Do citation counts matter? How to judge a research paper

In 2010, Psychological Science published a study of 42 people reporting that holding two expansive "power poses" for a minute each raised testosterone, lowered cortisol and made people take more risks. On 27 September 2026, OpenAlex counted 875 citations to it, in the top 1% of papers of its type, year and subfield. One of its three authors, Dana Carney, has posted a statement that reads: "I do not believe that 'power pose' effects are real" (Carney, n.d.).

So do citation counts matter? Yes, as a record of how much a paper has been used, argued with and built on; no, as a verdict on whether it is right. This guide, published by EdCitation, explains what a count measures, what a good number of citations looks like by field, what the h-index is, and how to judge a paper on what actually decides its reliability. Every count names its index and date.

EdCitation does two of these steps free, with no account: Find sources puts a "Cited by" figure on every result, with a Times cited filter and a Most cited sort, and Cite a source turns the paper you keep into a reference from the publisher's record, checking it for retraction on the way.

Do citation counts matter?

They matter as evidence of attention, not of quality. A citation count is the number of later works in one index whose reference lists were matched to the paper; OpenAlex (n.d.) defines its own figure as exactly that, a count of successful reference matches. Nothing in it says whether the citing author agreed.

Each index gives its own number. On 27 September 2026 Find sources showed the power posing paper as "Cited by 875" when it searched OpenAlex and "Cited by 692" when it searched Crossref, the publishers' registry. Neither is wrong: each counts the citing documents it holds and could match. Our guides to the most cited papers of all time and the first citation index show how wide those gaps get and where the indexes came from.

A count still tells you something. A paper cited thousands of times is one a reader in that field is expected to know. It does not tell you whether a marker would accept it as good evidence.

What is a good number of citations?

There is no single good number: it depends on the field, the paper's age and the kind of document. A count means something only next to the counts of similar papers.

Fields cite at very different rates

On 27 September 2026 we asked OpenAlex how many research articles published in 2015 each field holds, how many had 100 or more citations, and how many had none.

One hundred citations puts a 2015 mathematics article in roughly its field's top 1.3%; in biochemistry the same count is far less rare. Read the humanities row with care: OpenAlex files book reviews and short pieces as articles, those fields cite books, and its field labels are assigned by software.

The Leiden Manifesto's sixth principle makes the point with journals: top-ranked journals have impact factors of around 3 in mathematics and about 30 in cell biology, so indicators must be normalised, most robustly by percentiles within each field's citation distribution (Hicks et al., 2015).

Age and kind of paper change the count too

A count only grows. By OpenAlex's yearly figures the power posing paper gathered 97 citations in 2017 alone, and 31 more in 2026 by late September; last year's paper cannot compete, whatever its quality. Kind matters too: DORA (2013) asks those who supply metrics to account for article types, such as reviews against research articles, and for subject areas whenever counts are compared.

Field-weighted citation impact and percentiles

A normalised figure deals with all three. OpenAlex (n.d.) gives each work a field-weighted citation impact: citations received divided by those expected for works of the same type, year and subfield, where 1.0 is the world average, also given as a percentile. The power posing paper's was 93.3 on 27 September 2026. That answers "how much attention, for a paper of its kind?" and still not "is it true?"

What is the h-index? The h-index explained

The h-index describes a researcher, not a paper. Hirsch (2005) defined it: a scientist has index h if h of their papers have at least h citations each, and the rest no more than h each. A worked example, with invented figures, for six papers cited 25, 12, 8, 5, 3 and 1 times:

1. Sort the papers from most cited to least.
2. Number them down the list.
3. Find the last position where the count is still at least the position number: paper 4 has 5 citations, paper 5 only 3.
4. That position is the h-index: 4.

Working from physicists' records, Hirsch suggested, with "large error bars", about 12 as typical for tenure at a major research university and about 18 for a full professor. Life scientists run far higher: the 36 biologists and biomedical scientists elected to the US National Academy of Sciences in 2005 had a median h of 57 (Hirsch, 2005).

What the h-index cannot tell you

Hirsch named its limits himself: a single number "can never give more than a rough approximation" to a person's work, and an h built on many-author papers is treated "overly kindly" (Hirsch, 2005). The Leiden Manifesto's seventh principle adds that it rises with age even without new papers, and tops out near 200 for life scientists, 100 for physicists and 20 to 30 for social scientists (Hicks et al., 2015). For choosing sources it says nothing: a famous author's weakest study carries the same h.

Can a paper be highly cited because it is wrong?

Yes. A disputed paper is cited by everyone who tests, questions or corrects it, as well as by those who repeat it.

