Open Access Metrics: Downloads, Altmetrics, and Reach

Citations are slow and only count readers who publish. Downloads and altmetrics see further, but only if you know exactly what each number counts.

Explainer · 10 min read · Directive Publications Editorial Team

Why citations undersell open access

When a research office is asked what its open access spending achieved, citations are the first thing it reaches for. They are familiar, and they sit in databases the office already licenses. They are also slow, and they count only one kind of reader: someone who goes on to publish.

That matters most for open access, because the trial evidence points to readers rather than citers. In a randomised controlled trial across 11 American Physiological Society journals, published in the BMJ in 2008, open access articles received more full-text downloads and unique visitors in their first six months, but were no more likely to be cited in the first year. A larger trial by the same lead author across 36 journals, published in The FASEB Journal in 2011, found more downloads but no more citations within three years. It suggested the real beneficiaries may be "communities of practice that consume, but rarely contribute to, the corpus of literature."

In medicine, that means clinicians outside research institutions, guideline developers, patients, journalists, and policy staff, none of whom appear in a citation count. So measuring open access impact starts with a question no citation database answers: who was this work meant to reach, and what would show that it did?

Is there an open access citation advantage?

Authors are often told that open access papers are cited more. The evidence is less tidy.

A 2021 systematic review in PLOS ONE included 134 studies comparing citations to open access and non-open access articles. Of these, 64 (47.8%) found a citation advantage, 37 (27.6%) found none, 32 (23.9%) found one only in subsets of their sample, and one was inconclusive. Only three studies were judged at low risk of bias overall, and they split three ways: one advantage, one none, one in subsets. The authors concluded that the quality and heterogeneity of the studies make generalisation difficult. Observational comparisons also struggle to separate the effect of openness from authors' choices about which papers to make open.

What to do with that:

• Do not promise authors more citations as the reason to publish open access, and keep percentage "citation boosts" out of institutional reports.
• Make the case on what open access demonstrably changes: who can read the work, under what licence, and at what cost to the reader.
• When someone quotes a single figure, ask which study, which field, which kind of open access, and whether it was randomised.

A guaranteed citation boost is the mirror image of the belief that open access means low quality, one of the open access myths that still mislead research offices, and it deserves the same scepticism.

Article download statistics: what they count

A download figure means little until you know the counting rules behind it. The shared standard is the COUNTER Code of Practice. Release 5.1.1 is the current code; the Release 5.1 family has been the requirement for COUNTER compliance since January 2025. Three of its metrics do most of the work:

Compliant platforms count two clicks on the same link by the same user within 30 seconds as one action, and must exclude traditional bots and crawlers. Machine readers are the newer problem: COUNTER notes that generative and agentic AI systems "have proliferated since publication of Release 5.1", and its best practice on AI usage metrics, published in April 2026 and updated to apply to Release 5.1.1, is now available to use. It adds an optional "Agent" value for the Access_Method field, so AI access can be separated from human reading, and optional AI metric types covering responses generated and content investigated or requested by AI systems. These are reported only when a library asks for them, so ask your publishers and repository whether they supply them yet, and treat a sudden surge as a prompt to look closer, not as proof of new readers.

The feature that matters most for open access is global reporting. Usage that cannot be linked to an institution is attributed to "The World", and COUNTER's open access guidance recommends that publishers with open access content provide Global Item Reports, broken down by country. A country breakdown of global usage is direct evidence of geographic reach, so ask the publisher whether it provides one.

Download counts still do not add up neatly across platforms:

• Each platform counts its own traffic. One article can be read on the journal site, in PubMed Central, in a repository, and as a preprint. PMC reports usage of its copies to participating publishers separately. A 2013 study of 14 society journals in The FASEB Journal found that when NIH-funded articles became free in PMC, downloads from the journals' own websites fell. A publisher's figure alone cannot show every place an article is read.
• Not every platform follows COUNTER. A "views" counter on an article page may include abstract views or automated traffic.
• Downloads accumulate. Compare articles over the same number of months since publication.
• Versions split readers. A preprint, accepted manuscript, and version of record of the same work each collect their own usage.

