-level Scholar Metrics and person-level profiles. As of this rewrite, Metrics help is still dated to the July 2025 index and covers articles from 2020 through 2024. Scholar excludes court opinions, patents, books, and dissertations from those journal lists, and it drops titles with fewer than 100 articles in the five-year window. That is why a small but important journal can be missing from the top-100 lists even when Scholar search still finds its papers. The Google Scholar guide is the search workflow; this page is how to read the numbers that workflow surfaces.

On a personal profile, Scholar says it computes two versions — All and Recent — of three numbers: h-index, i10-index, and total citations. Recent is a shorter window on the same profile. Scholar counts more document types than Web of Science, so the same person usually looks "higher" on Scholar. That is coverage, not a promotion.

Altmetrics

Jason Priem, Dario Taraborelli, Paul Groth, and Cameron Neylon published altmetrics: a manifesto on 26 October 2010 (version 1.01, 28 September 2011; Zenodo record). They argued that citation filters were too slow for the online literature and that downloads, saves, posts, and policy mentions could be additional filters:

Commercial dashboards (Altmetric.com, PlumX) turn those events into a badge: news, policy documents, Wikipedia, blogs, patent citations, and social posts. A high score can mean a paper reached clinicians or reporters this week, which a two-year JIF cannot show. It can also mean a controversy or a press office. Altmetrics measure attention. They do not replace peer review or a reading of the methods.

Where to Look Up the Numbers

If your campus does not subscribe to JCR, do not scrape a random "impact factor" from a third-party SEO page. Those sites mix outdated JIFs, Scopus SNIP/SJR, and invented scores. Prefer the publisher's "Journal Impact Factor (Clarivate, JCR year)" line, or ask a librarian. For open full text after you have the citation, use the legal access paths.

Using Citation Metrics Without Fooling Yourself

DORA and the Leiden Manifesto

On 16 December 2012, journal editors meeting with the American Society for Cell Biology in San Francisco drafted what became the San Francisco Declaration on Research Assessment, posted in May 2013. The declaration's opening diagnosis is the sentence students need when a syllabus says "cite high-impact journals only":

DORA then lists four deficiencies: skewed citation distributions; field-specific composite scores; editorial gaming; and data that are not openly available. The general recommendation is the rule funding agencies and universities that signed DORA are supposed to follow:

The Leiden Manifesto (Hicks, Wouters, Waltman, de Rijcke, and Rafols, Nature 520, 429–431, 23 April 2015 issue; published 22 April) is the ten-principle counterpart for people who still use metrics. The official titles, in order:

1. Quantitative evaluation should support qualitative, expert assessment.
2. Measure performance against the research missions of the institution, group or researcher.
3. Protect excellence in locally relevant research.
4. Keep data collection and analytical processes open, transparent and simple.
5. Allow those evaluated to verify data and analysis.
6. Account for variation by field in publication and citation practices.
7. Base assessment of individual researchers on a qualitative judgement of their portfolio.
8. Avoid misplaced concreteness and false precision.
9. Recognize the systemic effects of assessment and indicators.
10. Scrutinize indicators regularly and update them.

Leiden principle 9 is the warning about gaming: a single indicator becomes the goal. Their example is 1990s Australia, which funded universities largely on paper counts and then watched output rise in less-cited journals. If your department still pays a bonus for "IF above 15," that is the pattern Leiden was written to stop.

How Metrics Get Distorted

Practices that move the number without moving the science

• Coercive citation. An editor or reviewer requires citations to the journal or to a preferred lab. That grows JIF and some h-indexes; it is research-integrity misconduct, not impact.
• Citation cartels. Groups cite one another to cross a threshold. Clarivate has suppressed JIF for anomalous citation patterns since the 2008 release.
• Salami slicing. One study becomes several least-publishable papers so that each can be cited. Counts rise; the evidence does not.
• Review-heavy issues and extra editorials. Because JIF's numerator counts citations to non-citable items, a journal can harvest citations that never enter the denominator.
• Predatory "impact factors." Fake metric brands mimic JCR. If the producer is not Clarivate, Scopus, or another named index, treat the badge as advertising.

Distortions that are not fraud

The Matthew effect — already-visible work attracting still more citations — is a property of attention, not a scam. English-language, U.S.-indexed journals are easier to cite in Web of Science; Leiden principle 3 exists because local labour-law or HIV-epidemiology work can be excellent and still look "low impact" in that index. Negative citations ("this experiment failed to replicate") still increment Times Cited. Citation counts do not carry a plus or minus sign. JCR 2026's own data underline the coverage problem: 58% of the 521 journals that received a first JIF that year sit outside the United States and Western Europe. Absence from last year's list was often an indexing timeline, not a quality verdict.

Citation Metrics for Students

You rarely need an Impact Factor to write a strong undergraduate or master's paper. You do need a way to find the papers a field still argues with.

• Use counts to navigate, not to judge. Sort Scholar by relevance first, then open Cited by on a promising hit. Highly cited older papers are often the methods or theory everyone assumes.
• Pair a classic with something recent. A 1998 paper with thousands of citations plus a 2024 test of the same claim is a literature, not a trophy case.
• Ignore journal IF when the article is what you are citing. DORA's point applies to bibliographies too: cite the work you used.
• Do not skip a 2026 paper because it has two citations. The two-year JIF window has not even closed.
• Niche and humanities work will look "small." Books, national-language journals, and archival essays are under-counted in JCR. That is a database limit, not a reason to drop the source.

Related Guides

• How to Use Google Scholar for Academic Research
• How to Evaluate Sources for Academic Research
• How to Tell If a Source Is Credible
• Scholarly vs Popular Sources
• Peer Review Process: What Students Need to Know
• How to Find Scholarly Sources