Negative and critical citations

Catalini et al. (2015) classified the 762,355 citations in 15,731 Journal of Immunology articles (1998 to 2007), training software on 15,000 citations sorted by immunologists. About 2.4% were negative, and 7.1% of the papers cited received at least one; negative citations tended to concern higher-quality papers and to target findings rather than theory or methods. Each still adds one to a count.

Silence distorts too. Greenberg (2009) mapped the 242 papers and 675 citations behind one claim in muscle disease: supportive primary studies received 94% of the 214 citations to primary data, and the six papers whose data weakened or refuted the claim received 6%.

The power posing paper

Ranehill et al. (2015) repeated the study in what their title calls a large sample of men and women, and found no effect on hormones or risk tolerance. Simmons and Simonsohn (2017) analysed the 33 successful studies the original authors had gathered in reply and found their p-values indistinguishable from no true effect plus selective reporting. Carney's statement lists the study's problems, including "The sample size is tiny", and states that the effect is no longer taught in Carney's classes (Carney, n.d.). The paper is not retracted, and its count keeps rising.

Retracted papers keep their counts

Mehra et al.'s 2020 hydroxychloroquine registry study in The Lancet was retracted that June, three of its authors writing that they could "no longer vouch for the veracity of the primary data sources" (Mehra et al., 2020). On 27 September 2026 OpenAlex still counted 1,255 citations to it, 672 from 2020, some of them about the retraction; the number cannot say which. See retracted papers still being cited for the studies, and our guide to AI research assistants for how often automatic "supporting" and "contrasting" labels miss.

How do I evaluate a research paper beyond its citation count?

Judge what decides whether the findings hold, and read the count last. This is how to tell if a paper is reliable enough to cite:

1. What kind of study is it? Trials, cohorts, case reports and opinion answer different questions with different risks of bias; see levels of evidence.
2. Where was it published, and was it peer reviewed? Check the journal's review policy and whether the item is an article, a letter or a preprint; see how to tell if an article is peer reviewed.
3. Do the methods and sample fit the claim? Forty-two participants cannot settle a question about everyone's hormones.
4. Has anyone repeated it? Search its central claim in Find sources and read what the replications and critiques say.
5. Has it been retracted or corrected? Cite a source screens each source as it builds the reference; Verify references flags retracted papers across a whole list.
6. Who paid? The ICMJE (n.d.) asks authors to disclose financial and non-financial relationships so that readers can judge whether they bear on the work.
7. Only then, the count, with its index and date, against papers of the same field and age.

How do the Times cited filter and Most cited sort work in Find sources?

Each does one exact thing, and knowing which saves a bad shortlist.

What each one does

• Times cited is a filter with steps Any, 10+, 50+, 100+ and 500+, plus the quick set "Cited 50 times or more". It keeps works the searched index counts as cited at least that often.
• Most cited is a sort. Results whose titles hold at least half as much of your search as the best-matching title come first, by count; looser matches follow by relevance.
• Each card shows "Cited by" from the index searched, and the page names it: "Results from OpenAlex."
• Retracted papers are left out unless you tick "Include them, marked as retracted".
• When the full index is busy, the search runs on Crossref, which cannot apply Times cited, and the page says which filters are not applied.

Our runs, 27 September 2026

1. Power posing, Best match: 2,672,450 results, provider OpenAlex. Carney et al. (2010), cited by 875, came first, with Ranehill et al. (2015), 244, and Simmons and Simonsohn (2017), 159, right under it.
2. Most cited: the full index was busy, so Crossref answered: 1,831,131 results, Carney et al. first with 692, then a 2015 follow-up by Cuddy and colleagues, 146, and Simmons and Simonsohn, 124. Ranehill et al., which Crossref's record counted at 181 that day, was not on the page; when a paper you expect is missing, search its title.
3. Times cited 100+ with Most cited: 63,032 results, provider OpenAlex. Six papers on power posing led, 875 down to 112; seventh was a 1994 book on power systems cited 19,556 times, kept below them because its title shares one word of the search.
4. Hydroxychloroquine macrolide COVID-19 registry analysis, Most cited: 1,363 results, with Mehra et al.'s retracted study absent. With retracted papers included: 1,377, and the study third, cited by 1,255, labelled Retracted.

Citing the paper you chose

We gave Cite a source the DOI 10.1177/0956797610383437. It took the record from Crossref, found no retraction and no notice, and produced this APA 7 entry:

Carney, D. R., Cuddy, A. J. C., & Yap, A. J. (2010). Power posing: Brief nonverbal displays affect neuroendocrine levels and risk tolerance. Psychological Science, 21(10), 1363–1368. https://doi.org/10.1177/0956797610383437

Note the clean retraction check: not retracted is not the same as reliable.

