“The Cost of Knowledge” is this year’s theme for International Open Access Week, and this “challenges us to consider the financial, human, and environmental costs of creating, sharing, and sustaining knowledge.” In health research, that conversation should extend to the cost of ensuring that knowledge is trustworthy, credible, and sustainable. The cost of knowledge includes not only the cost of making research accessible, but also the cost of making it trustworthy.
What does trustworthy health research cost?
Producing a trustworthy health research article requires considerable effort from multiple stakeholders. Authors design studies, obtain ethical approvals, collect and analyze data, report their methods and results, and respond to critique. Editors and reviewers assess quality, relevance, methodology, ethics, and competing interests. Journals check disclosures, reporting guidelines, references, metadata and permissions; and they also preserve the published record, investigate concerns, and issue corrections when necessary. Without this effort, discoveries cannot be effectively communicated, understood, assessed, replicated, or applied in clinical settings. Trust in the final output is produced through layers and layers of work; it is built by people, policies, safeguards, checks, and the willingness to correct the record.
Scholarly research is in an age of abundance. According to Digital Science’s Dimensions platform output in the biomedical and clinical sciences more than doubled from approximately 750,000 items in 2010 to 1.65 million in 2025. This growth reflects expanding research capacity — it does not, by itself, indicate declining quality, but it does however create an operational challenge: the expert capacity needed to assess the increasing volume is not growing at the same rate.
In healthcare, this scarcity of expertise matters. Published research is synthesized into systematic reviews, translated into clinical guidelines, cited in policy, and used to shape further research. Errors can, and do, become embedded in the evidence chain. A 2025 study published in The BMJ identified 1,330 retracted clinical trials and 847 systematic reviews that had quantitatively synthesized retracted trials.1 When those trials were removed, the direction of the pooled effect changed in 8.4% of replicable meta-analyses, statistical significance changed in 16%, and the magnitude of the effect changed by more than 50% in 15.7% of replicable meta-analyses. Evidence from 68 affected systematic reviews had also been used in 157 guideline documents.
Those findings should not be read as evidence that medical literature as a whole is unreliable. Rather, they show that publication is not the endpoint; trustworthy scholarly communications require continuing stewardship to detect problems, correct the record, and prevent retracted evidence from continuing to circulate.
How can AI support research integrity?
Generative AI may, to varying degrees, help support efforts to address some aspect of the expertise shortage by supporting screening, verification, and error detection. Its value, however, is greatest when it supports rather than replaces expert human judgement. The downside, to put it mildly, is AI can also create authoritative-sounding errors (or “hallucinations”) at scale. A widely discussed correspondence article in The Lancet audited 2.5 million biomedical papers and identified 4,046 fabricated references across 2,810 papers, with the authors noting that such references may be generated by paper mills, deliberate misconduct, and the improper use of AI.2 Technology can support the process, but there is no substitute for expert-verified evidence.
The primary guardrail for AI use is straightforward: AI may assist, but humans remain accountable. Current guidance from the International Committee of Medical Journal Editors (ICMJE) states that AI cannot be an author; that humans remain responsible for accuracy, attribution, and plagiarism checks; that AI use should be transparent; and that confidential manuscripts should not be uploaded to systems where confidentiality cannot be assured without the authors' permission.3
How does open access affect the cost of trustworthy knowledge?
How, then, does this relate to open access and the cost of knowledge? Open access does not, by itself, validate a method, detect an invented reference, or keep a clinical guideline current, but the value of open access content is dependent upon reliable editorial policies, peer review, ethical oversight, preservation, and visible corrections and retractions. Accessibility and quality are complementary, not competing, objectives. “Open” should therefore not be treated as synonymous with “costless.” As content migrates from closed to open models, costs shift from subscriptions to article processing charges (APCs) or other funding mechanisms. Whether open or closed, a journal must be able to support editorial independence, rigorous peer review, technical standards, integrity controls, discoverability, preservation, and post-publication correction. It must also enable researchers to participate regardless of geography or access to funding. These are the foundations that make knowledge usable and trustworthy.
Why sustainable investment in research quality matters
Open Access Week is an opportunity to make this often-invisible work visible. We should recognize peer review and editorial stewardship, build integrity checks into publishing workflows rather than adding them later, maintain transparent correction mechanisms, and use technology to focus scarce subject-matter expertise where it matters most. We should be candid that trustworthy publishing requires sustainable investment — and equally candid about who pays, what they are paying for, and how the system can support equitable participation.
The hidden cost of knowledge is the work required to make it trustworthy. In healthcare, evidence can move from a paper into a guideline and from a guideline into patient care. Trust is therefore the most valuable currency the research system has, and one that everyone involved in producing and communicating research must continually earn.
References
- Xu C, Fan S, Tian Y, et al. Investigating the impact of trial retractions on the healthcare evidence ecosystem (VITALITY Study I): retrospective cohort study. BMJ. 2025;389:e082068. doi: 10.1136/bmj-2024-082068.
- Topaz M, Roguin N, Gupta P, et al. Fabricated citations: an audit across 2·5 million biomedical papers. Lancet. 2026;407(10541):1779–1781. doi: 10.1016/S0140-6736(26)00603-3.
- International Committee of Medical Journal Editors (ICMJE). Recommendations: Use of Artificial Intelligence in Publishing. Updated January 2026. Available at: https://www.icmje.org/recommendations/browse/artificial-intelligence/.