Written and maintained by CASRAI Editorial Board
Last updated
There is no fixed number of citations that counts as “good.” A citation count only becomes meaningful once you fix four variables: field (citation density varies by an order of magnitude between disciplines), time since publication (citations accumulate for years, so a two-year-old paper and a fifteen-year-old paper are not comparable on raw counts), document type (reviews are cited far more than original research articles; editorials and letters far less), and database (Google Scholar, Scopus and Web of Science index different content and routinely produce three different counts for the same paper). Any answer that skips straight to a bare number — “100 citations is good,” “10 citations a year is average” — is answering a question that hasn’t actually been fixed yet.
This is why research-assessment specialists don’t judge papers on raw citation counts at all. They use field-normalized citation metrics, where the benchmark isn’t a specific number but a ratio against the world average for comparable work. That reframing is the actual answer to “how many citations is good”: good means above the normalized baseline for your field, document type and publication year, not above some fixed threshold.
The normalized-metric answer: what “average” actually means
The major citation databases each publish a field-normalized indicator built on the same underlying logic: take a paper’s raw citation count, divide it by the average citation count of a comparable set of papers (same field, same year, same document type), and the result centers on 1.0 as “exactly average.” A score above 1.0 means above-average performance for that specific comparison set; a score below 1.0 means below average. This is the closest thing research metrics has to a universal, field-independent answer.
| Metric | Provider / data source | What 1.0 means | Rough reading |
|---|---|---|---|
| Field-Weighted Citation Impact (FWCI) | Elsevier SciVal, built on Scopus coverage | Exactly the world average for comparable articles | FWCI 1.5 = 50% more citations than the global average for that field, year and document type |
| Category Normalized Citation Impact (CNCI) | Clarivate InCites, built on Web of Science Core Collection | Exactly the Web of Science category average | The Web of Science analogue of FWCI — same interpretation, different underlying database |
| Mean Normalized Citation Score (MNCS) | CWTS Leiden Ranking | Exactly the field/year/document-type average | MNCS of 2 = twice the field average; used at the institutional level, not typically quoted for single papers |
| Relative Citation Ratio (RCR) | NIH iCite, benchmarked against NIH-funded output | The median citation performance of NIH-funded papers in the same field and year | RCR of 1.0 = typical for an NIH-funded paper in that area; built from a co-citation network rather than a fixed subject category |
Read across all four rows: a paper scoring meaningfully above 1.0 — roughly 1.5 or higher, depending on the metric and how much variance is normal in that field — is doing better than the typical comparable paper. A score at or near 1.0 is unremarkable, not bad. A score well under 1.0 is under-cited relative to its peers. None of these thresholds translate to a fixed raw number, because the underlying “average” they’re measured against is different for every field, year and document type. See CASRAI’s Field-Weighted Citation Impact (FWCI) and Relative Citation Ratio (RCR) entries for how each is actually calculated.
Why raw citation counts mislead on their own
- Field-dependence. Fast-moving, large, heavily-referencing fields (much of biomedicine, for instance) generate far more citations per paper than smaller or slower-citing fields (parts of mathematics, some humanities disciplines). A count that looks unimpressive in one field can be near the top of its distribution in another. This is the same field-dependence problem documented on CASRAI’s good h-index entry, and it applies just as directly to raw citation totals.
- Time-dependence. Citations accumulate over years, sometimes over a decade or more for a paper’s full citation life. A paper published two years ago simply hasn’t had time to reach the count a fifteen-year-old paper in the same field has — comparing the two on raw counts conflates “recent” with “uncited,” which are not the same thing.
- Document-type-dependence. Review articles are structurally cited far more often than original research articles, because they’re written to be cited as a single reference point for a body of work. Editorials, letters and commentaries are typically cited far less. Comparing a review’s count against a primary research article’s count as if they’re the same kind of thing produces a meaningless comparison.
- Database-dependence. Google Scholar, Scopus and Web of Science apply different indexing scope and coverage rules, and can report meaningfully different citation counts for the identical paper. A count quoted without naming the source database is not fully specified.