Decision rule: combine download figures only when every source reports COUNTER Unique_Item_Requests for the same period, and label the result "tracked usage on [platforms]", never "total readership".

What are altmetrics?

Altmetrics record online attention to research beyond citations: news stories, blogs, policy documents, clinical guidelines, patents, Wikipedia entries, and social media posts. PlumX, from Elsevier, groups signals into five categories: Citations, Usage, Captures (such as bookmarks and saves), Mentions (such as news and blog posts), and Social Media. Altmetric condenses its data into the donut and a single Attention Score.

The Altmetric Attention Score is a weighted count. Its default minimum weights are 8 for a news story, 5 for a blog post or podcast, and 3 for a patent, a Wikipedia mention, or a policy document or clinical guideline (counted per source). A post on X, Bluesky, Facebook, or Reddit counts 0.25. Mendeley readers and Dimensions citations carry a weight of zero. Only one mention per person per source is counted, and news outlets are tiered by reach, so you cannot reproduce the score by adding up mentions.

Altmetric states plainly that the score "is not a measure of the quality of the research or the researcher," and that "attention can be both positive and negative." The score can also fall, when posts are deleted, spam is removed, or the algorithm is reweighted.

Reading an attention record properly

Open the details page and read the mentions by source. For a medical paper, check policy and guideline mentions first. A citation in a clinical guideline is direct evidence of reach into practice; a burst of social posts is evidence of circulation. The score blends both into one number.

A low score does not prove nobody used the work. Altmetric lists reasons mentions are missed:

• The mention had no working link to a page carrying the research identifier, such as the DOI. Unlinked mentions are matched by text mining only for some sources, such as news and policy documents.
• The link pointed to a press release or another news story, not to the research.
• The source was behind a login or paywall, or the mention was in a comment.
• A link was added to a news story after publication; news stories are scanned once.

So ask your communications team to link the DOI directly in every press release and post from the moment it goes out, and report trackable mentions that were missed: Altmetric's policy is to add them manually once notified.

Can altmetrics be gamed?

Yes, and social media is the easiest part to inflate. Altmetric says it combats gaming by "avoiding metrics that are easily gamed (e.g. Facebook likes, YouTube views)", "ignoring repeated mentions of the same article by the same account", and "detecting and ignoring X posts that exhibit the hallmarks of gaming". DORA asks organisations that supply metrics to "be clear that inappropriate manipulation of metrics will not be tolerated".

The institution's part is simpler. The surest way to invite gaming is to set a target, so keep attention scores out of promotion criteria, awards, and performance plans. When a record looks inflated, check its composition: attention made up almost entirely of social posts from a handful of accounts, with no news, policy, or guideline mentions, is promotion, not uptake.

How institutions can report reach fairly

The San Francisco Declaration on Research Assessment (DORA) asks institutions to "consider a broad range of impact measures including qualitative indicators of research impact, such as influence on policy and practice," and asks metric suppliers to account for variation in article types and subject areas. COUNTER's own guidance says that none of usage, citations, or altmetrics should be used alone. Build your reporting around both points:

1. Every number states its source, platform, metric name, and date retrieved
2. Downloads are COUNTER Unique_Item_Requests, global where available, and never summed across unlike platforms
3. Comparisons use the same article type, field, and months since publication
4. Attention is reported as named mentions, with the score as context at most
5. No citation claim assumes an open access advantage
6. Authors can see and correct the data before it is used about them
7. No metric is used alone to judge a paper or a person

For a single article, a short structured entry works better than a row of scores:

If your office funds article processing charges and wants a cost-per-use figure, COUNTER's guidance is to use global usage rather than your institution's own, and it warns that first-year usage will not show the long-term value of making work open. Keep that calculation alongside your decisions on APC waivers and fair funding for authors.

The first change to make

Stop reporting open access impact as bare numbers. Wherever a report says "Altmetric score: [n]" or "[n] downloads", replace it with the metric's name, the platform, the date, and the mentions behind the number. That one habit removes most of the ways these figures mislead, and it makes visible the reach that matters in medicine: into guidelines, policy, and practice.

Authors preparing work for open access publication can find submission guidance on the Directive Publications for authors page, and more articles for research offices are on the Directive Academy blog.