Does Open Access Increase Citations? 20 Years of Evidence (2026 Update)

"Does publishing open access increase citations?" is one of the oldest questions in bibliometrics. The honest 2026 answer is messier than either side of the policy debate tends to admit: there is a real effect, it is small on average, it varies sharply by field and OA route, and the best-evidenced version of the story contradicts the intuition that paying a Gold APC is how you buy more citations.

Where the famous numbers actually come from

The headline citation-advantage figures that still circulate in university OA guides trace back to two early studies.

Lawrence (2001, Nature) analysed 119,924 computer-science conference papers and reported that freely available articles were cited roughly 286% more often than those behind paywalls. Single field. Conference papers. Observational design. The "4.5×" figure often repeated in later summaries is a rough rephrasing of this single-field number. It was never a meta-analysis.

Eysenbach (2006, PLOS Biology) tracked a cohort of PNAS articles and found OA ones were about twice as likely to be cited at 4–10 months after publication, with the advantage sustained through 10–16 months. Single journal. Short window. Pre-Sci-Hub era, so non-OA access no longer means what it meant in 2004.

Both papers were observational and both carried the obvious selection confound: authors who pay APCs or self-archive are not a random sample of the research population. They launched a field of inquiry. They did not settle it.

The 2021 systematic review: heterogeneity is the finding

The most rigorous synthesis of this literature to date is Langham-Putrow, Bakker, and Riegelman (2021, PLOS ONE), a systematic review of 134 studies comparing citation rates of OA and subscription articles:

• 47.8% of studies found an OA citation advantage.
• 27.6% found no advantage.
• 23.9% found an advantage only in subsets of their sample.
• 0.8% found a citation disadvantage.

Averaged across 134 studies, OA helps. But the within-study variation is the signal that actually matters for a publishing decision. The effect depends on field, journal prestige, time window, and how "open access" is operationalised. A single pooled number hides everything that would help you predict whether it will hold in your paper's case.

Why paying for Gold doesn't buy what you think it buys

The biggest shift in the last decade has been evidence against the intuitive story that Gold OA journals cause the citation bump. Piwowar et al. (2018, PeerJ) analysed 67 million articles via the oaDOI service and reported that OA articles received 18% more citations on average, after controlling for age and discipline. But the effect was concentrated in Green and Hybrid OA, not pure Gold. Pure-Gold OA journals tended to cite at rates similar to subscription journals.

Counterintuitive, but replicated. Plausible mechanisms:

• Hybrid OA articles sit inside high-impact journals, so they get the journal's prestige and the access bump.
• Green self-archiving selects for author-engaged, actively-promoted papers; the act of depositing correlates with the kind of author who also promotes their work.
• Many pure-Gold journals are younger and less established; their APC-paying author pool does not mirror the elite cohort that self-archives in institutional repositories.

If you are choosing a route on citation grounds alone, the evidence does not recommend paying a Gold APC over submitting to the best journal you can reach and depositing a Green copy.

Broader readership, not just more citations

A distinct line of evidence asks whether OA expands who cites you, rather than just how many. The Science.org news coverage (2019) summarising the Huang–Neylon dataset reported that OA papers drew citations from roughly 31 institutions on average versus 21 for paywalled, and from 9 countries on average versus 7. Subsequent bibliometric work (see Scientometrics, 2024) has reinforced the pattern.

The readership benefit is not evenly distributed. A 2024 analysis of global reference patterns (Karlstrøm et al., Journal of Information Science) found researchers in lower-income countries cite OA works disproportionately: they are precisely the readership that cannot reach paywalled content. But a 2024 Nature news feature notes that OA's publishing side can actively disadvantage those same researchers, because APCs (~$2,000 on average) are prohibitive. OA expands reading access faster than it expands publishing access.

If your goal is reach beyond the handful of elite universities that subscribe to everything (clinical adoption, policy influence, global-south uptake), readership breadth may matter more to you than the headline citation count.

Preprints: the cheapest route to most of the benefit

For disciplines with established preprint servers (physics, biology, mathematics, computer science, increasingly medicine), a preprint is effectively zero-cost Green OA from the day of submission. A 2024 analysis of PLOS publications from 2018–2023 reported that articles with preprint versions received about 20% more citations than those without.

This is correlational (preprinting is not randomly assigned) and plausibly reflects both discoverability and selection: authors confident enough to preprint may also produce stronger work. But for citation-advantage purposes, a preprint captures most of the access benefit without an APC. See our guide on optimising your bioRxiv preprint for discoverability.

What 2026 changes

Several pressures are reshaping the calculation in ways the canonical 2001–2021 literature did not capture:

1. AI retrieval tools (ChatGPT, Perplexity, Claude) can only cite what they can read. Paywalled PDFs, particularly those behind aggressive anti-scraping infrastructure, are systematically excluded from AI-generated answers. See our review of generative engine optimisation for biomedical papers.
2. Google's AI Overviews reduce click-through to paywalled results that never render a readable snippet.
3. Funder mandates (Plan S, NIH, UKRI, cOAlition S) have pushed an increasing share of new biomedical papers to some form of OA compliance, shifting the baseline against which "OA vs paywalled" comparisons are even drawn.

None of these was measured in the 2001–2021 studies above. They make a well-indexed, openly-readable paper more discoverable per unit of research quality than at any previous point. They also change the relevant comparison: in 2026, the axis that matters is less "OA vs paywalled" and more "retrievable by humans and machines vs not." An open PDF with a badly chosen title and a hedging abstract opener is still hard to find.

The publishing decision, grounded in evidence

Post a preprint. If your field has a server (bioRxiv, medRxiv, arXiv, SSRN), this is the single highest-value step per unit of effort. You capture most of the OA advantage for free.

Publish in the best-fit journal you can reach. Journal quality, editorial fit, and peer-review calibre matter more than the Gold/subscription label. A paywalled paper in a strong field-specific journal will usually outperform a pure-Gold OA paper in a weaker venue.

Deposit a Green copy. Self-archive the accepted manuscript in your institutional repository or a disciplinary archive (PMC, arXiv, bioRxiv). On the Piwowar 2018 evidence, this is where the citation advantage actually lives.

Pay a Gold APC when the funder pays or the paper is borderline on other grounds. The citation-grounds argument for APCs is weak. The compliance, speed, and reader-experience arguments can still be valid on a per-paper basis; they are just different arguments from "Gold buys citations."

Optimise what the reader (and the machine) sees first. Access is necessary, not sufficient. If nobody can find your paper on Google, PubMed, or in ChatGPT, the licence is beside the point. See our evidence review on title optimisation and the recent calibrated citation study on 983 biomedicine papers.

Honest limitations

Most of the OA citation-advantage literature was written before large language models began consuming and citing scholarly content at scale. The 20-year empirical case is strong on direction (OA helps, modestly, with real heterogeneity) and weak on contemporary magnitude: the controls that worked in 2018 may not map cleanly onto a world where the marginal citation is generated by a retrieval-augmented model. A 2026-or-later replication of the OACA with AI-generated citations as an outcome is, as far as I am aware, not yet published. If you see one, send it over.