Which EdCitation tool helps me judge a source?

Find sources is the best free place to start judging a paper. Each "Cited by" figure is attached to a real record that opens, with its index stated on the page, so nothing is estimated or composed the way a chatbot's list can be; and one search on power posing put the failed replication and the p-curve analysis straight under the famous paper. It covers some 300 million published works, searched by subject or by the claim one of your sentences has to support, and narrowed by kind of source, years, open access and times cited. No account, no charge.

For the list itself, Cite a source formats each entry in APA 7, MLA 9, Chicago 18 author-date, Harvard, IEEE or Vancouver with a retraction screen, and Verify references returns every entry as verified, "check this" or not found, flags retractions, and keeps "could not check" apart from "not found". Neither costs anything. Upload a whole draft to References from a file, a Pro tool at $8 a month (see pricing), and each in-text citation is tied to its reference while every entry is checked.

EdCitation never writes any part of your work. It finds, cites and verifies; deciding which sources your argument can stand on is your judgement, and it is a skill worth having.

Quick questions

Does a high citation count mean a paper is good?

No. It means one index matched many later references to it, critiques included. The 2010 power posing study had 875 citations in OpenAlex on 27 September 2026, and one of its authors no longer believes the effect is real.

What is a good number of citations for a paper?

It depends on field, year and kind. In OpenAlex on 27 September 2026, 5.8% of 2015 biochemistry articles had 100 or more citations, against 1.3% in mathematics. Compare like with like, or use a percentile.

What is a good h-index?

It depends on field and career stage. Hirsch (2005) suggested about 12 for tenure and 18 for a full professor in physics, with large error bars. It describes a researcher, never a single paper.

How can I find only well-cited papers on a topic?

In EdCitation's Find sources, set Times cited to 10+, 50+, 100+ or 500+, or sort by Most cited. Then read them: a count shows what a field has used, not what is right.

Should I cite a retracted paper?

Only to write about the retraction. EdCitation's free Verify references flags retracted papers across a whole list before you hand it in.

References

• Carney, D. R. (n.d.). My position on "power poses". University of California, Berkeley, Haas School of Business. https://faculty.haas.berkeley.edu/dana_carney/pdf_my%20position%20on%20power%20poses.pdf
• Carney, D. R., Cuddy, A. J. C., & Yap, A. J. (2010). Power posing: Brief nonverbal displays affect neuroendocrine levels and risk tolerance. Psychological Science, 21(10), 1363–1368. https://doi.org/10.1177/0956797610383437
• Catalini, C., Lacetera, N., & Oettl, A. (2015). The incidence and role of negative citations in science. Proceedings of the National Academy of Sciences, 112(45), 13823–13826. https://doi.org/10.1073/pnas.1502280112
• DORA. (2013). San Francisco Declaration on Research Assessment. https://sfdora.org/read/
• Greenberg, S. A. (2009). How citation distortions create unfounded authority: Analysis of a citation network. BMJ, 339, Article b2680. https://doi.org/10.1136/bmj.b2680
• Hicks, D., Wouters, P., Waltman, L., de Rijcke, S., & Rafols, I. (2015). Bibliometrics: The Leiden Manifesto for research metrics. Nature, 520(7548), 429–431. https://doi.org/10.1038/520429a
• Hirsch, J. E. (2005). An index to quantify an individual's scientific research output. Proceedings of the National Academy of Sciences, 102(46), 16569–16572. https://doi.org/10.1073/pnas.0507655102
• ICMJE. (n.d.). Disclosure of financial and non-financial relationships and activities, and conflicts of interest. International Committee of Medical Journal Editors. https://www.icmje.org/recommendations/browse/roles-and-responsibilities/author-responsibilities--conflicts-of-interest.html
• Mehra, M. R., Ruschitzka, F., & Patel, A. N. (2020). Retraction: Hydroxychloroquine or chloroquine with or without a macrolide for treatment of COVID-19: A multinational registry analysis. The Lancet, 395(10240), 1820. https://doi.org/10.1016/S0140-6736(20)31324-6
• OpenAlex. (n.d.). Attributes: Works. OpenAlex Help Center. Retrieved September 27, 2026, from https://help.openalex.org/data/works/attributes/
• Ranehill, E., Dreber, A., Johannesson, M., Leiberg, S., Sul, S., & Weber, R. A. (2015). Assessing the robustness of power posing: No effect on hormones and risk tolerance in a large sample of men and women. Psychological Science, 26(5), 653–656. https://doi.org/10.1177/0956797614553946
• Simmons, J. P., & Simonsohn, U. (2017). Power posing: P-curving the evidence. Psychological Science, 28(5), 687–693. https://doi.org/10.1177/0956797616658563