Percentile indicators: a more robust alternative to a single ratio
Some research-evaluation frameworks go a step further than a normalized ratio and report where a paper sits in the citation distribution for its field, rather than just its distance from the mean. The CWTS Leiden Ranking’s PP(top 10%) indicator, for example, reports the proportion of an institution’s output that falls in the most-cited 10% of papers for its field, year and document type — and CWTS’s own methodology guidance treats this as more robust than a mean-based score like MNCS, because a single extremely highly-cited outlier paper can distort an average without meaningfully distorting a percentile rank. The same top-percentile logic underlies “highly cited” designations used elsewhere in bibliometrics (for example, Web of Science’s Highly Cited Researchers list, built from papers ranking in the top 1% by citations for their field and year). If you’re trying to judge a single paper or a body of work and a normalized-metric tool is available, a percentile rank is generally a more reliable signal than either a raw count or a single ratio.
A short decision flow for judging a citation count
- Do you know the field and the comparison set? If not, stop — a bare number can’t be judged without one. “Good for a first-author paper in this subfield within three years of publication” is an answerable question; “is 50 citations good” is not.
- Do you have access to a normalized metric? If the paper has an FWCI (Scopus/SciVal), CNCI (Web of Science/InCites) or RCR (NIH iCite, for NIH-funded biomedical work), use that instead of the raw count. A score meaningfully above 1.0 is a genuinely defensible “good.”
- If no normalized metric is available, compare against a same-field, same-year, same-document-type baseline manually — for example, the median citation count of other papers published in the same journal issue, or in the same subfield and year, pulled from the same database you’re quoting your number from.
- Account for time since publication. A count that looks low for a paper published last year may be entirely typical; citation accumulation is slow in the first one to two years after publication in most fields.
- Don’t use a single paper’s citation count, or your own h-index, as a career-evaluation tool in isolation. Responsible-metrics frameworks such as the Leiden Manifesto and DORA (the San Francisco Declaration on Research Assessment) caution specifically against reducing hiring, promotion or funding decisions to a single citation-based number, precisely because of the field-, time- and database-dependence covered above.
Is 100 citations good? Is that “a lot”?
The honest answer is: it depends entirely on the four variables above, and “100” carries no fixed meaning without them. A paper reaching 100 citations within two years in a small, slow-citing subfield would be an unusually strong result. The same 100 citations accumulated over fifteen years in a large, fast-citing biomedical field could be entirely unremarkable. There is genuinely no threshold — not 10, not 100, not 1,000 — that functions as a universal “good” across fields, because the underlying citation-rate baselines the field-normalized metrics above are built on differ by field by a large enough margin that no single raw number holds up across all of them. This is exactly why FWCI, CNCI, MNCS and RCR exist: they replace “how many citations” with “how many citations relative to the right comparison set,” which is the version of the question that actually has a stable answer.
Frequently asked questions
What is the average number of citations per paper?
There is no single figure that applies across fields, because average citation rates vary enormously by discipline, document type and time window. This is precisely why field-normalized metrics (FWCI, CNCI, MNCS, RCR) exist: instead of comparing a paper against a global average that mixes together fields with very different citation practices, they compare it only against papers in the same field, year and document type. A score of 1.0 on any of those metrics represents “average” for the correct, narrow comparison set — that is a more useful answer than any single cross-field number could be.
Is 100 citations a lot?
It can be, or it can be unremarkable — the same count means very different things depending on field, time since publication and document type. See the section above for how to actually evaluate a specific count rather than judging it against a fixed threshold.
How many citations is considered good for an h-index?
That’s a related but distinct question about a career-level composite metric rather than a single paper’s citation count. See CASRAI’s dedicated entry on what counts as a good h-index for the same field/career-stage/database caveats applied to that metric specifically.
What citation metric should I actually report if I need one number?
If the source database is available, a field-normalized indicator (FWCI for Scopus-based work, CNCI for Web of Science-based work, RCR for NIH-funded biomedical work) is preferable to a raw count, because it is comparable across fields and time in a way raw counts are not. See CASRAI’s comparison of Impact Factor vs. h-index for how journal-level and author-level metrics differ, and why neither should be applied to judge an individual paper’s citation count on its own.
Last verified 2026-08-16. Metric definitions confirmed against Elsevier SciVal/Scopus documentation, Clarivate InCites documentation, CWTS Leiden Ranking indicator methodology, and NIH iCite/RCR documentation (Hutchins et al., “Relative Citation Ratio,” PLOS Biology, 2016, doi:10.1371/journal.pbio.1002541). No universal “average citations per paper” figure is asserted on this page because none exists that holds across fields — that absence is itself the sourced, verified answer to the query.
See also: Inciteful citation network — Inciteful is a free citation-graph tool that maps related papers from a seed paper using co-citation, bibliographic coupling, and PageRank-style ranking.
